Loading presentation...

Present Remotely

Send the link below via email or IM


Present to your audience

Start remote presentation

  • Invited audience members will follow you as you navigate and present
  • People invited to a presentation do not need a Prezi account
  • This link expires 10 minutes after you close the presentation
  • A maximum of 30 users can follow your presentation
  • Learn more about this feature in our knowledge base article

Do you really want to delete this prezi?

Neither you, nor the coeditors you shared it with will be able to recover it again.


Swarm Intelligence: Aritificial Intelligence of the Future

No description

Akshay Sankar

on 9 December 2015

Comments (0)

Please log in to add your comment.

Report abuse

Transcript of Swarm Intelligence: Aritificial Intelligence of the Future

Drawing from biological principles for engineering and design

I think the biggest innovations of the 21st century will be at the intersection of biology and technology.
-- Steve Jobs
Artificial Intelligence
Machine Learning
Deep Learning

Genetic Algorithms

Neural networks

Robotics and Data Analytics

Collective Artificial Intelligence
Swarm Intellligence

Artificial Intelligence
Optimized and efficient
Turing Test

Swarm Intelligence
Data Analysis
Load balancing
Network routing
Swarm Robotics
Optimization problems
Swarm-based decision-making in healthcare reduces
false diagnosis rates
for breast cancer
outperforms best-performing radiologist
3 different swarm intelligence voting rules were used: quorum vote, weighted quorum vote, and majority vote
Data Analysis
modified version of the ACO algorithm
gathers optimum solution to search space (data set) using ACO techniques
includes weighting function to compare similar data points that cannot be resolved using implicit fitness/objective functions in ACO algorithm
Ant Colony Optmization
Ant Colonies
Foraging ants
release pheromones at a constant rate
give preference for more intense pheromone routes
Pheromones evaporate at a constant rate
Swarm Intelligence: Aritificial Intelligence of the Future
Shinkansen (Bullet) Train of Japan
Originally created sonic boom-like sound
Modeled after beak of Kingfisher
Slices through air silently
10% faster and 15% more energy efficient

Works with any smooth contour
One small piece can hold 700 pounds
Collective AI
Long and convoluted code

Computationally expensive
Use simple rules

Use multiple agents that work together
Form of Collective Artificial Intelligence

emerges from coordinated and random, local
of simple agents in the

Pioneering algorithms
Ant Colony Optimization (ACO)
Artificial Bee Colony (ABC)
ACO algorithm implements
Phermone Updating function
Pheromone Deposition
Pheromone evaporation
Path (Edge) Selection probability function
Computes a set of solutions to get selected path
Attractiveness of path (ie. evaulating quality of food souce)
Pheromon deposition
Traveling Salesman Problem
Shortest route through set of cities
Each city may be visited exactly one time
Artificial Bee Colony
Inspired by foraging behavior of honeybees
3 Basic components
employed bees
unemployed bees
quality of food source
ABC Scheme
abandon poor food sources
visit high-quality food sources more frequently
Businesses seek
Harvard Business Review
How do you think Swarm Intelligence can be used in Business?
Discuss with someone next to you for 4 minutes?
Swarm Robotics
Dr. Vijay Kumar
Motivation and Background
What are the practical applications of swarm intelligence in science, technology and business?

Swarm theory implementations can turn traditionally long and confusing code to relatively simple and equally effective code for computational algorithms and AI systems.

Been used to organize servers and manage network trafficking more efficiently, enable robotics systems to follow a simple set of rules to accomplish a complex task, and optimize current search algorithms.

Further research into swarm intelligence can improve advancements in science and business from faster and more accurate cancer diagnoses to the efficient delegation of tasks.
State-of-Art and Challenges
Swarm intelligence is incredibly powerful for optimization problems
Several solutions and applications from swarm intelligence:
Traveling Salesman Problem
Spatial Planning
Network Routing
Image Diagnostics and Healthcare
Information about pioneering swarm-based algorithms such as ACO and ABC.
I will use the basic principles of Ant Colony Optimization algorithm to run a simulation of a swarm findings its way through a map (possibly a maze)
I will do this using python

Challenges for this project include
Developing a simplified fitness function and other objective functions to evaluate the best path
Developing a valid map (maze) that is complex enough to show the intelligent capabilities of the swarm
Figuring out how to use graphics packages/modules in python.
Learning some graph theory and possibly some regression analysis
Ideas and Alternatives
Ideas re: this project include:
Simulating an intelligent swarm through a map (or maze)
Simulating a swarm of ants solving the Traveling Salesman Problem
Analyzing and compiling the time complexities of various swarm-based algorithms against their traditional counterparts for areas such as:
image diagnostics and feature selection
network routing
clustering and/or data analysis
Alternatives include:
porting certain aspects of swarm-based algorithms into R for data-analysis purposes
I am more comfortable with R
evaluating other modes of bioinspiration and its relation with computer science and technology
novel brain-to-computer interfaces
neural networks
Project Proposal
Problems with Some AI Techniques
General Scheme of ABC Algorithm
Initiation Phase
Population food sources are initialized by control parameters and information from
scout bees
. Solution vectors of n variables are generated and optimized using an
objective function
Employed Bee Phase
Employed bees find a new food source within the vicinity of food sources in memory. These bees evaluate the new food sources using a fitness funciton and apply a greedy selection between the old food sources and the new ones. A new set of food sources are committed to memory.
Unemployed Bee Phase
Onlooker Bees
Using information conveyed by employed bees, onlooker bees select food sources using a probability model. Once a food source is selected become employed bees and carryout the aforementioned processes of the employed bee phase. Positive feedback occurs.
Scout Bees
Randomly select food sources and evaluate chosen food sources using fitness function to find new solutions. Scout bees rule out poor solutions through negative feedback.
General Scheme of ACO Algorithm
Construct Ant Solutions
Creating graph or "map" of solutions. An initial set of solutions is created from a finite set of components. Partial solutions are connected to initialized graph using an probability model. This is an iterative process.
Daemon Actions
Centralized actions of the global system to perform problem-specific actions.
Update Phermones
Good solutions should receive a net increase in phermone value. Likewise, poor solutions should receive a net decrease in phermone values.Fitness funciton is used to evaluate solutions and increase phermone values associated with good solutions. A phermone evaporation function/constant is created to decrease values of all solutions.
Full transcript