NETSYN : a connectionist approach to synthesis knowledge acquisition and use

Size: px
Start display at page:

Download "NETSYN : a connectionist approach to synthesis knowledge acquisition and use"

Transcription

1 Carnegie Melln University Research CMU Department f Electrical and Cmputer Engineering Carnegie Institute f Technlgy 1992 NETSYN : a cnnectinist apprach t synthesis knwledge acquisitin and use Nenad Ivezic Carnegie Melln University James Henry. Garrett Carnegie Melln University.Engineering Design Research Center. Fllw this and additinal wrks at: This Technical Reprt is brught t yu fr free and pen access by the Carnegie Institute f Technlgy at Research CMU. It has been accepted fr inclusin in Department f Electrical and Cmputer Engineering by an authrized administratr f Research CMU. Fr mre infrmatin, please cntact research-shwcase@andrew.cmu.edu.

2 NOTICE WARNING CONCERNING COPYRIGHT RESTRICTIONS: The cpyright law f the United States (title 17, U.S. Cde) gverns the making f phtcpies r ther reprductins f cpyrighted material. Any cpying f this dcument withut permissin f its authr may be prhibited by law.

3 NETSYN - A Cnnectinist Apprach t Synthesis Knwledge Acquisitin and Use N. Ivezic, J. Garrett EDRC

4 NETSYN a Cnnectinist Apprach t Synthesis Knwledge Acquisitin and Use a Technical Reprt by Nenad Ivezic and James Garrett, Jr. July 1992 This wrk has been supprted by the Engineering Design Research Center, an NSF Engineering Research Center.

5 11

6 Abstract The gal f machine learning fr synthesis is the acquisitin f the relatinships between frm, functin, and behavir prperties that satisfy design requirements. The prpsed apprach creates a functin t estimate the prbability f each pssible value f each design prperty being used in a given design cntext. NETS YN uses a cnnectinist learning apprach t acquire and represent this prbability estimatinfunctin and exhibits gd perfrmance fr apsed artificial design prblem. The bjective f future wik is t apply NETS YN t realistic design dmains, specifically the dmain f cmputer system design as practiced by Ml. in

7 IV

8 Acknwledgments The authrs gratefully acknwledge the help prvided by Benit Julien and Yram Reich in evaluating the ILLS system and the ECOBWEB system fr the purpses f cmparisns presented in this reprt.

9 VI

10 Table f Cntents Chapter 1 Intrductin SYNTHESIS AND THE ROLE FOR MACHINE LEARNING Design Framewrk fr Synthesis A Rle fr Machine Learning OVERVIEW OF APPROACH 3 13 MOTIVATIONS FOR USING CONNECTIONIST APPROACH ORGANIZATION OF REPORT 5 Chapter 2 Backgrund CHARACTERISTICS OF THE SYNTHESIS PROCESS REQUIREMENTS FOR A LEARNING SYSTEM 8 23 INDUCTIVE LEARNING APPROACHES FOR SKAU 8 Chapter 3 A Prbabilistic Apprach T Synthesis ASSUMPTIONS ABOUT SYNTHESIS a psteriri PROBABILITY ESTIMATION FOR SYNTHESIS General Idea Required Accuracy fr the Prbability Estimatin Functin Appraches t Prbability Estimatin EXAMPLE USAGE: MICON SYNTHESIZER VERSION 1 (Ml) MICON Synthesizer Versin 1 (Ml) Example 1 Cmpleting a Design fr CGA cntrller Example 2 Estimating Behavir fr Partial Designs Example 3 Recgnizing Incnsistent Specificatins 15 Chapter 4 NETSYN a Cnnectinist Apprach t SKAU A CONNECTIONIST a psteriri PROBABILITY ESTIMATION THEORETICAL BASIS FORNETSYN Cmputatinal Thery Backgrund Learning Thery Backgrund CONNECTIONIST (NETSYN) REPRESENTATIONS Design Representatin Design Prcess Representatin NETSYN ARCHITECTURE NETSYN LEARNING 22 vu

11 Chapter 5 Evaluatin f NETSYN TEST METHODOLOGY - AN ARTIFICIAL DESIGN PROBLEM (ADP) Synthesis Space Synthesis Knwledge Training and Test Set Generatin Perfrmance Evaluatin Methdlgy Perfrmance Evaluatin Methdlgy NETSYN PERFORMANCE ON ADP TEST 1 Results TEST2 Results COMPARATIVE ANALYSIS: ILLS AND NETSYN COMPARATIVE ANALYSIS: ECOBWEB AND NETSYN DISCUSSION 33 Chapter 6 Future Wrk 35 Chapter 7 Summary 36 References 37 V1U

12 List f Tables Ikble 1 - Synthesis Rules fr Attribute E 27 Ikble 2 - Synthesis Rules fr Attribute F 27 Tkble 3 - Synthesis Rules fr Attribute G 27 Ikble 4 - Synthesis Rules fr Attribute H 27 IX

13 List f Figures Figure 1 Synthesis 2 Figure 2 Analysis 2 Figure 3 Evaluatin 2 Figure 4 Redesign 2 Figure 5 Relatinships Needed t Capture fr SKAU 3 Figure 6 Alternative Decisin Sequences fr a Synthesis Prcess 11 Figure 7 The Prbability Estimatin Functin 12 Figure 8 Iterative Design Cmpletin Using the Prbability Estimatin 13 Figure 9 Synthesis Hierarchy fr CGA Cntrller 16 Figure 10 Cmpletin f Design fr CGA cntrller 17 Figure 11 Estimating Behavir fr Partial Design 18 Figure 12 Recgnizing Incnsistent Specificatins fr CGA Cntrller 18 Figure 13 Mapping a Neural Netwrk n Prbability Estimatin 20 Figure 14 NETSYN Architecture 22 Figure 15 Mdular Cnstructin f a Neural Netwrk fr SKAU 23 Figure 16 Itaining f a Netwrk fr Synthesis Knwledge Acquisitin 24 Figure 17 NETSYN Perfrmance with Respect t "set-cverage" (TEST 1) 30 Figure 18 NETSYN Perfrmance with Respect t "max-crrect" (TEST 1) 30 Figure 19 NETSYN Perfrmance with Respect t "set-cverage" (TEST 2) 32 Figure 20 NETSYN Perfrmance with Respect t "max-crrect" (TEST 2) 32 Figure 21 Cmparisn f Perfrmance: NETSYN and ILLS 34 Figure 22 Cmparisn f Perfrmance: NETSYN and ECOBWEB 34

14 Chapter 1 Intrductin The bjectives f this chapter are t describe a general framewrk fr the synthesis prcesses cnsidered in this research and t identify a rle fr machine learning in design dmains with weak r nnexistent synthesis knwledge. In additin, this chapter includes an verview f, and the mtivatin fr, cnnectinist learning appraches fr synthesis knwledge acquisitin and use (SKAU). Finally, an utline f the cntents f the reprt is given. 1.1 SYNTHESIS AND THE ROLE FOR MACHINE LEARNING Design Framewrk fr Synthesis The main bjective f the synthesis prcess is t generate a descriptin f the frm f an artifact such that this descriptin satisfies a cllectin f design requirements. Frm here is used in the same sense as in [Hemming 92] t mean gemetry, tplgy, material, etc. The synthesis prcess is the initial stage f a multi-stage artifact design prcess cnsisting f the fllwing fur stages: synthesis, analysis, evaluatin, and redesign. In the synthesis stage, several pssible frms f an artifact are generated based n a given set f specificatins. In the analysis stage, the behavir f these pssible frms are determined fr the specified functinal cntext. In the evaluatin stage, the predicted behavirs are cmpared with thse desired and the deficiencies f the design are identified. In the redesign stage, changes t riginal pssible frms are made t address the deficiencies fund during evaluatin, r new realizatins are generated. If we cnsider the artifact resulting frm a design prcess t be described by its frm (e.g., tplgy, shape, material),/wrtcri0rt (e.g., lad resistance, vibratin islatin), and behavir prperties (e.g., stress and strain states), then we can describe each f the fur stages in the multi-stage prcess as fllws [Flemming92]: Synthesis is the prcess f mapping frm the design specificatin space (requirements n functin, behavir, frm) t the frm space (Fig. 1). Analysis is the prcess f mapping frm the frm and specificatin spaces t the behavir space (Fig. 2). Evaluatin is the prcess f cmparing the behavir prperties t the behavir requirements and the frm prperties t the frm requirements (Fig. 3). Redesign is the prcess f mapping frm the specificatin space t the frm space, with the knwledge f subspaces that lead t infeasible designs (Fig. 4) A Rle fr Machine Learning The specificatin fr an artifact used as input by the synthesis prcess may invlve any number f required functinalities, frm cnstraints, and behavir cnstraints. Fr prblems where all f the cnstraints and bjectives can be clearly articulated and expressed in algebraic frm, mathematical pti-

15 Functinal Specs Behaviral Specs Frm Specs Specificatin Space represents set f frm prperty values Frm Space Figure 1 Synthesis represents set f frm attribute values Frm Space represents a set f behaviral attribute values Functinal Specs Specificatin Space Figure 2 Analysis Behavir Space 1 + behavir frm functin i. ; IcmnareJ^ ^^.. -I cmpare!-. * ^ Specificatin Space Frm Space Figure 3 Evaluatin Behavir Space infeasible subspaces nal Specs iviral Specs represents set f frm attribute values Specificatin Space Frm Space Figure 4 Redesign

16 mizatin appraches can be emplyed t search ver the slutin space (i.e., the space f all pssible slutins) and select the "ptimal" slutin t the set f cnstraints and defined bjectives. Hwever, fr prblems where there are few cnstraints and the relatinships between frm and perfrmance are unknwn, a mathematical ptimizatin apprach is nt amenable. Hwever, if weak theries r heuristic knwledge abut synthesizing slutins t a prblem exist, knwledge-based synthesis appraches can be used t assist (thrugh the applicatin f heuristic knwledge) in efficiently searching thrugh the space f pssible design slutins, eliminating thse slutins that are nt prmising and fcussing n thse slutins that are. Hwever, frthse prblems in whichn significant dmain thery r cllectin f heuristics exists, the nly surce f synthesis knwledge tends t be past design experience [Reich 91]. One rle fr machine learning, identified in [Reich 91], lies within thse design dmains pssessing the fllwing characteristics: The design space is identified (i.e., all prperties and pssible values are knwn a priri). Synthesis knwledge is weak r nnexistent There exists a cllectin f design experiences. The rle f machine learning in dmains f this type is t acquire, frm past design cases, the relatinships between frm, functin, and behavir prperties that satisfy specified design requirements. These relatinships frm a cne f useful synthesis knwledge. Having such synthesis knwledge facilitates a mre direct mapping frm the specificatin space t the desired lcatins within the frm and behavir spaces. In ther wrds, less search is needed t find a gd slutin t the specificatin (Fig. 5). 1 behavir frm functin ; m t _^-~-~-H" satisfies l^-» *2> *** Specificatin Space Frm Space Behavir Space Figure 5 Relatinships Needed t Capture fr SKAU Hence, the gal f the research described in this reprt is t investigate and identify prmising learning appraches which are capable f "playing" the abve rle f acquiring synthesis knwledge. Next, we utline the apprach taken by this research in accmplishing this gal that will be presented in mre detail later in this reprt. 12 OVERVIEW OF APPROACH We prpse a prbabilistic apprach t synthesis that is based n the fllwing assumptins: Design can be sufficiently well represented as a cllectin f prperty-value pairs (i.e., a finite number f decisins). Cnsequently, the synthesis prcess is equivalent t a sequence f assignments f prperty values which ultimately lead t a cmplete design definitin.

