# Deep Episodic Value Iteration for Model-based Meta-Reinforcement Learning

## TL;DR

This paper introduces Deep Episodic Value Iteration (DEVI), a deep meta-reinforcement learning method that combines neural networks with a non-parametric model-based approach, enabling rapid adaptation to new reward and transition dynamics.

## Contribution

DEVI is a novel deep meta-RL algorithm that learns a similarity metric for model-based RL, trained end-to-end, and capable of one-shot transfer in high-dimensional state spaces.

## Key findings

- DEVI achieves rapid adaptation to changes in reward and transition structures.
- DEVI outperforms traditional model-free methods in transfer tasks.
- DEVI demonstrates effective one-shot transfer in high-dimensional environments.

## Abstract

We present a new deep meta reinforcement learner, which we call Deep Episodic Value Iteration (DEVI). DEVI uses a deep neural network to learn a similarity metric for a non-parametric model-based reinforcement learning algorithm. Our model is trained end-to-end via back-propagation. Despite being trained using the model-free Q-learning objective, we show that DEVI's model-based internal structure provides `one-shot' transfer to changes in reward and transition structure, even for tasks with very high-dimensional state spaces.