First Lecture of Machine Learning. Hung-yi Lee

Similar documents
Single Neuron. Hung-yi Lee

Lecture 1: Numerical Integration The Trapezoidal and Simpson s Rule

a dt a dt a dt dt If 1, then the poles in the transfer function are complex conjugates. Let s look at f t H t f s / s. So, for a 2 nd order system:

CSE 245: Computer Aided Circuit Simulation and Verification

Chapter 3 Common Families of Distributions

PFC Predictive Functional Control

Microscopic Flow Characteristics Time Headway - Distribution

SOLUTIONS. 1. Consider two continuous random variables X and Y with joint p.d.f. f ( x, y ) = = = 15. Stepanov Dalpiaz

Ma/CS 6a Class 15: Flows and Bipartite Graphs

EE 350 Signals and Systems Spring 2005 Sample Exam #2 - Solutions

AR(1) Process. The first-order autoregressive process, AR(1) is. where e t is WN(0, σ 2 )

On the Speed of Heat Wave. Mihály Makai

MEM 355 Performance Enhancement of Dynamical Systems A First Control Problem - Cruise Control

Double Slits in Space and Time

Voltage v(z) ~ E(z)D. We can actually get to this wave behavior by using circuit theory, w/o going into details of the EM fields!

Applied Statistics and Probability for Engineers, 6 th edition October 17, 2016

Physics 160 Lecture 3. R. Johnson April 6, 2015

CPSC 211 Data Structures & Implementations (c) Texas A&M University [ 259] B-Trees

Economics 302 (Sec. 001) Intermediate Macroeconomic Theory and Policy (Spring 2011) 3/28/2012. UW Madison

fakultät für informatik informatik 12 technische universität dortmund Petri Nets Peter Marwedel TU Dortmund, Informatik /10/10

4.1 The Uniform Distribution Def n: A c.r.v. X has a continuous uniform distribution on [a, b] when its pdf is = 1 a x b

Institute of Actuaries of India

5. An object moving along an x-coordinate axis with its scale measured in meters has a velocity of 6t

EE 434 Lecture 22. Bipolar Device Models

What does the data look like? Logistic Regression. How can we apply linear model to categorical data like this? Linear Probability Model

Midterm exam 2, April 7, 2009 (solutions)

46. Let y = ln r. Then dy = dr, and so. = [ sin (ln r) cos (ln r)

UNIT #5 EXPONENTIAL AND LOGARITHMIC FUNCTIONS

Ensamble methods: Boosting

MGM 562 Probability Theory [Teori Kebarangkalian]

Ensamble methods: Bagging and Boosting

An Indian Journal FULL PAPER. Trade Science Inc. A stage-structured model of a single-species with density-dependent and birth pulses ABSTRACT

3(8 ) (8 x x ) 3x x (8 )

Poisson process Markov process

Description of the MS-Regress R package (Rmetrics)

Ratio-Product Type Exponential Estimator For Estimating Finite Population Mean Using Information On Auxiliary Attribute

Transfer function and the Laplace transformation

Chapter 17 Handout: Autocorrelation (Serial Correlation)

General Article Application of differential equation in L-R and C-R circuit analysis by classical method. Abstract

6.1 Integration by Parts and Present Value. Copyright Cengage Learning. All rights reserved.

The transition:transversion rate ratio vs. the T-ratio.

Embedded Systems 5 BF - ES - 1 -

3.4 Repeated Roots; Reduction of Order

Review Lecture 5. The source-free R-C/R-L circuit Step response of an RC/RL circuit. The time constant = RC The final capacitor voltage v( )

Nikesh Bajaj. Fourier Analysis and Synthesis Tool. Guess.? Question??? History. Fourier Series. Fourier. Nikesh Bajaj

Innova Junior College H2 Mathematics JC2 Preliminary Examinations Paper 2 Solutions 0 (*)

First derivative analysis

Chapter 5 The Laplace Transform. x(t) input y(t) output Dynamic System

INTRODUCTION TO MACHINE LEARNING 3RD EDITION

= x. I (x,y ) Example: Translation. Operations depend on pixel s Coordinates. Context free. Independent of pixel values. I(x,y) Forward mapping:

Boyce/DiPrima 9 th ed, Ch 2.1: Linear Equations; Method of Integrating Factors

I) Title: Rational Expectations and Adaptive Learning. II) Contents: Introduction to Adaptive Learning

CS Homework Week 2 ( 2.25, 3.22, 4.9)

READING ASSIGNMENTS. Signal Processing First. Fourier Transform LECTURE OBJECTIVES. This Lecture: Lecture 23 Fourier Transform Properties

CptS 570 Machine Learning School of EECS Washington State University. CptS Machine Learning 1

(10) (a) Derive and plot the spectrum of y. Discuss how the seasonality in the process is evident in spectrum.