17 Each prperty value assignment (decisin) is made in sme design cntext (i.e., partially defined specificatin and/r design descriptin). It is pssible t cnstruct a sufficiently accurate estimatin f the prbability f each value f each design prperty being used in a given design cntext. Based n the abve assumptins, the gal f identifying learning appraches capable f acquiring synthesis knwledge is transfrmed int tw subgals: Identify a useful prbabilistic cncept which prvide an adequate measure f belief that a prperty value can be used in sme design cntext. Identify an effective methd fr the cnstructin f a prbability estimatin functin that implements the selected prbabilistic cncept. In this reprt we demnstrate that the slutin t these tw subgals (and t the riginal gal) may be achieved by applying cnnectinist learning appraches. Our cnnectinist-based slutin relies n the fllwing: A recent fundamental theretical research result demnstrates that a grup f cnnectinist learning methds is capable f estimating Bayesian a psteriri prbabilities. A representative methd frm the abve grup f cnnectinist learning methds may be successfully used t acquire and stre the prbability estimatin functin, thus, allwing inductive learning t be the methd f cnstructin fr this prbability estimatin functin. This reprt discusses and illustrates the use f a cnnectinist learning apprach t acquire and represent a prbability estimatin functin that can be used t achieve ur gal acquire synthesis knwledge fr its subsequent reuse. 1J MOTIVATIONS FOR USING CONNECTIONIST APPROACH Building a synthesis system in prblem dmains where there is a weak, r nnexistent, dmain thery, and where mathematical ptimizatin methds are nt amenable is a difficult task. Fr these types f prblems, an inductive machine learning apprach may pssibly be used t capture, and cnsequently permit reuse f, the synthesis knwledge embdied in the past design cases. While sme very gd research has been cnducted n hw symblic machine learning techniques can be applied t the synthesis prblem [Reich 91, Lu 92], little research has investigated hw neural netwrks might be used t capture and subsequently use synthesis knwledge. The research reprted here has fcuses n cnnectinist learning methds fr SKAU. The fllwing are the mtivatins fr investigating the cnnectinist learning methds fr SKAU: Cnnectinist systems are adaptive systems capable f capturing cmplex nnlinear relatinships. Bth the theretical results cncerning representatinal capabilities f cnnectinist systems [Cybenk 88, Hrnik 89], and the results f applicatins f cnnectinist mdels t difficult learning prblems in cntrl [Pmerleau 89], signal predictin [Lapedes 88] and pattern recgnitin [Sejnwski 87, Qian 88], illustrate that cnnectinist systems are capable f learning cmplex relatinships frm a sample f input-utput pairs representative f that relatinship. The relatinships between design variables can be very cmplex and it is felt that the pwer f neural netwrks can be emplyed t induce these relatinships frm a representative training sample f previus designs. The knwledge representatin paradigm in cnnectinist systems ffers attractive capabilities. The distributed knwledge representatin fund in cnnectinist systems has been shwn t lead t the kinds f peratins that are reminiscent f peratins f the human brain: cntent addressable memry recall (i.e., recnstructin), nise resistant cmputatin, and gracefully degrading cmputatin [Hintn 86]. Obviusly, a learning system fr SKAU culd beneficially emply these kinds f

18 peratins, particularly in the cntext f design dmains with weak synthesis knwledge. In additin, distributed knwledge representatin allws autmatic generalizatin [Hintn 86], which is ne f the key requirements fr inductive learning systems fr SKAU. Cnnectinist learning systems have exhibited cnsistently gd perfrmance n tasks similar t SKAU. The applicatin f cnnectinist systems that mtivates ur research is neural pattern recgnitin [Le Cun 89]. Since the dmains f interest t this research are assumed t have weak frmal knwledge f the synthesis prcess, ne apprach t acquire the actual synthesis knwledge is t search fr meaningful patterns f features in past successful designs. Recgnitin f such patterns and their generalizatin t new design scenaris culd assist the synthesis prcess in prducing "gd" designs. The existence f a lng list f successful neural pattern recgnitin applicatins, illustrates the feasibility f a cnnectinist apprach t SKAU [Le Cun 89, Tesaur 90, Pmerleau 89]. Cnnectinist learning systems estimate Bayesian a psteriri prbability. The result that a class f cnnectinist learning systems is capable f estimating Bayesian a psteriri prbability prvides justificatin fr emplying a neural pattern recgnitin system in perfrming SKAU. The fundamental difference between applying a pattern recgnitin system t the task f SKAU and ther similar tasks is the nnunique nature f the mappings used in synthesis. The acquired synthesis knwledge has t preserve all f the relatinships that hld between design prperties fr all f the multiple pssible alternatives. This is fundamentally different frm the requirement psed in ther tasks that search fr nly ne unique slutin t the prblem. By incrprating the cncept f a psteriri prbability estimatin we are able t satisfy the need t represent bth the relatinships and the multiple alternatives fr SKAU. In additin, this result prvides a rigrus theretical backgrund fr analysis f the perfrmance f cnnectinist systems and fr their use in practical applicatins. Cnnectinist systems allw the use f Bayesian learning methd. Sme f the recent research has fcussed its attentin n the analysis f learning in cnnectinist systems by fllwing the Bayesian learning methd [Buntine 91]. While we currently assume that the unifrm cnvergence methd applies (i.e., we assume the existence f a sufficiently large sample size and ignre factrs that appear as prirs in the learning mdel), the Bayesian learning methd prvides imprtant results fr the case f limited sample sizes. Althugh the Bayesian apprach has nt been applied in this research t date, we acknwledge the imprtance f these results fr future wrk. 1.4 ORGANIZATION OF REPORT The fllwing is the rganizatin f this technical reprt: Chapter 2 presents the synthesis prcess cnsidered in this research and discusses ne type f machine learning (i.e., inductive learning) apprach that is capable f SKAU under the cnditins specified in sectin 1.1. We present the characteristics f the synthesis prcess in general and f the synthesis prcess in dmains with weak r nnexistent synthesis knwledge. Frm this discussin we derive basic requirements fr a learning system capable f learning synthesis knwledge. We als describe tw alternative inductive learning appraches which have been emplyed fr SKAU. Chapter 3 develps a prbabilistic apprach t synthesis. The apprach is based n the assumptin that it is pssible t cnstruct a sufficiently accurate estimatin f the prbability f each pssible value f each design prperty being used in a given design cntext. The assumptins abut the synthesis prcess cnsidered in this apprach are presented and a prbabilistic apprach fr synthesis knwledge acquisitin and use (SKAU) is described. We discuss the minimal requirements n the accuracy f the prbability estimatin functin t be useful fr synthesis. The apprach is illustrated n a number f examples taken frm a realistic design dmain. Chapter 4 prvides a detailed accunt f NETSYN - a neural netwrk apprach fr synthesis knwledge acquisitin and use (SKAU). The basic idea f the apprach is t use a cnnectinist learning

19 system t acquire and represent the prbability estimatin functin intrduced in Chapter 3. Theretical results relevant t the expected perfrmance f this apprach are presented alng with a discussin f hw this learning apprach fits int a mre general machine learning framewrk. Cnnectinist representatins are then discussed. Discussin f the architecture and the training f NETSYN is prvided. Chapter 5 describes the test methdlgy used fr evaluating the perfrmance f NETSYN and its cmparisn t tw symblic inductive learning appraches: ILLS and ECOBWEB. A series f results is presented fr the perfrmance and cmparisn f NETSYN t these symblic learning appraches. Chapter 6 gives directins fr the future wrk cncerning the verall gal f NETSYN: applicatin t a realistic design synthesis task. Chapter 7 summarizes the results presented in this technical reprt.

20 Chapter 2 Backgrund In this Chapter, we describe in mre detail the synthesis prcess cnsidered in this research and discuss ne a type f machine learning apprach that is capable f SKAU under the cnditins specified in the previus chapter. First we present the characteristics f synthesis prcess in general and f the synthesis prcess in dmains with weak r nnexistent synthesis knwledge. Then, we present basic requirements fr a learning system capable f learning synthesis knwledge in such a dmain. Finally, we describe tw alternative inductive learning appraches which have been emplyed fr SKAU. 2.1 CHARACTERISTICS OF THE SYNTHESIS PROCESS The engineering synthesis prcess is ften viewed as an explratin f a space f pssible slutins fr a psed design prblem. Hwever, the directins fr perfrming this search are rarely given in an explicit frm fr any engineering design dmain. Design prblems are ill-structured and the design prblem frmulatin is itself a dynamic prcess where new design knwledge causes the refrmulatin f the design prblem and, subsequently, refrmulatin f the synthesis search space [Cyne 90]. In certain design dmains, it may be assumed that the dynamics f the design prblem frmulatin are slwer than the actual synthesis space traversal. This bviusly hlds in mre mature design dmains where the knwledge f the design synthesis space (i.e., design variables and the relatinships between them) has been reinfrced thrugh the many successful utcmes f synthesizing designs fr given design specificatins. The synthesis prcess in many design dmains is characterized by generating alternatives which "satisflee" the given design requirements [Simn 81]. Therefre, the utcme f a synthesis prcess in such a design dmain is a cllectin f candidate designs that satisfy initial design requirements. The nnunique nature f the synthesized designs pses a requirement n the designer t cnsider and evaluate mre than ne alternative slutin. Anther characteristic f the synthesis prcess is the utilizatin f varius pieces f infrmatin f different nature. Bth qualitative and quantitative data are cmbined by a designer during the design synthesis prcess. The ways in which such data are cmbined t decide which part f the search space t traverse are ften hard t express r analyze rigrusly using classical mathematical appraches. In the prcess f generating alternatives during the synthesis prcess, the designer will narrw the search space by cnsidering nly selected subintervals f values f design variables. Eventually, the design variables will be assigned specific values t give a mre cmplete design cntext The decisin t select a particular variable value will influence the feasible ranges f ther design variables. The relatinships between design variables are in general cmplex and many-t-many: cnstraining ne design variable will affect the selectin prcess f all ther variables, while a particular design variable assignment may be incnsistent with previusly cnstrained design variables and cause an infeasible design. These relatinships are results f designer's experience and are reinfrced thrughut many successful synthesis prcesses. The designer's ability t capture and t generalize these relatinships frm a finite number

