Using Deep RL to teach machines an understanding of the world

Work in Progress!

Image

Description

Work with Viviane Clay in Dr. Peter Konig's group at RTG Computational Cognition, Osnabruck, Germany. Inspired by Turing's idea to 'try to model a child's mind rather than an adult's', we aim to make agents learn concepts in an unsupervised way, using curiosity based exploration. This could eventually lead to language grounding, without any explicit supervision. To achieve this, Deep Reinforcement Learning is used to make the agent explore Obstacles Tower Environment using a curiosity module proposed by Pathak et al.

How

  • Deep Reinforcement Learning
  • Curiosity based Learning
  • Unity MLAgents

Github Link