Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Ai swarm intelligence


Published on

Inspired from social behavior of insects and other creatures, artificial intelligence community has learned a series of techniques commonly known as Swarm Intelligence. Here we give a 15 minute introduction to this area.

Published in: Technology
  • Be the first to comment

Ai swarm intelligence

  1. 1. Introduction to Swarm Intelligence What can we learn from natural intelligence? Venkatesh Vinayakarao Rekha Tokas Haroon Rashid (In the order of presentation)
  2. 2. Swarms Loosely connected interactive agents Collective behavior Swarms • Display collective behavior • Have loosely connected interactive agents Why Swarm? • Seek food • Migrate • Defense (even attack) 2
  3. 3. Concepts  Stigmergy – Stimulation by work  No direct communication  Agents react to environment  work that does not depend on specific agents  Eg  Ant colonies find shortest path to food and maximum distance from colony entrances to dispose dead ants (midden piles).  Queen ant reproduces. Worker ants (of whom, queen is the mother), fetch food and dispose waste.  Queen does not give any orders. Queen has no authority or decision making control.  Emergence  Complex patterns from simple interactions Eg., structures of termite colonies  Self Organization  Feedback Oriented  Multiple Interactions based on simple rules 3
  4. 4. What is Swarm Intelligence?  “any attempt to design algorithms or distributed problem-solving devices inspired by the collective behavior of social insect colonies and other animal societies” [Bonabeau etal.]  Swarm  A loosely structured collection of interacting agents  Agents:  Individuals that belong to a group (but are not necessarily identical)  They contribute to and benefit from the group  They can recognize, communicate, and/or interact with each other  The instinctive perception of swarms is a group of agents in motion – but that does not always have to be the case. 4
  5. 5. Understanding Swarm Behavior • How do these aggregations work? • Selfish Herd Theory (W.D.Hamilton 1971 – citations > 2000) • Reduce predation risk • Who is protected in the center? Who is at the periphery? • Models for simulation 5
  6. 6. Algorithms & Applications  Few Algorithms  Ant Colony Optimization  Particle Swarm Optimization  Intelligent Water Drops  The Bees  Bat  Termites  Sample Applications  Traveling Salesman  Crown Simulation  Ad-hoc Networks 6
  7. 7. Swarm Intelligence in Termites  After human beings, Only Termites are the living beings who can make standing structure 100 times bigger than its original size.  They build complex structure without having blueprint or outside intervention. 7
  8. 8. Swarm Intelligence in Termites  Not centralized. Don’t follow any order i.e Termites have no supervisors.  Stigmergy  Kind of implicit communication i.e they observe each other’s changes to the environment and act accordingly). 8
  9. 9. System inspired by Termites • Inspired by termites, Harvard University Researchers have designed a construction crew of tiny robots able to build complicated structures (Towers, Castle & Pyramids) without blueprints. • They are 8 inches long and 4.5 inches wide and have pinwheeled shaped tyres. • “Every robot acts independently but together they will end up building what you want.” said team leader Justin Werfel. • To sense its surroundings, each robot is equipped with an infrared sensor, an ultrasound sensor & an accelerometer. • The robots can sense the bricks they carry and the other robots nearby. 9
  10. 10. System inspired by Termites  Each robot can walk around the structure until it sees something that needs to be done and then does it.  A human user need to only design a structure. Software automatically generates the rules that guide the robots.  They usually can recognize mistakes they make and correct them.  Advantage : These can be extremely useful in situations where human intervention is difficult, dangerous such as building structures in space and in disaster zones. 10
  11. 11. SI in Ad hoc networks • Probabilistic based algorithm (PERA) for Ad hoc networks firstly proposed in 2002 , and this gave an edge to TERMITE based algorithm proposed in 2003[13]. • Termite based routing algorithm achieve better adaptively, lower control overhead, and better packet delivery than contemporary solutions. • Each node maintains pheromone table. Network Topology Pheromone Table at Node S 11
  12. 12. Recent developments in SI • U.S. Military is applying SI techniques to control of unmanned vehicles. • SI techniques are applied to load balancing in telecommunication networks[7]. • Entertainment industry is applying SI techniques for battle and crowd scenes. E.g., used in Lord of the Rings • Medical Research is trying SI based controls for nanobots to fight cancer[6]. Unmanned Vehicles Nanobot on Brain cells 12
  13. 13. SI Concerns o Convergence is guaranteed but Time to Convergence is uncertain. o Not suitable for time critical applications E.g., nuclear reactor temperature controller. o Parameter Tuning. o Most of the parameters are problem dependent. E.g., PID(Proportional Integral derivative) controller used in real world control problems.[1] o Stagnation: Premature convergence to a local optimum. o Caused due to lack of central coordination. o Multi-Objective optimization problems. o E.g., Optimize f1(x) = x1 and f2(x) = x2/ax1. o Can we design agent-level behaviours in order to obtain a certain desired behaviour at the collective-level ? 13
  14. 14. Conclusion 14 In Conclusion • Heuristics to solve difficult optimization problems. • Has wide variety of applications. • Basic theme of Natural Computing: Observe nature, mimic nature.
  15. 15. References 7. Ant-based Load Balancing in Telecommunications Networks, HP Labs Technical Reports. 8. A Probabilistic Emergent Routing Algorithm for Mobile Ad Hoc Networks, Baras, Mehta. 9. Geometry for the Selfish Herd, W. D. Hamilton. 10. Swarm Smarts, Bonabeau, Theraulaz. 11. Swarm Intelligence: From Natural to Artificial Systems, Bonabeau, Dorigo, Theraulaz. 12. Swarm Intelligence, Introduction & Applications, Christian Blum, Daniel Merkle. 13. TERMITE: A Swarm Intelligent Routing Algorithm For Mobile Wireless Networks. M. H. Roth. Ph.D. thesis, Cornell University 2005. 15 1. Auto Tuning of PID Controller Using Swarm Intelligence, M. H. T. Omar, W. M. Ali, M. Z. Mostafa. 2. Semantic Web Reasoning by Swarm Intelligence, Kathrin Dentler, Christophe Gu eret, and Stefan Schlobach. 3. DCT-Based Robust Watermarking with Swarm Intelligence Concepts, Hsiang- Cheh Huang , Kaohsiung, Yueh-Hong Chen, Guan-Yu Lin. 4. Robot Swarms in an Uncertain World: Controllable Adaptability, Olga Bogatyreva, Alexandr Shillerov. 5. xTune 6. Video on Nanobots: A cancer treatment revolution powered by tiny robot
  16. 16. Q & A Thank You! 16
  17. 17. 17 Annexure
  18. 18. 18 Particle Swarm Optimization Source: Professor David Wolfe Corne's Talk on PSO Goal: Find the deepest part of the ocean to catch the fish.
  19. 19. 19 Particle Swarm Optimization Source: Professor David Wolfe Corne's Talk on PSO Step1: Place the particles. Step2: Initialize velocities. Step3: Particiles find the best neighbours and move towards them in discrete time. Step4: Repeat Step2 (find a better often slower velocities) and Step3. Challenge: Parameter Tuning!
  20. 20. 20 BAT Algorithm How? Control flying velocity and frequency of sound generation. Why? Search for food.
  21. 21. 21 Comparison of Algorithms Source: Comparison of algorithms, Khan, Sahai, 2012. Conclusion: Bat algorithm outperforms all other algorithms for training feed forward Neural networks!