21 f previus design episdes and, subsequently, t use this knwledge in new design situatins is a crucial capability fr successful synthesis. In the framewric f a frmulated synthesis prblem, where the dimensinality f the search space (i.e., prperties t be specified) is knwn but the synthesis knwledge (i.e., relatinships between design prperties) is weak r nnexistent, the nly surce f synthesis knwledge tends t be the past design experience. In such a design situatin, where an experienced designer is faced with a new design prblem, the first step is t perfrm sme limited (t the extent t which the synthesis thery is available) analysis f the design prblem and then t prceed with the synthesis prcess. The knwledge f first principles f behavir in the cnsidered engineering dmain is f limited utility in cnstructing the design alternatives. T slve the design prblem, the designer relies n his r her past experience t recgnize the similarities f the new situatin t the design prblems he r she has encuntered in the past. The designer recgnizes the imprtant characteristics f the design prblem and fcuses his r her attentin n these imprtant cnsideratins. This subjective, experience-based knwledge f the designer will gvern the synthesis prcess f generating alternative design slutins. The identified nature f the synthesis prcess in dmains with nnexistent r weak dmain thery frms the basis fr deriving the requirements fr systems that are t capable f capturing synthesis knwledge and facilitating its reuse in new design prblems. These requirements are presented next. 2.2 REQUIREMENTS FOR A LEARNING SYSTEM Three requirements are identified as necessary ingredients f any system capable f capturing synthesis knwledge and using this knwledge in nvel design situatins [Reich, 1991]: The ability t capture many-t-many mappings between frm, functin and behavir design prperties. In general, ne must cnsider all design prperties fr which values have been fixed r ranges cnstrained when making decisins abut yet unbund design prperties. Cnsidering nly the design specificatins and disregarding values f design attributes fixed in the curse f synthesis will in general lead t infeasible slutins. The ability t recall r generate multiple design alternatives satisfying the particular cmbinatin f specificatins. As described earlier, the synthesis prcess is cncerned with generating multiple alternatives fr the same design specificatins. The ability t generalize frm the individual design experiences i.e., generating acceptable slutins in design situatins nt seen befre. A learning system is expected t behave well in the new design situatins that are similar t, but nt identical t, previus successful design episdes. 2.3 INDUCTIVE LEARNING APPROACHES FOR SKAU Inductive learning appraches are candidate machine learning appraches fr acquiring synthesis knwledge and allwing the reuse f this knwledge in new design situatins. A brief discussin f tw alternative inductive learning appraches fr synthesis knwledge acquisitin is presented next (fr a mre exhaustive treatment f learning methds fr this task, see [Reich 91]). Inductive learning appraches can be gruped in tw main classes: The first is the class f supervised learning appraches. In supervised learning appraches, learning is based n the availability f a crrect answer fr any input descriptin f a situatin that needs t be classified r t which a value is t be assciated. The learning system has t adjust itself accrding t the representative training cases. The gal f learning is that the system ultimately learns crrect assciatins between input and utput patterns. The secnd is the class f unsupervised learning appraches. In this case, the learning target is nt specified explicitly in terms f crrect results fr a set f training cases. The nly available inf rma-

22 tin is in the crrelatins f the input data. The system has t discver regularities in the training set and t create categries based n the discvered crrelatins. These categries are the basis fr classificatin f the new input. An example f an unsupervised learning apprach t synthesis knwledge acquisitin is the ECOB WEB learning algrithm used in the Bridger system [Reich 91]. The ECOBWEB learning algrithm is used t frm meaningful clusters ut f design training cases. These clusters can then be used t cmplete the synthesis prcess fr given partial design descriptins. ECOBWEB utilizes a prbabilistic apprach and it uses a heuristic perfrmance measure t cmpare alternative classificatins with respect t their utility in the classificatin f designs. These classes are then used t classify a new, partially cmplete design specificatin. Once the input specificatin is classified, the whle classificatin hierarchy can be used t synthesize design prperty values cnsistent with members f that class. The supervised learning apprach has been criticized as being inapprpriate fr learning synthesis knwledge [Reich91]. Early learning systems used this apprach in a straightfrward manner which prevented them frm learning "cmplete" relatinships between design prperties: nly the mapping frm the set f fixed specificatin prperties t design descriptin prperties was captured. N attentin was paid t the influence that fixing sme design attribute values might have n the feasible ranges f ther design prperties. Hence, many-t-many relatinships between design prperties culd nt be captured and the use f the supervised learning apprach was rightfully judged as inapprpriate fr learning synthesis knwledge. In ur research we have fllwed the supervised learning apprach. Hwever, the fundamental difference between ur apprach and the previus appraches using supervised learning is that ur basis fr classificatin is nt just the subset f design prperties declared t be specificatins. Rather, all design prperties knwn in a given design cntext are used t determine the values f the remaining, unknwn design prperties. The apprach is basically ne f predicting the a psteriri prbabilities f each pssible value f each unknwn prperty fr the given set f knwn prperty values. Using this apprach, it is pssible t accunt fr the many-t-many relatinships within synthesis knwledge. The next sectin gives a detailed descriptin f this apprach.

23 Chapter 3 A Prbabilistic Apprach T Synthesis In this chapter we develp aprbabilistic apprach t synthesis. The apprach is based n the assumptin that it is pssible t cnstruct a sufficiently accurate estimatin f the prbability f each pssible value f each design prperty being used in a given design cntext. First, assumptins abut the synthesis prcess cnsidered in this research are presented. Then, a prbabilistic apprach fr synthesis knwledge acquisitin and use (SKAU) is described. Next, the minimal requirements n the accuracy f the prbability estimatin functin t be useful fr synthesis are given. Finally, the apprach is illustrated n a number f examples taken frm the dmain f Ml [Gupta 91] - a knwledge based system fr synthesizing single-bard cmputer systems. 3.1 ASSUMPTIONS ABOUT SYNTHESIS In this sectin we list several assumptins abut the kind f synthesis prcesses we address in this research and fr which we attempt t create a learning system. We view a design as a cllectin f prperty-value pairs r, equivalently, cllectin f design decisins. The design synthesis prcess is then equivalent t a sequence f assignments f prperty values which ultimately leads t a cmplete design definitin. Sme prperty values are determined a priri, representing design specificatin values. All design decisins are made in sme design cntext (i.e., a partial design descriptin). T further elabrate n ur assumptins, cnsider a representatin f a simplified design prcess shwn in Fig. 6. On the left side f the figure, an initial design cntext is shwn. T cmplete the design descriptin, the values f three design prperties need t be determined (ne specificatin prperty and tw design descriptin prperties). Each design prperty may take n ne ut f three distinct values (graphically represented by an rectangle, triangle, and an val). In the initial cntext, nly the design specificatin value is determined (dented by a slid shape). The first task in the design prcess is t cnsider all f the design prperties fr which values have nt been determined and t select ne and assign its value. In the mst general case the designer wuld like t cnsider every unreslved design prperty and decide which values are feasible fr the given design cntext. Then, based n sme design strategy, the designer culd select a prperty and its value which he believes will lead t a gd design. In the figure, cmpleting this task gives the intermediate state f design cntext (1) and the new, mre cmplete, design cntexts (2 r 3). The intermediate state (1), represents cnsideratin f unbund design prperties. The different shades f symbls representing values depict that each value may appear in a given design cntext (the darker shades indicate greater level f belief that the value applies fr the given design cntext). Alternative design cntexts (2 and 3) are results f different design decisins. Similarly, ging frm either f tw partial design cntexts (2 r 3), the designer cnsiders the values that the remaining design prperty may assume. Intermediate states f the design cntext (2a) and (3a) represent a degree f belief that the designer may have that each unbund prperty culd be assigned each f its pssible values in design cntext (2) and (3), respectively. The final design cntexts (2b and 3b) are representatins f utcmes f tw alternative decisin sequences. 10

24 Design attributes edesign sped 01 jaol Initial design cntext PAO1 valued LI1 Final design cntexts Figure 6 Alternative Decisin Sequences fr a Synthesis Prcess A particular pint wrth nting here is that a designer usually has a number f alternative chices fr making synthesis decisins. The designer utilizes his r her subjective knwledge f the synthesis prcess (i.e., knwledge f relatinships amng prperty values) t determine which prperty value will be used in the given cntext (which changes as decisins are made); cnsequently, he r she decides which particular prperty value t determine next. We are specifically interested in acquiring this subjective knwledge that is used t select particular prperty values fr given design cntexts. 32 a psteriri PROBABILITY ESTIMATION FOR SYNTHESIS General Idea We cast the described synthesis prcess int a prbabilistic framewrk. The prbabilistic cncept we select t emply in this research is Bayesian a psteriri prbability. This cncept has been lng recgnized as useful relative t engineering planning and design, as it incrprates engineering judgment and bservatinal data in a frmal framewrk [Ang 75]. T use the cncept f a psteriri prbability in a traditinal way, ne has t knw a priri prbabilities f classes f phenmena ne deals with as well as likelihd functins (i.e., class cnditinal prbability functins) f the measurements (i.e., dependent variables) that ne can btain fr these phenmena. Measurements may be cntinuus r discrete values and they are represented in the frm f a measurement vectr. The Bayesian a psteriri prbability p(q IX) represents the cnditinal prbability f class Q given the input measurement vectr X. Use f Bayes rule allws it t be expressed as fllws: w _p (X I Q) p(c t ) where X is a measurement vectr p (XI Q) is the likelihd f prducing the measurement X if the class is Q p (Ci) is the a priri prbability f class Q p (X) is the uncnditinal prbability f the measurement In ur synthesis framewrk, the phenmena in which we are interested are the design prperty value assignments fr which we knw in advance the classes (i.e., values) and the measurements are all bund 11

