Informatics and computing I 501 Introduction to Informatics

Informatics and computing I 501 – Introduction to Informatics From swarming to collaborative filtering. http: //www. csml. ucl. ac. uk/images/Netflix_Prize. jpg jbollen@indiana. edu http: //informatics. indiana. edu/jbollen/I 501 Lecture 11 – Fall 2009

Informatics and computing I 501 – Introduction to Informatics: a possible parsing X-Informatics or Computational X Health. HCID Bio- § towards problem solving § beyond computing § into the natural and social § synthesis of information technology jbollen@indiana. edu http: //informatics. indiana. edu/jbollen/I 501 Data & Search Informatics Security Computer Science Social Informatics Geo- Data Mining Complex Systems Music- Chem- Lecture 11 – Fall 2009

Informatics and computing I 501 – Introduction to Informatics Let’s Observe Nature! What do you see? § Plants typically branch out § How can we model that? § Observe the distinct parts § Color them § Assign symbols § Build Model § Initial State: b § b -> a § a -> ab § Doesn’t quite Work! Psilophyta/Psilotum b b a a b jbollen@indiana. edu http: //informatics. indiana. edu/jbollen/I 501 a b a b Lecture 11 – Fall 2009 b b

Informatics and computing I 501 – Introduction to Informatics Complex systems approach: looking at nature § A complex system is any system featuring a large number of interacting components (agents, processes, etc. ) whose aggregate activity is nonlinear § not derivable from the summations of the activity of individual components § Network identity: Components form aggregate structures or functions that requires more explanatory devices than those used to explain the components § Genetic networks, Immune networks, Neural networks, Social insect colonies, Social networks, Distributed Knowledge Systems, Ecological networks § Bottom-up Methodology § Collections of simple units interacting to form a more complex hole § Study of Simple Rules that Produce Complex Behavior § Discovery of Global Patterns of behavior jbollen@indiana. edu http: //informatics. indiana. edu/jbollen/I 501 Lecture 11 – Fall 2009

Informatics and computing I 501 – Introduction to Informatics What about our plant? § An Accurate model requires § Varying angles § Varying stem lengths § Randomness § The Fibonacci Model is similar § Sneezewort: b b a a b b a b a ba a b a Psilophyta/Psilotum b jbollen@indiana. edu http: //informatics. indiana. edu/jbollen/I 501 Lecture 11 – Fall 2009

Informatics and computing I 501 – Introduction to Informatics Fibonacci Numbers! § Rewriting production rules § Initial State: A § A -> B § B -> AB § § § § n=0 : A n=1 : B n=2 : AB n=3 : BAB n=4 : ABBAB n=5 : BABABBAB n=6 : ABBABBAB n=7 : BABABBABBAB § The length of the string is the Fibonacci Sequence § 1 1 2 3 5 8 13 21 34 55 89. . . § Fibonacci numbers in Nature § Livio (2003) The Golden Ratio: The Story of PHI, the World's Most Astonishing Number jbollen@indiana. edu http: //informatics. indiana. edu/jbollen/I 501 Lecture 11 – Fall 2009

Informatics and computing I 501 – Introduction to Informatics Another example: flocking in nature § § Flocking occurs when large groups of animals of the same species form aggregates that behave like a coherent, single entity § Herds, flocks, schools, swarms, humans Properties: § Collective flight, migration, foraging, “drafting” § Coherence: aggregate has its own distinguishable system behavior and form § Adaptive: behavior of aggregate responds and adapts to external events (predators) § Coordination: behavior of individuals seems to be indicative of central control or symbolic/long-range communication, but isn’t jbollen@indiana. edu http: //informatics. indiana. edu/jbollen/I 501 Lecture 11 – Fall 2009

Informatics and computing I 501 – Introduction to Informatics How to model flocking behavior? § § § Describing properties of aggregate behavior will only go so far: § Study shapes of aggregate § Situations in which it occurs § Dynamics, features of behavior § Biologists fixing radios? Lessons from complex systems: § Complex systems behavior: not derivable from the summations of the activity of individual components § Network identity: Components form aggregate structures or functions that requires more explanatory devices than those used to explain the components ~ emergence Bottom-up Methodology: § § Parrish(2002) – Self-organized fish schools Collections of simple units interacting to form a more complex hole Study of Simple Rules that Produce Complex Behavior jbollen@indiana. edu http: //informatics. indiana. edu/jbollen/I 501 Lecture 11 – Fall 2009

Informatics and computing I 501 – Introduction to Informatics Models of flocking behavior § § Boids: Craig Reynolds “Flocks, Herds and schools”, SIGGRAPH 21(4), 1987 Visual model of bird flocks § Lack of centralized control § Lack of symbolic communication General approach: Local computation, i. e. each individual maximizes: § Collision avoidance: steer away from impact § Velocity matching: match speed of neighboring birds § Flock centering: steer towards perceived flock center Flock behavior = emerges from interactions of large groups of such construed individuals jbollen@indiana. edu http: //informatics. indiana. edu/jbollen/I 501 Lecture 11 – Fall 2009

Informatics and computing I 501 – Introduction to Informatics Ant trails: emergent organizaton driven by communication § § § Problem: optimize location and extraction of food source § Lack of centralized control § Lack of symbolic communication General modeling approach: § Local computation leads to higher order emergent computation § Walk algorithm probabilistic, but biased by pheromone concentraion § Ants leave pheromone trail when food is found § Pheromone evaporates with time § Find shortest path Note: § ~ greedy algorithm: hill-climbing on trail strength leads to adaptive, collective behavior § Approaches to address traveling salesman problem: BIOS group: S. Kaufmann (Santa Fe), see also M. Dorigo(2006) Ant Colony Optimization-IEEE Computational Intelligence Magazine for overview jbollen@indiana. edu http: //informatics. indiana. edu/jbollen/I 501 Lecture 11 – Fall 2009

