Title: Towards Interpretable, Transparent and Unbiased AI

Date: Tuesday, December 11 2018

Time: 4:30PM - 6:30PM (ET)

Location: CCB 312A

Ramprasaath R. Selvaraju

Ph.D. Student in Computer Science

School of Interactive Computing 

Georgia Institute of Technology

ramprs.github.io

Committee:

Dr. Devi Parikh (Advisor, School of Interactive Computing, Georgia Institute of Technology)

Dr. Dhruv Batra (School of Interactive Computing, Georgia Institute of Technology)

Dr. Stefan Lee (School of Interactive Computing, Georgia Institute of Technology)

Dr. Been Kim (Sr. Research Scientist, Google Brain)

Abstract:

Deep networks have enabled unprecedented breakthroughs in a variety of computer vision tasks. While these models enable superior performance, their increasing complexity and lack of decomposability into individually intuitive components makes them hard to interpret. Consequently, when today's intelligent systems fail, they fail spectacularly disgracefully, giving no warning or explanation.

Towards the goal of making deep networks Interpretable, Transparent and Unbiased, in my thesis I will present my work on building algorithms that provide explanations for decisions emanating from deep networks in order to — 

1. understand why the model did what it did,

2. diagnose network errors,

3. help users build appropriate trust, and

4. enable knowledge transfer between humans and AI. 

In my proposed work, I will show how we can leverage explanations to teach AI systems to correct unwanted biases learned during training, thus improving visual grounding in these systems and making them more trustworthy.