TD-FALCON

TD-FALCON is a fusion architecture that incorporates temporal difference methods and self-organizing neural networks for reinforcement learning with delayed rewards. It learns by creating cognitive codes across sensory input, actions, and rewards.

The implementation of TD-FALCON can be found in the minefield navigation simulator available below, alongside with its reactive version R-FALCON, and a backpropagation network based Q-learning system.

     Download Minefield Navigation Simulator (2008) (written in JAVA)

ARAM API and Benchmark Program

Adaptive Resonance Associative Map (ARAM) is a neural network model that can be used for pattern recognition, predictive modeling, and associative pattern storage and recall. It can be seen as a compressed form of ARTMAP network. I developed ARAM as part of my PhD project and has since improved it over the past years. The latest version comes complete with voting capabilities and Application Program Interface (API) that allows one to build applications as well as a benchmarking program designed for running intensive experiments.

     Download vARAM (2003) (written in C)

ARTMAP Source Codes

... that I implemented while I was a graduate student at the Department of Cognitive and Neural Systems, Boston University.

     CMU AI Repository
     Original at CNS, Boston University

Talks Handout

     Advances in Adaptive Resonance Theory. Invited Plenary Talk at SCIS & ISIS 2006, Tokyo, 23 September 2006.

     Intelligent Technologies for Media Fusion and Analysis. Open Lecture at I2R-NTU Joint Lab Symposium, Singapore, 4 August 2006.

     Artificial Intelligence Technologies for Web Intelligence. Guest lecture to Singapore-MIT Alliance (CS) Program, Singapore, 11 October 2002