25 design prperties that make a current design cntext (i.e., the measurement vectr). The a priri prbabilities can in general be estimated fairly well; hwever, the likelihd functins invlve assumptins abut the prbability distributins f values f the phenmena (design prperties) with respect t the measurements (design cntext). Instead f indirectly estimating a psteriri prbabilities f each prperty value assignment by cmputing prbabilities accrding t Bayes rule, we will cnsider an alternative apprach f directly estimating a psteriri prbabilities (hereafter, prbabilities) using a cnnectinist learning mdel. This is a subject f Chapter 4. One way t use the a psteriri prbability cncept is t use it as a prbability estimatin f each value f each unbund design prperty being used in a given design cntext Such prbabilities will be very useful infrmatin during synthesis. Cnsider Fig. 7, which illustrates this idea. Each prperty value is dented by a circle. The prperties are divided int specificatins and frm and behavir prperties. The assigned prperties frm the current design cntext which is perated upn by the prbability estimatin functin. The result f this peratin is a set f estimated prbabilities f each value f each unbund design prperty being used in the given design cntext The shaded circles n the right hand side f the figure represent these prbability estimatins fr each prperty value. Daiker shades represent higher prbability f the value appearing in the given design cntext. Fr a given set f design specificatins, r any partial descriptin f design, by using a prbability estimatin functin we are in a psitin t d the fllwing: Acquire the prbabilities f the values appearing fr each unbund frm and behavir prperty. Reasn abut which values f unbund design prperties are viable alternatives in a given design cntext specificatins s^ frm and behavir prperties s4 dl". values. OO O O OOO OOO OO values " Prbability Estimatin _ values _ OO O O O «O OO CO values I sl ' s2! s3! s4 : di J d2 ; d3 ; d4 specificatins; frm and J behavir ; prperties! Legend: m = attribute value selected = P(dijlSn&Si2&.. &S 42 ) Figure 7 - The Prbability Estimatin Functin Given such a tl, ne can imagine a synthesis prcess being cmpleted by iteratively applying the fllwing steps, starting frm sme design cntext (i.e., design specificatins) and ending with the cmpletin f the design (see Fig. 8): Apply the prbability estimatin functin t acquire the prbability estimatin f each value f each unbund design prperty fr the given design cntext. Cnsider prbability distributins ver the values fr each unbund design prperty and decide t which prperty t assign a value next (this step may be subject t a specific strategy f selecting the next prperty t bind). D

26 Initial Design Cntext Iteratin 1: Prbab. Estimat. OO Iteratin 2: Prbab. Estimat Iteratin 3: O9O Prbab. Estimat Figure 8 Iterative Design Cmpletin Using the Prbability Estimatin Cnsider the prbabilities fr values f the selected prperty and decide which value t assign t the prperty. Update the design cntext by fixing the prperty value Required Accuracy fr the Prbability Estimatin Functin Frm the perspective f the usability f this apprach, the prbability estimatin functin shuld: (1) estimate nn-zer prbabilities fr design prperty values which have appeared in a given design cntext; (2) estimate zer prbabilities fr the prperty values that have nt been in that design cntext; and (3) crrectly rank the mst prbable design prperty value. This minimalist accuracy requirement is necessary as the relatinships between design prperties in a realistic design dmain may be very cmplex, and cnstructin f a prbability estimatin f an arbitrarily high accuracy may be impssible. Althugh the abve requirement is sufficient t render the prpsed apprach useful, it still may be t hard fr practical design prblems. It is likely that fr any prblem f this cmplexity nly an apprximatin f the the prbability estimatin functin will be pssible. We will return t this accuracy requirement again in Chapter 5, where we deal with evaluatin f the prpsed apprach Appraches t Prbability Estimatin One may emply different appraches t cnstruct the prbability estimating functin. These appraches may in general be classified as exact prbabilistic methds, apprximatins f exact prbabilistic meth- 13

27 ds, r heuristic methds. By fllwing an exact apprach, ne needs t fllw axims f prbability thery and prvide a rigrus prf that the cnstructed prbability estimating functin fr the cnsidered prblem has the desired prperties. While this may be pssible fr small prblems, the difficulty f real synthesis prblems prevents the actual usage f this apprach. Methds fr apprximating these prbabilistic methds may be a mre viable apprach t cmplex tasks such as synthesis. The quality f the perfrmance f these apprximating methds depends n the prblem typically, the size f the sample that is used t cnstruct the prbability estimatin functin and the dimensinality f the prblem. Therefre, the usefulness f this apprximating apprach varies with the cnditins that the actual synthesis prblem may impse. Finally, the heuristic methds sacrifice exactness fr feasibility and decreased cmplexity in cnstructing the prbability estimating functin. In additin, these methds are smetimes emplyed t intrduce prblem-dependent biases, which cause the estimatin f the prbabilities t fllw sme assumed prbability distributins peculiar t that prblem. In the next chapter we present a cnnectinist apprach t cnstructing the prbability estimatin functin that actually estimates a psteriri prbability. 3.3 EXAMPLE USAGE: MICON SYNTHESIZER VERSION 1 (Ml) This sectin illustrates the use f the prpsed prbabilistic apprach within the framewrk f a synthesis system fr the design f cmputer systems. First, a descriptin f the synthesis tl in MICON (Ml) and the system-level synthesis task perfrmed by this tl are given (this part is based n [Gupta 91]) MICON Synthesizer Versin 1 (Ml) Ml (MICON Synthesizer Versin 1) is a part f the MICON system that designs single-bard cmputer systems [Gupta 91]. Specifically, M1 is a knwledge-based synthesis tl that satisfies high level specificatins by perfrming system-level synthesis f cmputer systems frm a set f cmpnents. The bjective f system-level synthesis is t create a cmplete and peratinal cmputer system capable f perfrming general-purpse r special-applicatin cmputing. The cmpnents f design are integrated circuits (i.e., ICs) r applicatin-specific IC libraries. Input specificatins t Ml are based n a functinal descriptin f the elements; n details f cmpnents and their specifics are included in the specificatins. Typically, the result f the system-level synthesis cnsists f a cnfiguratin f CPU (i.e., central prcessing unit), memry system, I/O (input utput) cmpnent, bus interfaces, and ther supprting circuitry. Tw characteristics f this synthesis task are: There exists n mdel r language t clearly define the functin f an artifact being designed. There exists n cmplete and well-defined thery relating structure (i.e., frm) and behavir f the artifact. Being a realistic prblem, the fllwing are sme prperties that make the synthesis prcess in this dmain hard: The design space is large as the parts can be used in a variety f ways t fulfill functinal specificatins. Cmbinatrial explsin is a result f an attempt t thrughly search ver the whle design space. There are interactins between the design sub-prblems. Ml divides the synthesis prblem int a set f sub-prblems and cannt, in general, determine that the design fr a subsystem will yield a satisfactry cmplete design. There are cmplex interactins amng the cmpnents. The relatinships between cmpnents depend n several factrs, including the way in which a cmpnent is prgrammed. The knwledge base is rapidly evlving. New ICs are cnstantly being develped, leading t imprved cmpnents and new design styles. 14

28 Based n the abve descriptin, this synthesis task belngs t the class f synthesis prblems with well defined design spaces but weak synthesis knwledge. In the fllwing subsectins, we illustrate several ways in which the prpsed prbabilistic apprach fr SKAU culd be used by M1. We cnsider the least cmplex subdmain f M1 CGA cntrller design fr which the synthesis space descriptin is given in the frm f the synthesis hierarchy in Fig. 9. In general, Ml uses an "AND-OR" hierarchy t represent the design space. The synthesis hierarchy fr CGA cntrller design represents a part f this "AND-OR" hierarchy. The state f the Mi's knwledge base fr the CGA cntrller has dictated the shape f, and the size f, this hierarchy. The arch acrss the tp level branches indicates the tp nde being "AND" nde. The lwer level ndes are "OR" ndes (indicated by heavier lines with rdinal numbers in circles in frnt f the alternatives) and "AND" ndes (indicated by lighter lines). The nly decisins that need t be made crrespnd t "OR" ndes. Therefre, nly thse cmpnents crrespnding t "OR" ndes are cnsidered in the fllwing examples. In additin, a subset f perfrmances and specificatins listed in the same figure are included in the illustratins: cst, RAM (Randm Access Memry) size, ROM (Read Only Memry) size, and the speed f the bard Example 1 Cmpleting a Design fr CGA cntrller The initial example illustrates the idea f cmpleting a design fr a given design specificatin (Fig. 10). Fr the given specificatins (cst range, RAM size, ROM size, and desired speed f the bard), the task is t estimate in every step f the design cmpletin prcess the prbability f each value f each unbund design prperty (RAM chip, ROM chip, Address Decder, and number f Address Decder cmpnents) being used in the current design cntext The utcme f the represented design cmpletin prcess can be read frm the final iteratin (4) in Fig. 13: RAM chip = 6264, ROM chip = 27512, Address Decder = OR gate, number f Address Decder cmpnents = 1. Als nte that the synthesized slutin is an utcme f develping nly a single alternative within the synthesis space Example 2 Estimating Behavir fr Partial Designs In the case where a behavir design prperty is included amng the design prperties fr describing design cntext, it is pssible t estimate the behavir f the partially cmpleted designs in the early stages f the synthesis prcess. In Fig. 11, an example f this use f a prbability estimatin functin is illustrated. The indicated input specificatin cnsists f desired RAM size, ROM size, bard speed, and a cnstraint n the desired ROM chip type. By applying the prbability estimatin functin, the prbability f cst being abve $ 150 was estimated t be the highest f all cnsidered cst values, representing the predictin that the cst fr this cnfiguratin will be abve $ Example 3 Recgnizing Incnsistent Specificatins If we estimate the prbabilities f all prperties (including specificatin prperties) based n all ther prperties, then it is pssible t determine whether the given specificatins are cnsistent. Fig. 12 illustrates this by shwing hw the utput f the prbability estimatin functin indicates different values fr prperties fr which values are fixed as a part f specificatin (e.g., the cst was set t be in the range $5(>-$60 n the input, but the estimated prbability indicates that this value fr the cst is nt cnsistent with the rest f the specificatins). T elabrate n this example use, cnsider a subpattern f the input frmed by mitting ne specificatin input This is exactly the input that enters the prbability estimatin functin fr this specificatin. The functin then estimates the prbability fr each value f that specificatin f being a part f cnsidered pattern (i.e., design cntext). Therefre, the estimatin functin is perfectly capable f estimating as nnzer thse prbabilities f design specificatin values which are nt set n input (and vice versa, estimating as zer thse prbabilities f design specificatin values which are set n input) since it des nt cnsider the input values fr this design specificatin in its estimatin prcess. The cause fr this difference 15