Vector autoregression VAR. Case 1

The stable manifold approach for optimal swing up and stabilization of an inverted pendulum with input saturation

Coherence and interactions in diffusive systems. Lecture 4. Diffusion + e-e interations

ON DETERMINATION OF SOME CHARACTERISTICS OF SEMI-MARKOV PROCESS FOR DIFFERENT DISTRIBUTIONS OF TRANSIENT PROBABILITIES ABSTRACT

Asymptotic Equipartition Property - Seminar 3, part 1

Voltage v(z) ~ E(z)D. We can actually get to this wave behavior by using circuit theory, w/o going into details of the EM fields!

What Ties Return Volatilities to Price Valuations and Fundamentals? On-Line Appendix

Real time estimation of traffic flow and travel time Based on time series analysis

whereby we can express the phase by any one of the formulas cos ( 3 whereby we can express the phase by any one of the formulas

Fourier. Continuous time. Review. with period T, x t. Inverse Fourier F Transform. x t. Transform. j t

Online Convex Optimization Example And Follow-The-Leader

Final Exam : Solutions

Reliability of Technical Systems

Elementary Differential Equations and Boundary Value Problems

8. Queueing systems. Contents. Simple teletraffic model. Pure queueing system

CHAPTER CHAPTER14. Expectations: The Basic Tools. Prepared by: Fernando Quijano and Yvonn Quijano

Control System Engineering (EE301T) Assignment: 2

Learning a Class from Examples. Training set X. Class C 1. Class C of a family car. Output: Input representation: x 1 : price, x 2 : engine power

Age (x) nx lx. Age (x) nx lx dx qx

Lecture 2: Current in RC circuit D.K.Pandey

Chapter 3: Fourier Representation of Signals and LTI Systems. Chih-Wei Liu

of Manchester The University COMP14112 Hidden Markov Models

Bias in Conditional and Unconditional Fixed Effects Logit Estimation: a Correction * Tom Coupé

Frequency Response. Lecture #12 Chapter 10. BME 310 Biomedical Computing - J.Schesser

Lecture 37 (Schrödinger Equation) Physics Spring 2018 Douglas Fields

Note: For all questions, answer (E) NOTA means none of the above answers is correct.

Section. Problem Representation. Substation. Protection Device. protection equipments. Substation. Client. EPDS divided in blocks connected by

Inverse Fourier Transform. Properties of Continuous time Fourier Transform. Review. Linearity. Reading Assignment Oppenheim Sec pp.289.

Routing in Delay Tolerant Networks

Charging of capacitor through inductor and resistor

Introduction D P. r = constant discount rate, g = Gordon Model (1962): constant dividend growth rate.

EXERCISE - 01 CHECK YOUR GRASP

Pattern Classification (VI) 杜俊

EE 315 Notes. Gürdal Arslan CLASS 1. (Sections ) What is a signal?

An recursive analytical technique to estimate time dependent physical parameters in the presence of noise processes

Introduc)on to Bayesian methods (con)nued) - Lecture 16

10. If p and q are the lengths of the perpendiculars from the origin on the tangent and the normal to the curve

3-2-1 ANN Architecture

Math 106: Review for Final Exam, Part II. (x x 0 ) 2 = !

7.4 QUANTUM MECHANICAL TREATMENT OF FLUCTUATIONS *

Midterm Examination (100 pts)

1. Consider a pure-exchange economy with stochastic endowments. The state of the economy

H is equal to the surface current J S

( A) ( B) ( C) ( D) ( E)

Transcription:

Firs Lcur of Machin Larning Hung-yi L

Larning o say ys/no Binary Classificaion

Larning o say ys/no Sam filring Is an -mail sam or no? Rcommndaion sysms rcommnd h roduc o h cusomr or no? Malwar dcion Is h sofwar malicious or no? Sock rdicion Will h fuur valu of a sock incras or no wih rsc o is currn valu? Binary Classificaion

Eaml Alicaion: Sam filring f : X Y { ys, no} E-mail Sam No sam 2 f ys 2 (h://sam-filr-rviw.onrviws.com/) f no

Eaml Alicaion: Sam filring f : X Y { ys, no} Wha dos h funcion f look lik? y f ys no P P ys ys 0.5 0.5 How o sima P(ys )?

