lecture 22 -inspired
Sections I485/H400 course outlook Assignments: 35% Students will complete 4/5 assignments based on algorithms presented in class Lab meets in I1 (West) 109 on Lab Wednesdays Lab 0 : January 14 th (completed) Introduction to Python (No Assignment) Lab 1 : January 28 th Measuring Information (Assignment 1) Graded Lab 2 : February 11 th L-Systems (Assignment 2) Graded Lab 3: March 25 th Cellular Automata & Boolean Networks (Assignment 3) Graded Lab 4: April 8 th Genetic Algorithms (Assignment 4) Being graded Lab 5: April 22 nd Ant Clustering Algorithm (Assignment 5) Due May 4 th
Readings until now Class Book Nunes de Castro, Leandro [2006]. Fundamentals of Natural Computing: Basic Concepts, Algorithms, and Applications. Chapman & Hall. Chapters 1,2,3,7,8 Chapter 5, all sections Section 7.7, 8.3.1,8.3.6,8.3.8-10 Lecture notes Chapter 1: What is Life? Chapter 2: The Logical Mechanisms of Life Chapter 3: Formalizing and Modeling the World Chapter 4: Self-Organization and Emergent Complex Behavior Chapter 5: Reality is Stranger than Fiction posted online @
final project schedule ALIFE 15 Projects Due by May 6 th in Oncourse ALIFE 15 (14) Actual conference due date: 2016 http://blogs.cornell.edu/alife14nyc/ 8 pages (LNCS proceedings format) http://www.springer.com/computer/lncs?sgwi D=0-164-6-793341-0 Preliminary ideas overdue! Individual or group With very definite tasks assigned per member of group
differences and explanations biological, social and complexity explanations Emergent behavior Intricate structures and behavior from the interaction of many simple agents or rules Examples Cellular Automata, Ant colonies, development, morphogenesis, brains, immune systems, economic markets Mechanism Parallelism, multiplicity, stigmergy, multi-solutions, redundancy Design causes Natural selection, self-organization, epigenetics, culture
based on dead body cleaning ant clustering algorithm (ACA) Very simple rules for colony clean up Pick dead ant. if a dead ant is found pick it up (with probability inversely proportional to the quantity of dead ants in vicinity) and wander. Drop dead ant. If dead ants are found, drop ant (with probability proportional to the quantity of dead ants in vicinity) and wander. Data vector: X x 1 x 2 x 3 x n-1 x n x 1 x 2 x 3 x n-1 x n Lumer, E. D. and Faieta, B. 1994. Diversity and adaptation in populations of clustering ants. In From Animals To Animats 3, pp. 501-508. x 1 x 2 x 3 x n-1 x n Cluster data (N samples) according to ant clean up rules
for multivariate data ant clustering algorithm (ACA) Group n-dimensional data samples in 2-dimensional grid Data vector: X 1 x 1,1 x 1,2 x 1,3 x 1,n-1 x 1,n Distance between two data samples (in original space): ) D( x ) ( i,x j = xi, k x j, k Data vector: X 2 e.g. Euclidean k = 1 x 2,1 x 2,2 x 2,3 x 2,n-1 x 2,n Ants see data points in a certain neighborhood n 2 s 2 : area of neighborhood (side s, radius 1)
using thresholds Clustering rules Pick data sample If there are few similar Drop data sample. If there are many similar Reduces dimensionality No a priori number of clusters Overshoots number of clusters p f d ( x ) i ant clustering algorithm (ACA) Probability of picking up p p ( x ) i = k Probability of dropping ( ) ( ) 1 2 = s x j Neigh 0 ( s s) 1 ( x ) k1 + f ( x, x ) ( x ) 2 f xi if f xi < k2 ( xi ) = 1 otherwise Neighborhood Similarity or density measure p D 1 i j α if otherwise Discrimination factor Improved with d i = k Threshold i 2 f + 2 ( xi ) f ( x ) f > 0 Different moving speeds, Shortterm memory, Behavioral switches Cooling cycle for thresholds, progressive vision, pheromone reinforcement 2 i
The workings ant clustering algorithm (ACA) 1. Project high-dimensional data items onto 2-dimensional grid randomly 2. Distribute N ants randomly on grid 3. repeat For every ant i in colony Compute neighborhood density f(x i ) If ant i is unloaded and its cell is occupied with data item x i then pick up x i with probability p p (x i ) Else if ant i is loaded with x i and its cell is empty drop x i with probability p d (x i ) Move randomly to neighbor cell with no ant 4. Until maximum iterations
by brood sorting Same principle as Clustering Rules Pick data sample of type t If there are few of type t Drop data sample of type t. If there are many of type t p p d p ( x t) i = k sorting with ants 1 k + f 1 t ( x ) i Probability of picking up item of type t ( x t) i = k 2 ft + ( xi ) f ( x ) t i 2 2 f t ( x ) i Probability of dropping item of type t 1 2 = s x j Neight 0 ( s s) D 1 ( x, x ) otherwise i α Neighborhood density of type t j if f > 0
based on ant algorithm sorting swarm-robots Holland O. & Melhuish C. (1999) Stigmergy, Self-organisation, and Sorting in Collective Robotics Journal of Adaptive Behaviour. 5(2). Bristol Robotics Laboratory. See Also: J. L. Deneubourg, S. Goss, N. Franks, A. Sendova-Franks, C. Detrain, L. Chretien. The Dynamics of Collective Sorting Robot-Like Ants and Ant-Like Robots. From Animals to Animats: Proc. of the 1st Int. Conf. on Simulation of Adaptive Behaviour. 356-363 (1990).
