Dileep George: Cortex, CAPTCHAs and Connectomics

The ultimate goal of A.I. research is to build machines that exceed the flexibility and dynamism of the human brain. Currently, the predominant approach in A.I. is to use unlimited data to solve narrowly defined problems. To progress towards human-like intelligence, A.I. benchmarks will need to be extended to focus more on data efficiency, flexibility of reasoning, and transfer of knowledge between tasks -- the constraints on a solution to a problem are as important as the solution itself.

In the first part of the talk, I will describe how these principles are reflected in the way we cracked text-based CAPTCHAs, a class of problems that demonstrate the dynamism of human perception. Capturing the inductive biases of the visual cortex can be critical for achieving human-level flexibility.  I will describe a generative model, Recursive Cortical Network (RCN), that captures many of the organizational properties of the visual cortex. RCN cracks text-based CAPTCHAs in a fundamental way because it is able to generatively segment out the characters with very little training data, potentially making text-based CAPTCHAs obsolete. In addition, RCN demonstrates excellent one-shot  recognition and generation on multiple standard benchmarks, shows superior occlusion-reasoning, and outperforms deep learning on multiple scene-text recognition benchmarks while requiring 300-fold less training data.

In the second part of the talk I will try to establish connections of RCN to cortical circuits. Connectomics is an important endeavor for understanding the human brain, and model-based top-down approaches are complementary to the bottom up data-gathering approach. I will describe the broad connections of RCN to cortical anatomy and give an overview of our ongoing work on creating a "virtual connectome" derived from the generative model. 

Bio: Dileep George is a co-founder of Vicarious, an AI company that is building a unified algorithmic architecture to achieve human level intelligence in vision, language, and motor control. He has authored 29 patents and several influential papers on the mathematics of brain circuits. Before Vicarious, Dileep was CTO of Numenta, an AI company he cofounded with Jeff Hawkins and Donna Dubinsky. Dileep's research on hierarchical models of the brain earned him a PhD in Electrical Engineering from Stanford University and a position as a Research Fellow at the Redwood Neuroscience Institute. He earned his MS in EE from Stanford and his BS from IIT in Bombay.