29 ^0188_AT_INTF_(HG_SRAM_0 GROM_0 S0188_IO_INTF *0188_PROC_ _ADDR_DEC l # ADDRDEC 3 s I I f n Ci ; CLOCK BUS 3/8DEC 16L8 NOTO FL_FL CONSTANT! I RS_LATCHl j RS_LATCH j (2) L-(T)27020 RES SPDT! CONSTANT!! I t BUS CONSTANT CLOCK 1 BUTTON ~IL DS1232 CONSTANT j OR21 0 [J;CMOS_OSC_2OM 2)CMOS_OSC_16M ScMOS OSC 24M 2 4 DESIGN SPECIFICATIONS AND PERFORMANCES 1) Pwer dissipatin 2) Bard area 3) Cst 4) RAM size 5) ROM size 6) Speed - (L)8O

30 Design Specs: Cst: $80-$90 RAM: 8 KB ROM: 512 KB Speed: 10 MHz «rtcin PmiuHic* RAM chin: ROM chip: Addr. Dec: #Addr.Dec: TC OR 1 L8 & 4 Iteratin 1: * O 0 > < < i <» D O Iteratin 2: t 1 C) O O -JiU IP Iteratin 3: c Lr 1D O O i # O ( Iteratin 4: 4 Figure 10 ler Cmpletin f Design fr CGA cntrl- 17

31 Cst: RAM: 8KB ROM: 512KB Speed: 10 MHz ROM( ;hip: O O 2764' C O Cst: > $15f 9 O 9 Figure 11 Estimating Behavir fr Partial Design between the input value and the estimated prbability f that value is that the selected design prperties have created a design cntext that is incnsistent with the remaining selected prperty value. This incnsistency wuld nt have been encuntered if the decisin making prcess (i.e., prperty assignment) had been cnducted ne decisin at a time. Pst-$50-$60 RAM- 8KB ROM- 512 KB Speed: 10 MHz ROM chip: O O 2764: ( O ROM O 9 O 128 O 512 O 0 Figure 12 Recgnizing Incnsistent Specificatins fr CGA Cntrller Frm the examples shwn, this apprach based n prbability estimatin culd supply useful infrmatin during the synthesis prcess that can be used t guide the prcess mre directly t the lcatin in frm and behavir space that satisfy the given requirements. The next chapter describes a cnnectinistbased apprach that acquires and represents this prbability estimatin functin. 18

32 Chapter 4 NETSYN a Cnnectinist Apprach t SKAU This chapter prvides a detailed accunt f NETSYN - a neural netwrk apprach fr synthesis knwledge acquisitin and use (SKAU). First, the basic idea f the apprach is given by mapping a cnnectinist learning system nt the prbability estimatin functin intrduced in the last sectin. Next, sme theretical results relevant t the expected perfrmance f this apprach are presented alng with a discussin f hw this learning apprach fits int a mre general machine learning framewrk. Cnnectinist representatins are then discussed. Finally, a discussin f the architecture and the training f NETSYN is prvided. 4.1 A CONNECTIONIST a psteriri PROBABILITY ESTIMATION As discussed in subsectin 3.2.3, cnstructin f a prbability estimatin functin is a difficult task fr realistic synthesis prblems. This difficulty is acnsequence f tw prperties f realistic synthesis prblems: The dimensinality f the design space may be very high there may be a large number f interacting design prperties that need t be cnsidered simultaneusly in the synthesis prcess. Relatinships between design prperties may be highly cmplex. Our apprach t cnstructing the prbability estimatin functin is illustrated in Fig. 13. We emply feed-frward neural netwrks (discussed belw in mre detail) as a mechanism by which t acquire and represent the prbability estimatin functin. Tw principal benefits f using a neural netwik in the rle f prbability estimatin functin are: The prbability estimatin functin is acquired thrugh inductive learning. We use existing recrds f previus designs t train the neural netwik t estimate the desired prbabilities. The trained netwrk, under assumptins presented belw, estimates Bayesian a psteriri prbabilities, thus allwing us fr reliance the established theretical backgrund f Bayesian learning appraches. 42 THEORETICAL BASIS FOR NETSYN This sectin presents sme results f fundamental research n which the prpsed cnnectinist apprach t cnstructing the prbability estimatin functin fr SKAU are funded Cmputatinal Thery Backgrund In the paper by Richard and Lippmann [Richard 91 ], the fllwing result has been presented, rigrusly prven, and experimentally tested: "Fr an M class prblem, Bayesian prbabilities are estimated when the netwrk has ne utput fr each pattern class, desired utputs are 1 fm (ne utput unity crrespnding t the crrect class, all thers zer), and an apprpriate cst functin is used. " 19

the results to larger systems due to prop'erties of the projection algorithm. First, the number of hidden nodes must

the results to larger systems due to prop'erties of the projection algorithm. First, the number of hidden nodes must M.E. Aggune, M.J. Dambrg, M.A. El-Sharkawi, R.J. Marks II and L.E. Atlas, "Dynamic and static security assessment f pwer systems using artificial neural netwrks", Prceedings f the NSF Wrkshp n Applicatins

More information

Churn Prediction using Dynamic RFM-Augmented node2vec

Churn Prediction using Dynamic RFM-Augmented node2vec Churn Predictin using Dynamic RFM-Augmented nde2vec Sandra Mitrvić, Jchen de Weerdt, Bart Baesens & Wilfried Lemahieu Department f Decisin Sciences and Infrmatin Management, KU Leuven 18 September 2017,

More information

Technical Bulletin. Generation Interconnection Procedures. Revisions to Cluster 4, Phase 1 Study Methodology

Technical Bulletin. Generation Interconnection Procedures. Revisions to Cluster 4, Phase 1 Study Methodology Technical Bulletin Generatin Intercnnectin Prcedures Revisins t Cluster 4, Phase 1 Study Methdlgy Release Date: Octber 20, 2011 (Finalizatin f the Draft Technical Bulletin released n September 19, 2011)

More information

Revision: August 19, E Main Suite D Pullman, WA (509) Voice and Fax

Revision: August 19, E Main Suite D Pullman, WA (509) Voice and Fax .7.4: Direct frequency dmain circuit analysis Revisin: August 9, 00 5 E Main Suite D Pullman, WA 9963 (509) 334 6306 ice and Fax Overview n chapter.7., we determined the steadystate respnse f electrical

More information

WRITING THE REPORT. Organizing the report. Title Page. Table of Contents

WRITING THE REPORT. Organizing the report. Title Page. Table of Contents WRITING THE REPORT Organizing the reprt Mst reprts shuld be rganized in the fllwing manner. Smetime there is a valid reasn t include extra chapters in within the bdy f the reprt. 1. Title page 2. Executive

More information

Part 3 Introduction to statistical classification techniques

Part 3 Introduction to statistical classification techniques Part 3 Intrductin t statistical classificatin techniques Machine Learning, Part 3, March 07 Fabi Rli Preamble ØIn Part we have seen that if we knw: Psterir prbabilities P(ω i / ) Or the equivalent terms

More information

Math Foundations 20 Work Plan

Math Foundations 20 Work Plan Math Fundatins 20 Wrk Plan Units / Tpics 20.8 Demnstrate understanding f systems f linear inequalities in tw variables. Time Frame December 1-3 weeks 6-10 Majr Learning Indicatrs Identify situatins relevant

More information

CAUSAL INFERENCE. Technical Track Session I. Phillippe Leite. The World Bank

CAUSAL INFERENCE. Technical Track Session I. Phillippe Leite. The World Bank CAUSAL INFERENCE Technical Track Sessin I Phillippe Leite The Wrld Bank These slides were develped by Christel Vermeersch and mdified by Phillippe Leite fr the purpse f this wrkshp Plicy questins are causal

More information

Subject description processes

Subject description processes Subject representatin 6.1.2. Subject descriptin prcesses Overview Fur majr prcesses r areas f practice fr representing subjects are classificatin, subject catalging, indexing, and abstracting. The prcesses

More information

ENSC Discrete Time Systems. Project Outline. Semester

ENSC Discrete Time Systems. Project Outline. Semester ENSC 49 - iscrete Time Systems Prject Outline Semester 006-1. Objectives The gal f the prject is t design a channel fading simulatr. Upn successful cmpletin f the prject, yu will reinfrce yur understanding

More information

Bootstrap Method > # Purpose: understand how bootstrap method works > obs=c(11.96, 5.03, 67.40, 16.07, 31.50, 7.73, 11.10, 22.38) > n=length(obs) >

Bootstrap Method > # Purpose: understand how bootstrap method works > obs=c(11.96, 5.03, 67.40, 16.07, 31.50, 7.73, 11.10, 22.38) > n=length(obs) > Btstrap Methd > # Purpse: understand hw btstrap methd wrks > bs=c(11.96, 5.03, 67.40, 16.07, 31.50, 7.73, 11.10, 22.38) > n=length(bs) > mean(bs) [1] 21.64625 > # estimate f lambda > lambda = 1/mean(bs);

More information

Eric Klein and Ning Sa

Eric Klein and Ning Sa Week 12. Statistical Appraches t Netwrks: p1 and p* Wasserman and Faust Chapter 15: Statistical Analysis f Single Relatinal Netwrks There are fur tasks in psitinal analysis: 1) Define Equivalence 2) Measure

More information

A New Evaluation Measure. J. Joiner and L. Werner. The problems of evaluation and the needed criteria of evaluation

A New Evaluation Measure. J. Joiner and L. Werner. The problems of evaluation and the needed criteria of evaluation III-l III. A New Evaluatin Measure J. Jiner and L. Werner Abstract The prblems f evaluatin and the needed criteria f evaluatin measures in the SMART system f infrmatin retrieval are reviewed and discussed.