Eaml Alicaion: Sam filring To sima P(ys ), collc amls firs.. Earn fr fr Ys (Sam) Som words frqunly aar in h sam.g., fr 2 Win fr Ys (Sam) Us h frquncy of fr o dcid if an -mail is sam 3. Talk Ming. No (No Sam) Esima P(ys fr = k) fr is h numbr of fr in -mail

Rgrssion In raining daa, hr is no - mail conaining 3 fr. (ys fr ) (ys fr = ) = 0.4 (ys fr = 0 ) = 0. Frquncy of Fr ( fr ) in an -mail Problm: Wha if on day you rciv an -mail wih 3 fr.

Rgrssion f( fr ) = w fr + b (f( fr ) is an sima of (ys fr ) ) Sor w and b (ys fr ) Rgrssion Frquncy of Fr ( fr ) in an -mail

Rgrssion f( fr ) = w fr + b Th ouu of f is no bwn 0 and (ys fr ) Rgrssion Frquncy of Fr ( fr ) in an -mail Problm: Wha if on day you rciv an -mail wih 6 fr.

Logi ln fr vrical lin: Probabiliy o b sam (ys fr ) () is always bwn 0 and vrical lin: logi() logi ln

Logi f ( fr ) = w fr + b (f ( fr ) is an sima of logi() ) fr fr vrical lin: Probabiliy o b sam (ys fr ) () is always bwn 0 and vrical lin: logi() logi ln

Logi Sor w and b fr f 3 fr w 3 b.5 logi ln.5 0.87 > 0.5, so ys f ( fr ) = w fr + b (f ( fr ) is an sima of logi() ) f fr w fr b 0 fr ln 0 0.5 ys vrical lin: logi() logi ln

Mulil Variabls Considr wo words fr and hllo comu (ys fr, hllo ) () logi ln hllo fr

Mulil Variabls Considr wo words fr and hllo comu (ys fr, hllo ) () logi ln f fr, hllo Rgrssion w fr w2 hllo b hllo fr

Mulil Variabls Of cours, w can considr all words {, 2, N } in a dicionary b w w w f N N N 2 2 2, b w w N w w w 2 N 2 is o aroima logi() N ys 2, P :

Logisic Rgrssion w b aroima logi ln : P ys If h robabiliy = or 0, ln(/-) = +infiniy or infiniy Can no do rgrssion, 2 N Th robabiliy o b sam is always or 0. aars 3 ims 2 aars 0 im N aars im P ys 2 N 3 0

Logisic Rgrssion b w b w b w ln Sigmoid Funcion

Logisic Rgrssion Ys (Sam) 7 0 3 2 N b w clos o b w No (no Sam) 2 2 b w 2 clos o 0

Logisic Rgrssion Ys No 2 N faur w w 2 w N wb b bias w This is a nuron in nural nwork. b 0

Mor han saying ys/no Muliclass Classificaion

Mor han saying ys/no Handwriing digi classificaion This is Muliclass Classificaion

Mor han saying ys/no Handwriing digi classificaion Simlify h qusion: whhr an imag is 2 or no Dscrib h characrisics of inu objc 2 Each il corrsonds o on dimnsion in N h faur faur of an imag

Mor han saying ys/no Handwriing digi classificaion Simlify h qusion: whhr an imag is 2 or no 2 2 or no N

Mor han saying ys/no Handwriing digi classificaion Binary classificaion of, 2, 3 If y 2 is h ma, hn h imag is 2. y or no 2 y 2 2 or no N y 3 3 or no

This is no good nough

Limiaion of Logisic Rgrssion w w 2 2 b a ys no a a 0.5 0.5 ys no w w2 2 b 0 0 Inu Ouu 2 0 0 No 0 Ys 0 Ys No Ys No

So w nd nural nwork Inu Layr Layr 2 Layr L Ouu y 2 y 2 N y M D mans many layrs

Thank you for your lisning!

Andi

Mor rfrnc h://www.ccs.nu.du/hom/vi/ach/mlcours/2_ GD_REG_on_NN/lcur_nos/logisic_rgrssion_l oss_funcion/logisic_rgrssion_loss.df h://mahgochas.blogso.w/20/0/why-is-rrorfuncion-minimid-in.hml hs://cs.nyu.du/~yann/alks/lcun-2007207- nonconv.df h://www.cs.columbia.du/~bli/fogm/lcurs/glms.df h://grgor.chruala.m/ars/ml4nl/linarclassifirs.df