Research & Innovation

The Research and Development Group at Cortexica has been operating for over 10 years, advancing its expertise in various fields such as, Artificial Intelligence, Deep Learning, Machine Learning, Geometric and Statistical techniques. This drives sophisticated AI solutions for our clients and simply put; “World-class, best-of-breed technology”.

Publications

Cover well-known journals (PAMI, JOSA, Phys Rev E, PloS Computational Biology/Biology, Neuroimage, Royal society journals, etc.) and respected conferences such as NIPS, CoSyNe, AISTATS, ECCV, ICCV, etc.

The group is structured into, i) Applied, and ii) Theoretical teams, with a commercial focus and support for our clients’ AI roadmaps.

Applied Research

Infrastructure Optimization

Utilizing differential geometry, we have shown how rankings on a set of such images can be improved by using Wasserstein distances. Further on, bringing on ideas from tensor factorization, we have shown that these images can be encoded to save storage footprint along with producing encodings that have at par retrieval performance as that demonstrated by advanced methodologies such as Fisher encodings.

Ranking of Images

On the very end of applied research, we have productized a deep learning based recommendation framework that can provide suggestions to end-users based on their personal preference of colour, texture, style for a variety of apparels.

Video Behavioural Analysis

Our “pillar networks” for video based action recognition have demonstrated state-of-the-art predictions on data-sets that have varying camera viewing angles, video quality, etc.

Theoretical Research

Hybrid approach combining Differential Geometry with Statistical Bayesian techniques

Our ICML workshop papers have concentrated on fusing differential geometry based parallel transport with variational Bayes, apart from coming up with a Bayesian belief updating scheme to predict spatiotemporal dynamics such as seizures that originate in the cortex of epileptic patients.

Reinforcement learning for human decision making

In other unpublished work, we have looked into active inference, a neuroscience aided reinforcement learning framework, for agents that are spread on a spatiotemporal field – using PDE analysis and optimal transport theory. This has significant consequences for reinforcement learning, deep and otherwise, which has always ignored the spatial structures of agents.

Discover the recent publications from our Research team

Download here