Akarsh Kumar

Welcome! I want to understand intelligence and all the processes that emerge it.

Research Interests

Broadly, I want to emerge intelligence by creating+analyzing embodied AI agents that reason+learn more intelligently in increasingly complex environments.
To this end, I employ AI Generating Algorithms (AI-GAs) (meta-learning architectures, learning algorithms, and environments) and Open-Ended quality-diverse evolutionary algorithms.

I'm espescially interested in sample efficiency, extrapolative generalization, intelligent exploration, and specialized neural architectures.
I'm also interested in generating effective learning environments to automate curriculum design.

Bio

I'm a Ph.D. student at MIT CSAIL Embodied Intelligence working with Phillip Isola.
My research is funded by the NSF GRFP!

I received my B.S. from UT Austin ECE.
Senior year, I did NN architecture research in VITA with Atlas Wang.
Senior year summer, I did neuroevolution and RL research in LARG with Peter Stone and Risto Miikkulainen.
Sophomore and junior year, I worked with Alexander Huth on visually estimating contact forces in hand-object interaction.

Before that, I went to highschool at the Arkansas School for Mathematics, Sciences, and the Arts.

Here is my CV.