Cortical Learning Algorithms (CLAs) are an attempt by Numenta Inc. to create a computational model of the neocortex of the brain. The goal of this project was to demonstrate that adding a higher level region implementing CLAs, which is being fed input from a lower level region implementing CLAs, helps in improving prediction accuracy in comparison to using just one region as described in the CLAs paper.
This was used to translate/project a high dimensional sparse vector to a high dimensional dense integer vector, appropriate for storage in an Integer Sparse Distributed Memory. The idea used was inspired from Random Projection technique of translation of vectors in high-dimensional spaces.
In this project we are trying to use a Sparse-Autoencoder with a Logistic Regression classifier to build a self-taught learning system for gene expression data. We use a huge amount of data to train the unsupervised step which will help the classification step to solve multiple problems.
I worked on a NASA funded project to understand how can we assist NASA astronauts in completing their tasks while on deep space exploration missions. With the help of interviews of astronauts and a cognitive architecture (SIGMA) and deep neural networks we were able to give cognitive capabilities to a software agent and accomplish the task of helping astronauts in completing their Intra-vehicular and Extra-vehicular Assignments. This project was completed at nFlux AI and the technology built is now being used in manufacturing industry to help workers.
The hierarchical nature of neocortex served as an inspiration to build a predictive model based on Numenta Inc.'s Cortical Learning Algorithms to provide signals for trading for a futures trading platform built at Intelletic Trading Systems. I lead the development of the idea as their Head Researcher and consultant to build a highly profitable trading strategy which is currently deployed and trading.