teach yourself ant clustering in ten minutes
play

Teach Yourself Ant Clustering in Ten Minutes Karan K. Budhraja - PowerPoint PPT Presentation

Teach Yourself Ant Clustering in Ten Minutes Karan K. Budhraja What to Expect Basic model Brief associated content No mathematical equations! Motivation Interest in nature inspired algorithms applied to Machine Learning


  1. Teach Yourself Ant Clustering in Ten Minutes Karan K. Budhraja

  2. What to Expect ● Basic model ● Brief associated content ● No mathematical equations!

  3. Motivation ● Interest in nature inspired algorithms ○ applied to Machine Learning or Artificial Intelligence ○ explored in personal work ● Clustering ○ unsupervised learning ○ important in context of Information Retrieval

  4. Definitions ● Clustering ○ grouping of objects based on similarity ● Swarm Intelligence ○ use many unintelligent agents to perform intelligent task ○ software agents ○ often nature inspired

  5. Ants MVP! (for this presentation)

  6. Related Work ● Ant Colony Optimization ○ ant foraging behavior ○ ants use chemical trails ○ shortest path emerges

  7. Ant Clustering in Nature ● Cemetery formation ○ form piles of corpses ○ keep nests clean! ● Brood sorting ○ sort minion ants

  8. Ant Clustering (Basic Model) 1. Scatter items randomly in 2D space 2. Ants begin to move 3. Ants pick up items with probability P pick 4. Ants drop items with probability P drop ● P pick increases in dissimilar item locality ● P drop increases in similar item locality

  9. Ant Clustering (Standard Model) ● Basic Model + modifications ○ no communication between ants ■ distributed approach ○ short term memory ■ remember recent drop locations ■ use them while carrying item ■ move towards similar drops

  10. Variations (brief) ● Sense environment complexity ○ complexity based on item similarity in locality ○ agents focus on high complexity areas ● Sense environment / cluster density ● Sense ant odor ● Hybrid with traditional methods ○ K-Means clustering ○ Fuzzy C-Means clustering

  11. Applications ● General benefits ○ linear complexity with #features ○ #clusters / seed not specified ○ inherently parallel ● Anomaly detection ● Exploratory data analysis

  12. Example ● Ant based clustering by Christian Nitschke ● Example image ○ just in case video doesn’t work :)

  13. Future Work ● Comparison with traditional methods ● Exploration of real world applications ● Parameter tuning, sensitivity ● Combination with traditional methods ○ best of both worlds

  14. Further Reading ● Plethora of references in document OR ● Ignore everything and read this ○ Jafar, OA Mohamed, and R. Sivakumar. "Ant-based clustering algorithms: A brief survey." International journal of computer theory and engineering 2.5 (2010): 1793-8201.

Recommend


More recommend