Devendra R. Parkar

Devendra R. Parkar

Research Assistant

Arizona State University

Welcome!

I am a PhD student studying Computer Science at Arizona State University. I am currently working as a researcher at Daymude Lab under Prof. Joshua Daymude and previously worked at LabV2 and SOPSLab.

My broad research interest lies in understanding and building complex systems with distributed agents that can learn. I wish to study biological phenomena that arise due to interactions of such dynamic agents, specifically, the brain networks as a collective of distributed agents.

My current research explores techniques from multi-agent optimization, stochastic processes and reinforcement learning to build and study multi-agent behaviors. Previously I looked at ways to incorporate symbolic logic and human knowledge to improve machine learning algorithms.

Interests
  • Complex Systems
  • Distributed Systems
  • Multi-agent Reinforcement Learning
  • Learning Theory
Education
  • PhD in Computer Science, (Expected) 2029

    Arizona State University

  • M.S. in Computer Science, 2024

    Arizona State University

  • B.E. in Computer Engineering, 2018

    University of Mumbai

Projects

*
IARPA HAYSTACK - Anamoly Detection
Our team at LabV2(ASU) [collaboration with Leidos Inc.] works on modeling agent movements in real-world maps. We use knowledge graphs to model the maps in combination with symbolic rule learning and heuristic based graph traversal algorithms to model and learn agent trajectories(both anamolous and normal).
IARPA HAYSTACK - Anamoly Detection
EvoSOPS: Evolving global collective behaviors using restricted agents
Using genetic algorithms to find distributed local algorithms that generate global collective behaviors
EvoSOPS: Evolving global collective behaviors using restricted agents
PyReason-Gym: Symbolic logic simulator for Reinforcement Learning
A temporal and annotated-logic based simulation proxy that enables interpretable reinforcement learning
PyReason-Gym: Symbolic logic simulator for Reinforcement Learning
DMAR: Decentralized Multi-agent Rollout Algorithm
A new decentralized and online reinforcement learning algorithm for vehicle routing problem in unmapped environments
DMAR: Decentralized Multi-agent Rollout Algorithm
Neuro-Evolutionary Swarms
Explores evolutionary algorithms to build neural network robot controllers for building collective behaviors
Neuro-Evolutionary Swarms