More information

NUROP CONGRESS PAPER CHINESE PINYIN TO CHINESE CHARACTER CONVERSION

NUROP CONGRESS PAPER CHINESE PINYIN TO CHINESE CHARACTER CONVERSION NUROP Chinese Pinyin T Chinese Character Cnversin NUROP CONGRESS PAPER CHINESE PINYIN TO CHINESE CHARACTER CONVERSION CHIA LI SHI 1 AND LUA KIM TENG 2 Schl f Cmputing, Natinal University f Singapre 3 Science

More information

Chapter 3: Cluster Analysis

Chapter 3: Cluster Analysis Chapter 3: Cluster Analysis } 3.1 Basic Cncepts f Clustering 3.1.1 Cluster Analysis 3.1. Clustering Categries } 3. Partitining Methds 3..1 The principle 3.. K-Means Methd 3..3 K-Medids Methd 3..4 CLARA

More information

Pattern Recognition 2014 Support Vector Machines

Pattern Recognition 2014 Support Vector Machines Pattern Recgnitin 2014 Supprt Vectr Machines Ad Feelders Universiteit Utrecht Ad Feelders ( Universiteit Utrecht ) Pattern Recgnitin 1 / 55 Overview 1 Separable Case 2 Kernel Functins 3 Allwing Errrs (Sft

More information

CHAPTER 3 INEQUALITIES. Copyright -The Institute of Chartered Accountants of India

CHAPTER 3 INEQUALITIES. Copyright -The Institute of Chartered Accountants of India CHAPTER 3 INEQUALITIES Cpyright -The Institute f Chartered Accuntants f India INEQUALITIES LEARNING OBJECTIVES One f the widely used decisin making prblems, nwadays, is t decide n the ptimal mix f scarce

More information

Module 4: General Formulation of Electric Circuit Theory

Module 4: General Formulation of Electric Circuit Theory Mdule 4: General Frmulatin f Electric Circuit Thery 4. General Frmulatin f Electric Circuit Thery All electrmagnetic phenmena are described at a fundamental level by Maxwell's equatins and the assciated

More information

7 TH GRADE MATH STANDARDS

7 TH GRADE MATH STANDARDS ALGEBRA STANDARDS Gal 1: Students will use the language f algebra t explre, describe, represent, and analyze number expressins and relatins 7 TH GRADE MATH STANDARDS 7.M.1.1: (Cmprehensin) Select, use,

More information

Collocation Map for Overcoming Data Sparseness

Collocation Map for Overcoming Data Sparseness Cllcatin Map fr Overcming Data Sparseness Mnj Kim, Yung S. Han, and Key-Sun Chi Department f Cmputer Science Krea Advanced Institute f Science and Technlgy Taejn, 305-701, Krea mj0712~eve.kaist.ac.kr,

More information

EDA Engineering Design & Analysis Ltd

EDA Engineering Design & Analysis Ltd EDA Engineering Design & Analysis Ltd THE FINITE ELEMENT METHOD A shrt tutrial giving an verview f the histry, thery and applicatin f the finite element methd. Intrductin Value f FEM Applicatins Elements

More information

, which yields. where z1. and z2

, which yields. where z1. and z2 The Gaussian r Nrmal PDF, Page 1 The Gaussian r Nrmal Prbability Density Functin Authr: Jhn M Cimbala, Penn State University Latest revisin: 11 September 13 The Gaussian r Nrmal Prbability Density Functin

More information

Multiple Source Multiple. using Network Coding

Multiple Source Multiple. using Network Coding Multiple Surce Multiple Destinatin Tplgy Inference using Netwrk Cding Pegah Sattari EECS, UC Irvine Jint wrk with Athina Markpulu, at UCI, Christina Fraguli, at EPFL, Lausanne Outline Netwrk Tmgraphy Gal,

More information

Lab 1 The Scientific Method

Lab 1 The Scientific Method INTRODUCTION The fllwing labratry exercise is designed t give yu, the student, an pprtunity t explre unknwn systems, r universes, and hypthesize pssible rules which may gvern the behavir within them. Scientific

More information

The blessing of dimensionality for kernel methods

The blessing of dimensionality for kernel methods fr kernel methds Building classifiers in high dimensinal space Pierre Dupnt Pierre.Dupnt@ucluvain.be Classifiers define decisin surfaces in sme feature space where the data is either initially represented

More information

Weathering. Title: Chemical and Mechanical Weathering. Grade Level: Subject/Content: Earth and Space Science

Weathering. Title: Chemical and Mechanical Weathering. Grade Level: Subject/Content: Earth and Space Science Weathering Title: Chemical and Mechanical Weathering Grade Level: 9-12 Subject/Cntent: Earth and Space Science Summary f Lessn: Students will test hw chemical and mechanical weathering can affect a rck

More information

The standards are taught in the following sequence.

The standards are taught in the following sequence. B L U E V A L L E Y D I S T R I C T C U R R I C U L U M MATHEMATICS Third Grade In grade 3, instructinal time shuld fcus n fur critical areas: (1) develping understanding f multiplicatin and divisin and

More information

A Scalable Recurrent Neural Network Framework for Model-free

A Scalable Recurrent Neural Network Framework for Model-free A Scalable Recurrent Neural Netwrk Framewrk fr Mdel-free POMDPs April 3, 2007 Zhenzhen Liu, Itamar Elhanany Machine Intelligence Lab Department f Electrical and Cmputer Engineering The University f Tennessee

More information

COMP 551 Applied Machine Learning Lecture 5: Generative models for linear classification

COMP 551 Applied Machine Learning Lecture 5: Generative models for linear classification COMP 551 Applied Machine Learning Lecture 5: Generative mdels fr linear classificatin Instructr: Herke van Hf (herke.vanhf@mail.mcgill.ca) Slides mstly by: Jelle Pineau Class web page: www.cs.mcgill.ca/~hvanh2/cmp551

More information

ENG2410 Digital Design Sequential Circuits: Part B

ENG2410 Digital Design Sequential Circuits: Part B ENG24 Digital Design Sequential Circuits: Part B Fall 27 S. Areibi Schl f Engineering University f Guelph Analysis f Sequential Circuits Earlier we learned hw t analyze cmbinatinal circuits We will extend

More information

A Matrix Representation of Panel Data

A Matrix Representation of Panel Data web Extensin 6 Appendix 6.A A Matrix Representatin f Panel Data Panel data mdels cme in tw brad varieties, distinct intercept DGPs and errr cmpnent DGPs. his appendix presents matrix algebra representatins

More information

NUMBERS, MATHEMATICS AND EQUATIONS

NUMBERS, MATHEMATICS AND EQUATIONS AUSTRALIAN CURRICULUM PHYSICS GETTING STARTED WITH PHYSICS NUMBERS, MATHEMATICS AND EQUATIONS An integral part t the understanding f ur physical wrld is the use f mathematical mdels which can be used t

More information

Building to Transformations on Coordinate Axis Grade 5: Geometry Graph points on the coordinate plane to solve real-world and mathematical problems.

Building to Transformations on Coordinate Axis Grade 5: Geometry Graph points on the coordinate plane to solve real-world and mathematical problems. Building t Transfrmatins n Crdinate Axis Grade 5: Gemetry Graph pints n the crdinate plane t slve real-wrld and mathematical prblems. 5.G.1. Use a pair f perpendicular number lines, called axes, t define

More information

CS 477/677 Analysis of Algorithms Fall 2007 Dr. George Bebis Course Project Due Date: 11/29/2007

CS 477/677 Analysis of Algorithms Fall 2007 Dr. George Bebis Course Project Due Date: 11/29/2007 CS 477/677 Analysis f Algrithms Fall 2007 Dr. Gerge Bebis Curse Prject Due Date: 11/29/2007 Part1: Cmparisn f Srting Algrithms (70% f the prject grade) The bjective f the first part f the assignment is

More information

SUPPLEMENTARY MATERIAL GaGa: a simple and flexible hierarchical model for microarray data analysis

SUPPLEMENTARY MATERIAL GaGa: a simple and flexible hierarchical model for microarray data analysis SUPPLEMENTARY MATERIAL GaGa: a simple and flexible hierarchical mdel fr micrarray data analysis David Rssell Department f Bistatistics M.D. Andersn Cancer Center, Hustn, TX 77030, USA rsselldavid@gmail.cm

More information

Least Squares Optimal Filtering with Multirate Observations

Least Squares Optimal Filtering with Multirate Observations Prc. 36th Asilmar Cnf. n Signals, Systems, and Cmputers, Pacific Grve, CA, Nvember 2002 Least Squares Optimal Filtering with Multirate Observatins Charles W. herrien and Anthny H. Hawes Department f Electrical

More information

COMP 551 Applied Machine Learning Lecture 11: Support Vector Machines

COMP 551 Applied Machine Learning Lecture 11: Support Vector Machines COMP 551 Applied Machine Learning Lecture 11: Supprt Vectr Machines Instructr: (jpineau@cs.mcgill.ca) Class web page: www.cs.mcgill.ca/~jpineau/cmp551 Unless therwise nted, all material psted fr this curse

More information

k-nearest Neighbor How to choose k Average of k points more reliable when: Large k: noise in attributes +o o noise in class labels

k-nearest Neighbor How to choose k Average of k points more reliable when: Large k: noise in attributes +o o noise in class labels Mtivating Example Memry-Based Learning Instance-Based Learning K-earest eighbr Inductive Assumptin Similar inputs map t similar utputs If nt true => learning is impssible If true => learning reduces t

More information

MODULE FOUR. This module addresses functions. SC Academic Elementary Algebra Standards:

MODULE FOUR. This module addresses functions. SC Academic Elementary Algebra Standards: MODULE FOUR This mdule addresses functins SC Academic Standards: EA-3.1 Classify a relatinship as being either a functin r nt a functin when given data as a table, set f rdered pairs, r graph. EA-3.2 Use

More information

NGSS High School Physics Domain Model

NGSS High School Physics Domain Model NGSS High Schl Physics Dmain Mdel Mtin and Stability: Frces and Interactins HS-PS2-1: Students will be able t analyze data t supprt the claim that Newtn s secnd law f mtin describes the mathematical relatinship

More information

Emphases in Common Core Standards for Mathematical Content Kindergarten High School

Emphases in Common Core Standards for Mathematical Content Kindergarten High School Emphases in Cmmn Cre Standards fr Mathematical Cntent Kindergarten High Schl Cntent Emphases by Cluster March 12, 2012 Describes cntent emphases in the standards at the cluster level fr each grade. These

More information

Lecture 2: Supervised vs. unsupervised learning, bias-variance tradeoff

Lecture 2: Supervised vs. unsupervised learning, bias-variance tradeoff Lecture 2: Supervised vs. unsupervised learning, bias-variance tradeff Reading: Chapter 2 STATS 202: Data mining and analysis September 27, 2017 1 / 20 Supervised vs. unsupervised learning In unsupervised

More information

Admin. MDP Search Trees. Optimal Quantities. Reinforcement Learning

Admin. MDP Search Trees. Optimal Quantities. Reinforcement Learning Admin Reinfrcement Learning Cntent adapted frm Berkeley CS188 MDP Search Trees Each MDP state prjects an expectimax-like search tree Optimal Quantities The value (utility) f a state s: V*(s) = expected

More information

Data Mining: Concepts and Techniques. Classification and Prediction. Chapter February 8, 2007 CSE-4412: Data Mining 1

Data Mining: Concepts and Techniques. Classification and Prediction. Chapter February 8, 2007 CSE-4412: Data Mining 1 Data Mining: Cncepts and Techniques Classificatin and Predictin Chapter 6.4-6 February 8, 2007 CSE-4412: Data Mining 1 Chapter 6 Classificatin and Predictin 1. What is classificatin? What is predictin?