Informatics and computing I 501 – Introduction to Informatics Probabilistic cleaning: ants § 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. 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 1 st Int. Conf. on Simulation of Adaptive Behaviour. 356 -363 (1990). Figure by Marco Dorigo in Real ants inspire ant algorithms jbollen@indiana. edu http: //informatics. indiana. edu/jbollen/I 501 Lecture 11 – Fall 2009

Informatics and computing I 501 – Introduction to Informatics Ant-inspired robots § Rules (Becker et al, 1994) § Move: with no sensor activated move in straight line § Obstacle avoidance: if obstacle is found, turn with a random angle to avoid it and move. § Pick up and drop: Robots can pick up a number of objects (up to 3) § If shovel contains 3 or more objects, sensor is activated and objects are dropped. Robot backs up, chooses new angle and moves. § Results in clustering § The probability of dropping items increases with quantity of items in vicinity Figure from R Beckers, OE Holland, and JL Deneubourg [1994]. “From local actions to global tasks: Stigmergy and collective robotics”. In Artificial Life IV. jbollen@indiana. edu http: //informatics. indiana. edu/jbollen/I 501 Lecture 11 – Fall 2009

Informatics and computing I 501 – Introduction to Informatics becker et al experiments jbollen@indiana. edu http: //informatics. indiana. edu/jbollen/I 501 Lecture 11 – Fall 2009

Informatics and computing I 501 – Introduction to Informatics Luc Steels et al: ant algorithms http: //www. youtube. com/watch? v=93 Lwvux. Dbf. U jbollen@indiana. edu http: //informatics. indiana. edu/jbollen/I 501 Lecture 11 – Fall 2009

Informatics and computing I 501 – Introduction to Informatics Adaptive information systems Swarm Smarts. 78. Scientific American March 2000. ERIC BONABEAU Johan Bollen (1994): adaptive hypertext systems jbollen@indiana. edu http: //informatics. indiana. edu/jbollen/I 501 Lecture 11 – Fall 2009

Informatics and computing I 501 – Introduction to Informatics Recommender systems: general principles • § § People ~ n-dimensional vectors Shameboy § Person = { CD/book purchases, DVDs rented, …} § Vector is a representation of consumer. Entries can be weighted (TFIDF etc) “Vector Space Model” § Calculate similarity of users: § Correlation of user vectors § Cosine similarity § Group consumers according to similarity: clustering Angle: Consumer Similarity § Similar users: discrepancies in vectors are recommendations Used for all sorts of applications § Similar problem to “bad of words” Plastic Operator § Multiple user personalities? § Orthogonality? [Shameboy, Plastic Operator, Figurine, …] § Same = better? ? Buyer 1 [1, 1, 0, 0, 0, …] Buyer 2 [1, 0, 0, …] jbollen@indiana. edu http: //informatics. indiana. edu/jbollen/I 501 Lecture 11 – Fall 2009

Informatics and computing I 501 – Introduction to Informatics Tracking scientists (they are people too!) http: //informatics. indiana. edu/jbollen/PLos. ONEmap André Skupin Borner/Ketan (2004) PNAS 101(1) Highly recommended: http: //www. scimaps. org/ jbollen@indiana. edu http: //informatics. indiana. edu/jbollen/I 501 Lecture 11 – Fall 2009

Informatics and computing I 501 – Introduction to Informatics We’re all ants now? • § § § User vectors: § Represent individual trail/exploration in n-dimension information space Recommender systems: § bias probabilistic exploration paths of users based on others’ actions § Higher probability of following existing trails Analogy: § Set of user vectors + recommender system ~ ant trails § Solving traveling salesman in n dimensions? ; -) Modeling fads, hypes, flashcrowds in cyberspace, self-fulfilling prophecies, but also long tail effects, more optimized exploration of information space? § Which features of recommender systems promote either of the above? § Cf. youtube. com: “other users are watching” vs. batchprocessed recommendations jbollen@indiana. edu http: //informatics. indiana. edu/jbollen/I 501 documents interface recommender Lecture 11 – Fall 2009

Informatics and computing I 501 – Introduction to Informatics Readings: Questions: - Atlantic (2009) “Is google making us stupid”: As a scientist how would you falsify Carr’s theory that “google is changing the way we think”? Has google changed the way you think? (notions of sampling, plagiarism, etc) - Bettencourt (2008), PNAS: The proposed model results in a scenario in which cities undergo cycles of expansion followed by crisis as a result of the exhaustion of resources. Cycle length shortening with each generation. Speculate: where does this process “break”? What’s a way out? jbollen@indiana. edu http: //informatics. indiana. edu/jbollen/I 501 Lecture 11 – Fall 2009

Informatics and computing I 501 – Introduction to Informatics Next week readings 1. Gouth (2009) Training for Peer Review. Science Signaling 2 (85), tr 2. [DOI: 10. 1126/scisignal. 285 tr 2] 2. MONASTERSKY (2005) The number that is devouring science. Chronicle of higher education, Section: Research & Publishing Volume 52, Issue 8, Page A 12 3. Eysenbach G, 2006 Citation Advantage of Open Access Articles. PLo. S Biol 4(5): e 157. doi: 10. 1371/journal. pbio. 0040157 4. Lance Fortnow (2009) Time for Computer Science to Grow Up. Communications of the ACM, august, 52(8) doi: 10. 1145/1536616. 1536631 jbollen@indiana. edu http: //informatics. indiana. edu/jbollen/I 501 Lecture 11 – Fall 2009