bio-inspired collective robotics Box pushing tasks Taxis-based action (reflex translation or rotation in response to stimulus) and kinesthetic-based action (or proprioception) + realignment and repositioning C. Ronald Kube, Chris A. Parker, Tao Wang and Hong Zhang. "Biologically Collective Robotics," Chapter 15 in Recent Developments in Biologically Computing, de Castro, Leandro N. and Von Zuben, Fernando J., editors, Idea Group Publlishing, 456 pages, 2005.
natural organization bee s nets
by stigmergy self-assembly Self-assembly algorithm Agents move randomly on a 3D grid of sites. An agent deposits a brick every time it finds a stimulating configuration. Rule table contains all such configurations A rule table defines a particular self-assembly algorithm. Rule space is very large From E. Bonabeau. Swarm Intelligence.
space-station by dynamic concepts (dynamic-concepts.com) robotic self-assembly Phase 1: Simulating construction rules
by dynamic concepts (phase two) robotic self-assembly Phase 2: prototype robots
swarm cognition and art Vitorino Ramos: Pheromone Fields as Swarm Cognitive Maps Artificial Ants in Digital Image Habitats A strange Metamorphosis [From Kafka 2 Red Ant] epostcard: V.Ramos CVRM-IST [http://alfa.ist.utl.pt/~cvrm/staff/vramos]; June 2001. Created with an Artificial Ant Colony, that uses images as Habitats, being sensible to their gray levels. At the second row, Kafka is replaced as a substrate, by Red Ant. In black, the higher levels of pheromone (a chemical evaporative sugar substance used by swarms on their orientation trought out the trails). It s exactly this artificial evaporation and the computational ant collective group sinergy realocating their upgrades of pheromone at interesting places, that allows for the emergence of adaptation and perception of new images. Only some of the 6000 iterations processed are represented. The system does not have any type of hierarchy, and ants communicate only in indirect forms, through out the sucessive alteration that they found on the Habitat.
Leonel Moura swarm art
Leonel Moura s RAP (Robotic Action Painter) @ The American Museum of Natural History sensors to avoid obstacles, to perceive the presence of visitors near the case, to check the paper, and most important to detect color. Two modes Random until color threshold is detected. Random sketching Random seed from relative direction measured by an onboard compass. Reactive After passing color threshold Does not go back Draws only where color exceeds threshold. Stopping criteria Pattern in color sensor grid signs off at the corner and flashes lights
Dirk Helbing s Group Modeling traffic and human group behavior Vehicles and people modeled as particles in a fluid medium Free traffic: behaves as a gas Particles move freely Congested traffic: behaves as a liquid movement of particles strongly depends on surrounding dynamics Shock waves emerge from density variations Example in congested traffic The velocity change of a vehicle propagates (with a homogenous time delay) in the opposite direction of traffic as downstream vehicle respond to changes in upstream vehicles propagation speed aprox. -15 km/h (In free traffic = free vehicle velocity). D. Helbing: Traffic and related self-driven many-particle systems. Reviews of Modern Physics 73, 1067-1141 (2003).
Dirk Helbing s Group People modeled as self-driven many-particle systems Testing individualistic vs herding behavior as well as environmental solutions Modeling crowd disasters D. Helbing, A. Johansson and H. Z. Al- Abideen (2007) The Dynamics of Crowd Disasters: An Empirical Study. Physical Review E 75, 046109.
exploring similarities across nature Natural design principles self-similar structures Trees, plants, clouds, mountains morphogenesis Mechanism Iteration, recursion, feedback Unpredictability From limited knowledge or inherent in nature? Mechanism Chaos, measurement Emergence, and self-organization Complex behavior from collectives of many simple units or agents Cellular Automata, development, morphogenesis, brains Mechanism Parallelism, multiplicity, multi-solutions, redundancy (open-ended) Evolution Adaptation, novelty, creativity, learning Mechanism Reproduction, transmission, variation, selection Collective behavior, network causality Behavior derived from many inseparable sources Environment, ant colonies, embodiment, epigenetics, culture, immune systems, economic markets Mechanism Interactivity, stigmergy, non-holonomic constraints
readings Next lectures Class Book Nunes de Castro, Leandro [2006]. Fundamentals of Natural Computing: Basic Concepts, Algorithms, and Applications. Chapman & Hall. Chapter 6 Lecture notes Chapter 1: What is Life? Chapter 2: The logical Mechanisms of Life Chapter 3: Formalizing and Modeling the World Chapter 4: Self-Organization and Emergent Complex Behavior Chapter 5: Reality is Stranger than Fiction Chapter 6: Von Neumann and Natural Selection Chapter 7: Modeling Evolution: Evolutionary Computation posted online @ http://informatics.indiana.edu/rocha/ibic