More information

Support-Vector Machines

Support-Vector Machines Supprt-Vectr Machines Intrductin Supprt vectr machine is a linear machine with sme very nice prperties. Haykin chapter 6. See Alpaydin chapter 13 fr similar cntent. Nte: Part f this lecture drew material

More information

8 th Grade Math: Pre-Algebra

8 th Grade Math: Pre-Algebra Hardin Cunty Middle Schl (2013-2014) 1 8 th Grade Math: Pre-Algebra Curse Descriptin The purpse f this curse is t enhance student understanding, participatin, and real-life applicatin f middle-schl mathematics

More information

COMP 551 Applied Machine Learning Lecture 9: Support Vector Machines (cont d)

COMP 551 Applied Machine Learning Lecture 9: Support Vector Machines (cont d) COMP 551 Applied Machine Learning Lecture 9: Supprt Vectr Machines (cnt d) Instructr: Herke van Hf (herke.vanhf@mail.mcgill.ca) Slides mstly by: Class web page: www.cs.mcgill.ca/~hvanh2/cmp551 Unless therwise

More information

Elements of Machine Intelligence - I

Elements of Machine Intelligence - I ECE-175A Elements f Machine Intelligence - I Ken Kreutz-Delgad Nun Vascncels ECE Department, UCSD Winter 2011 The curse The curse will cver basic, but imprtant, aspects f machine learning and pattern recgnitin

More information

Sequential Allocation with Minimal Switching

Sequential Allocation with Minimal Switching In Cmputing Science and Statistics 28 (1996), pp. 567 572 Sequential Allcatin with Minimal Switching Quentin F. Stut 1 Janis Hardwick 1 EECS Dept., University f Michigan Statistics Dept., Purdue University

More information

You need to be able to define the following terms and answer basic questions about them:

You need to be able to define the following terms and answer basic questions about them: CS440/ECE448 Sectin Q Fall 2017 Midterm Review Yu need t be able t define the fllwing terms and answer basic questins abut them: Intr t AI, agents and envirnments Pssible definitins f AI, prs and cns f

More information

Assessment Primer: Writing Instructional Objectives

Assessment Primer: Writing Instructional Objectives Assessment Primer: Writing Instructinal Objectives (Based n Preparing Instructinal Objectives by Mager 1962 and Preparing Instructinal Objectives: A critical tl in the develpment f effective instructin

More information

Lecture 2: Supervised vs. unsupervised learning, bias-variance tradeoff

Lecture 2: Supervised vs. unsupervised learning, bias-variance tradeoff Lecture 2: Supervised vs. unsupervised learning, bias-variance tradeff Reading: Chapter 2 STATS 202: Data mining and analysis September 27, 2017 1 / 20 Supervised vs. unsupervised learning In unsupervised

More information

Document for ENES5 meeting

Document for ENES5 meeting HARMONISATION OF EXPOSURE SCENARIO SHORT TITLES Dcument fr ENES5 meeting Paper jintly prepared by ECHA Cefic DUCC ESCOM ES Shrt Titles Grup 13 Nvember 2013 OBJECTIVES FOR ENES5 The bjective f this dcument

More information

5 th grade Common Core Standards

5 th grade Common Core Standards 5 th grade Cmmn Cre Standards In Grade 5, instructinal time shuld fcus n three critical areas: (1) develping fluency with additin and subtractin f fractins, and develping understanding f the multiplicatin

More information

Writing Guidelines. (Updated: November 25, 2009) Forwards

Writing Guidelines. (Updated: November 25, 2009) Forwards Writing Guidelines (Updated: Nvember 25, 2009) Frwards I have fund in my review f the manuscripts frm ur students and research assciates, as well as thse submitted t varius jurnals by thers that the majr

More information

Differentiation Applications 1: Related Rates

Differentiation Applications 1: Related Rates Differentiatin Applicatins 1: Related Rates 151 Differentiatin Applicatins 1: Related Rates Mdel 1: Sliding Ladder 10 ladder y 10 ladder 10 ladder A 10 ft ladder is leaning against a wall when the bttm

More information

How do scientists measure trees? What is DBH?

How do scientists measure trees? What is DBH? Hw d scientists measure trees? What is DBH? Purpse Students develp an understanding f tree size and hw scientists measure trees. Students bserve and measure tree ckies and explre the relatinship between

More information

Comprehensive Exam Guidelines Department of Chemical and Biomolecular Engineering, Ohio University

Comprehensive Exam Guidelines Department of Chemical and Biomolecular Engineering, Ohio University Cmprehensive Exam Guidelines Department f Chemical and Bimlecular Engineering, Ohi University Purpse In the Cmprehensive Exam, the student prepares an ral and a written research prpsal. The Cmprehensive

More information

Coalition Formation and Data Envelopment Analysis

Coalition Formation and Data Envelopment Analysis Jurnal f CENTRU Cathedra Vlume 4, Issue 2, 20 26-223 JCC Jurnal f CENTRU Cathedra Calitin Frmatin and Data Envelpment Analysis Rlf Färe Oregn State University, Crvallis, OR, USA Shawna Grsspf Oregn State

More information

Enhancing Performance of MLP/RBF Neural Classifiers via an Multivariate Data Distribution Scheme

Enhancing Performance of MLP/RBF Neural Classifiers via an Multivariate Data Distribution Scheme Enhancing Perfrmance f / Neural Classifiers via an Multivariate Data Distributin Scheme Halis Altun, Gökhan Gelen Nigde University, Electrical and Electrnics Engineering Department Nigde, Turkey haltun@nigde.edu.tr

More information

initially lcated away frm the data set never win the cmpetitin, resulting in a nnptimal nal cdebk, [2] [3] [4] and [5]. Khnen's Self Organizing Featur

initially lcated away frm the data set never win the cmpetitin, resulting in a nnptimal nal cdebk, [2] [3] [4] and [5]. Khnen's Self Organizing Featur Cdewrd Distributin fr Frequency Sensitive Cmpetitive Learning with One Dimensinal Input Data Aristides S. Galanpuls and Stanley C. Ahalt Department f Electrical Engineering The Ohi State University Abstract

More information

Resampling Methods. Cross-validation, Bootstrapping. Marek Petrik 2/21/2017

Resampling Methods. Cross-validation, Bootstrapping. Marek Petrik 2/21/2017 Resampling Methds Crss-validatin, Btstrapping Marek Petrik 2/21/2017 Sme f the figures in this presentatin are taken frm An Intrductin t Statistical Learning, with applicatins in R (Springer, 2013) with

More information

Fall 2013 Physics 172 Recitation 3 Momentum and Springs

Fall 2013 Physics 172 Recitation 3 Momentum and Springs Fall 03 Physics 7 Recitatin 3 Mmentum and Springs Purpse: The purpse f this recitatin is t give yu experience wrking with mmentum and the mmentum update frmula. Readings: Chapter.3-.5 Learning Objectives:.3.

More information

Reinforcement Learning" CMPSCI 383 Nov 29, 2011!

Reinforcement Learning CMPSCI 383 Nov 29, 2011! Reinfrcement Learning" CMPSCI 383 Nv 29, 2011! 1 Tdayʼs lecture" Review f Chapter 17: Making Cmple Decisins! Sequential decisin prblems! The mtivatin and advantages f reinfrcement learning.! Passive learning!

More information

PSU GISPOPSCI June 2011 Ordinary Least Squares & Spatial Linear Regression in GeoDa

PSU GISPOPSCI June 2011 Ordinary Least Squares & Spatial Linear Regression in GeoDa There are tw parts t this lab. The first is intended t demnstrate hw t request and interpret the spatial diagnstics f a standard OLS regressin mdel using GeDa. The diagnstics prvide infrmatin abut the

More information

MATCHING TECHNIQUES. Technical Track Session VI. Emanuela Galasso. The World Bank

MATCHING TECHNIQUES. Technical Track Session VI. Emanuela Galasso. The World Bank MATCHING TECHNIQUES Technical Track Sessin VI Emanuela Galass The Wrld Bank These slides were develped by Christel Vermeersch and mdified by Emanuela Galass fr the purpse f this wrkshp When can we use

More information

Activity Guide Loops and Random Numbers

Activity Guide Loops and Random Numbers Unit 3 Lessn 7 Name(s) Perid Date Activity Guide Lps and Randm Numbers CS Cntent Lps are a relatively straightfrward idea in prgramming - yu want a certain chunk f cde t run repeatedly - but it takes a

More information

Particle Size Distributions from SANS Data Using the Maximum Entropy Method. By J. A. POTTON, G. J. DANIELL AND B. D. RAINFORD

Particle Size Distributions from SANS Data Using the Maximum Entropy Method. By J. A. POTTON, G. J. DANIELL AND B. D. RAINFORD 3 J. Appl. Cryst. (1988). 21,3-8 Particle Size Distributins frm SANS Data Using the Maximum Entrpy Methd By J. A. PTTN, G. J. DANIELL AND B. D. RAINFRD Physics Department, The University, Suthamptn S9

More information

T Algorithmic methods for data mining. Slide set 6: dimensionality reduction

T Algorithmic methods for data mining. Slide set 6: dimensionality reduction T-61.5060 Algrithmic methds fr data mining Slide set 6: dimensinality reductin reading assignment LRU bk: 11.1 11.3 PCA tutrial in mycurses (ptinal) ptinal: An Elementary Prf f a Therem f Jhnsn and Lindenstrauss,

More information

Department of Electrical Engineering, University of Waterloo. Introduction

Department of Electrical Engineering, University of Waterloo. Introduction Sectin 4: Sequential Circuits Majr Tpics Types f sequential circuits Flip-flps Analysis f clcked sequential circuits Mre and Mealy machines Design f clcked sequential circuits State transitin design methd

More information

Rangely RE 4 Curriculum Development 5 th Grade Mathematics

Rangely RE 4 Curriculum Development 5 th Grade Mathematics Unit Title Dctr We Still Need t Operate... Length f Unit 12 weeks Fcusing Lens(es) Inquiry Questins (Engaging Debatable): Structure Systems Standards and Grade Level Expectatins Addressed in this Unit

More information

A Quick Overview of the. Framework for K 12 Science Education

A Quick Overview of the. Framework for K 12 Science Education A Quick Overview f the NGSS EQuIP MODULE 1 Framewrk fr K 12 Science Educatin Mdule 1: A Quick Overview f the Framewrk fr K 12 Science Educatin This mdule prvides a brief backgrund n the Framewrk fr K-12

