Nearby in the stack

Iterative Model-Based Reinforcement Learning Using Simulations in the Differentiable Neural Computer · arXivDesk

1906.07248Jun 17, 2019Accepted at the Workshop on Multi-Task and Lifelong Reinforcement Learning, 36th International Conference on Machine Learning, Long Beach, California, 2019

Iterative Model-Based Reinforcement Learning Using Simulations in the Differentiable Neural Computer

Adeel Mufti, Svetlin Penkov, Subramanian Ramamoorthy

Machine Learning Machine Learning

arXiv PDF

Abstract

We propose a lifelong learning architecture, the Neural Computer Agent (NCA), where a Reinforcement Learning agent is paired with a predictive model of the environment learned by a Differentiable Neural Computer (DNC). The agent and DNC model are trained in conjunction iteratively. The agent improves its policy in simulations generated by the DNC model and rolls out the policy to the live environment, collecting experiences in new portions or tasks of the environment for further learning. Experiments in two synthetic environments show that DNC models can continually learn from pixels alone to simulate new tasks as they are encountered by the agent, while the agents can be successfully trained to solve the tasks using Proximal Policy Optimization entirely in simulations.

Briefing

What this paper is actually doing

This is a heuristic sketch from the abstract. Sign in for a live briefing.

GraduateNovelty 3/5

Researchers working in cs.LG and adjacent areas.

Claims

The paper addresses: Iterative Model-Based Reinforcement Learning Using Simulations in the Differentiable Neural Computer.
We propose a lifelong learning architecture, the Neural Computer Agent (NCA), where a Reinforcement Learning agent is paired with a predictive model of the environment learned by a Differentiable Neural Computer (DNC).

Methods

Method details are not extracted without an AI key — open the PDF or HTML version.
Author comment: Accepted at the Workshop on Multi-Task and Lifelong Reinforcement Learning, 36th International Conference on Machine Learning, Long Beach, California, 2019

Limitations

This is a heuristic briefing generated without a language model. Treat it as a sketch, not an analysis.

Why it matters

Newly listed under cs.LG. Connect your OpenRouter key to unlock a full arXivDesk briefing.

Open questions

What is the strongest empirical result, and on which baseline?
Which assumption would most change the conclusion if it failed?

Socratic margin

Ask the abstract

Questions are grounded in the title and abstract. If the PDF is required, arXivDesk will say so.

Nearby in the stack

arXiv did not return neighbours for this paper. Try again in a bit.