More information

Computational modeling techniques

Computational modeling techniques Cmputatinal mdeling techniques Lecture 4: Mdel checing fr ODE mdels In Petre Department f IT, Åb Aademi http://www.users.ab.fi/ipetre/cmpmd/ Cntent Stichimetric matrix Calculating the mass cnservatin relatins

More information

Resampling Methods. Chapter 5. Chapter 5 1 / 52

Resampling Methods. Chapter 5. Chapter 5 1 / 52 Resampling Methds Chapter 5 Chapter 5 1 / 52 1 51 Validatin set apprach 2 52 Crss validatin 3 53 Btstrap Chapter 5 2 / 52 Abut Resampling An imprtant statistical tl Pretending the data as ppulatin and

More information

GENESIS Structural Optimization for ANSYS Mechanical

GENESIS Structural Optimization for ANSYS Mechanical P3 STRUCTURAL OPTIMIZATION (Vl. II) GENESIS Structural Optimizatin fr ANSYS Mechanical An Integrated Extensin that adds Structural Optimizatin t ANSYS Envirnment New Features and Enhancements Release 2017.03

More information

Distributions, spatial statistics and a Bayesian perspective

Distributions, spatial statistics and a Bayesian perspective Distributins, spatial statistics and a Bayesian perspective Dug Nychka Natinal Center fr Atmspheric Research Distributins and densities Cnditinal distributins and Bayes Thm Bivariate nrmal Spatial statistics

More information

Five Whys How To Do It Better

Five Whys How To Do It Better Five Whys Definitin. As explained in the previus article, we define rt cause as simply the uncvering f hw the current prblem came int being. Fr a simple causal chain, it is the entire chain. Fr a cmplex

More information

Aristotle I PHIL301 Prof. Oakes Winthrop University updated: 3/14/14 8:48 AM

Aristotle I PHIL301 Prof. Oakes Winthrop University updated: 3/14/14 8:48 AM Aristtle I PHIL301 Prf. Oakes Winthrp University updated: 3/14/14 8:48 AM The Categries - The Categries is ne f several imprtant wrks by Aristtle n metaphysics. His tpic here is the classificatin f beings

More information

SAMPLING DYNAMICAL SYSTEMS

SAMPLING DYNAMICAL SYSTEMS SAMPLING DYNAMICAL SYSTEMS Melvin J. Hinich Applied Research Labratries The University f Texas at Austin Austin, TX 78713-8029, USA (512) 835-3278 (Vice) 835-3259 (Fax) hinich@mail.la.utexas.edu ABSTRACT

More information

A Correlation of. to the. South Carolina Academic Standards for Mathematics Precalculus

A Correlation of. to the. South Carolina Academic Standards for Mathematics Precalculus A Crrelatin f Suth Carlina Academic Standards fr Mathematics Precalculus INTRODUCTION This dcument demnstrates hw Precalculus (Blitzer), 4 th Editin 010, meets the indicatrs f the. Crrelatin page references

More information

Physics 2B Chapter 23 Notes - Faraday s Law & Inductors Spring 2018

Physics 2B Chapter 23 Notes - Faraday s Law & Inductors Spring 2018 Michael Faraday lived in the Lndn area frm 1791 t 1867. He was 29 years ld when Hand Oersted, in 1820, accidentally discvered that electric current creates magnetic field. Thrugh empirical bservatin and

More information

English 10 Pacing Guide : Quarter 2

English 10 Pacing Guide : Quarter 2 Implementatin Ntes Embedded Standards: Standards nted as embedded n this page are t be cntinuusly spiraled thrughut the quarter. This des nt mean that nging explicit instructin n these standards is t take

More information

ECE 545 Project Deliverables

ECE 545 Project Deliverables ECE 545 Prject Deliverables Tp-level flder: _ Secnd-level flders: 1_assumptins 2_blck_diagrams 3_interface 4_ASM_charts 5_surce_cde 6_verificatin 7_timing_analysis 8_results

More information

Methods for Determination of Mean Speckle Size in Simulated Speckle Pattern

Methods for Determination of Mean Speckle Size in Simulated Speckle Pattern 0.478/msr-04-004 MEASUREMENT SCENCE REVEW, Vlume 4, N. 3, 04 Methds fr Determinatin f Mean Speckle Size in Simulated Speckle Pattern. Hamarvá, P. Šmíd, P. Hrváth, M. Hrabvský nstitute f Physics f the Academy

More information

Kinetic Model Completeness

Kinetic Model Completeness 5.68J/10.652J Spring 2003 Lecture Ntes Tuesday April 15, 2003 Kinetic Mdel Cmpleteness We say a chemical kinetic mdel is cmplete fr a particular reactin cnditin when it cntains all the species and reactins

More information

The Kullback-Leibler Kernel as a Framework for Discriminant and Localized Representations for Visual Recognition

The Kullback-Leibler Kernel as a Framework for Discriminant and Localized Representations for Visual Recognition The Kullback-Leibler Kernel as a Framewrk fr Discriminant and Lcalized Representatins fr Visual Recgnitin Nun Vascncels Purdy H Pedr Mren ECE Department University f Califrnia, San Dieg HP Labs Cambridge

More information

What is Statistical Learning?

What is Statistical Learning? What is Statistical Learning? Sales 5 10 15 20 25 Sales 5 10 15 20 25 Sales 5 10 15 20 25 0 50 100 200 300 TV 0 10 20 30 40 50 Radi 0 20 40 60 80 100 Newspaper Shwn are Sales vs TV, Radi and Newspaper,

More information

1996 Engineering Systems Design and Analysis Conference, Montpellier, France, July 1-4, 1996, Vol. 7, pp

1996 Engineering Systems Design and Analysis Conference, Montpellier, France, July 1-4, 1996, Vol. 7, pp THE POWER AND LIMIT OF NEURAL NETWORKS T. Y. Lin Department f Mathematics and Cmputer Science San Jse State University San Jse, Califrnia 959-003 tylin@cs.ssu.edu and Bereley Initiative in Sft Cmputing*

More information

22.54 Neutron Interactions and Applications (Spring 2004) Chapter 11 (3/11/04) Neutron Diffusion

22.54 Neutron Interactions and Applications (Spring 2004) Chapter 11 (3/11/04) Neutron Diffusion .54 Neutrn Interactins and Applicatins (Spring 004) Chapter (3//04) Neutrn Diffusin References -- J. R. Lamarsh, Intrductin t Nuclear Reactr Thery (Addisn-Wesley, Reading, 966) T study neutrn diffusin

More information

Name: Block: Date: Science 10: The Great Geyser Experiment A controlled experiment

Name: Block: Date: Science 10: The Great Geyser Experiment A controlled experiment Science 10: The Great Geyser Experiment A cntrlled experiment Yu will prduce a GEYSER by drpping Ments int a bttle f diet pp Sme questins t think abut are: What are yu ging t test? What are yu ging t measure?

More information

Sections 15.1 to 15.12, 16.1 and 16.2 of the textbook (Robbins-Miller) cover the materials required for this topic.

Sections 15.1 to 15.12, 16.1 and 16.2 of the textbook (Robbins-Miller) cover the materials required for this topic. Tpic : AC Fundamentals, Sinusidal Wavefrm, and Phasrs Sectins 5. t 5., 6. and 6. f the textbk (Rbbins-Miller) cver the materials required fr this tpic.. Wavefrms in electrical systems are current r vltage

More information

Chemistry 20 Lesson 11 Electronegativity, Polarity and Shapes

Chemistry 20 Lesson 11 Electronegativity, Polarity and Shapes Chemistry 20 Lessn 11 Electrnegativity, Plarity and Shapes In ur previus wrk we learned why atms frm cvalent bnds and hw t draw the resulting rganizatin f atms. In this lessn we will learn (a) hw the cmbinatin

More information

Homology groups of disks with holes

Homology groups of disks with holes Hmlgy grups f disks with hles THEOREM. Let p 1,, p k } be a sequence f distinct pints in the interir unit disk D n where n 2, and suppse that fr all j the sets E j Int D n are clsed, pairwise disjint subdisks.

More information

Admissibility Conditions and Asymptotic Behavior of Strongly Regular Graphs

Admissibility Conditions and Asymptotic Behavior of Strongly Regular Graphs Admissibility Cnditins and Asympttic Behavir f Strngly Regular Graphs VASCO MOÇO MANO Department f Mathematics University f Prt Oprt PORTUGAL vascmcman@gmailcm LUÍS ANTÓNIO DE ALMEIDA VIEIRA Department

More information

Medium Scale Integrated (MSI) devices [Sections 2.9 and 2.10]

Medium Scale Integrated (MSI) devices [Sections 2.9 and 2.10] EECS 270, Winter 2017, Lecture 3 Page 1 f 6 Medium Scale Integrated (MSI) devices [Sectins 2.9 and 2.10] As we ve seen, it s smetimes nt reasnable t d all the design wrk at the gate-level smetimes we just

More information

Associated Students Flacks Internship

Associated Students Flacks Internship Assciated Students Flacks Internship 2016-2017 Applicatin Persnal Infrmatin: Name: Address: Phne #: Years at UCSB: Cumulative GPA: E-mail: Majr(s)/Minr(s): Units Cmpleted: Tw persnal references (Different

More information

TRAINING GUIDE. Overview of Lucity Spatial

TRAINING GUIDE. Overview of Lucity Spatial TRAINING GUIDE Overview f Lucity Spatial Overview f Lucity Spatial In this sessin, we ll cver the key cmpnents f Lucity Spatial. Table f Cntents Lucity Spatial... 2 Requirements... 2 Supprted Mdules...

More information

o o IMPORTANT REMINDERS Reports will be graded largely on their ability to clearly communicate results and important conclusions.

o o IMPORTANT REMINDERS Reports will be graded largely on their ability to clearly communicate results and important conclusions. BASD High Schl Frmal Lab Reprt GENERAL INFORMATION 12 pt Times New Rman fnt Duble-spaced, if required by yur teacher 1 inch margins n all sides (tp, bttm, left, and right) Always write in third persn (avid

More information

Analysis on the Stability of Reservoir Soil Slope Based on Fuzzy Artificial Neural Network

Analysis on the Stability of Reservoir Soil Slope Based on Fuzzy Artificial Neural Network Research Jurnal f Applied Sciences, Engineering and Technlgy 5(2): 465-469, 2013 ISSN: 2040-7459; E-ISSN: 2040-7467 Maxwell Scientific Organizatin, 2013 Submitted: May 08, 2012 Accepted: May 29, 2012 Published:

More information

Accreditation Information

Accreditation Information Accreditatin Infrmatin The ISSP urges members wh have achieved significant success in the field t apply fr higher levels f membership in rder t enjy the fllwing benefits: - Bth Prfessinal members and Fellws

More information