Online Robustness Training for Deep Reinforcement Learning

11/03/2019
by   Marc Fischer, et al.
0

In deep reinforcement learning (RL), adversarial attacks can trick an agent into unwanted states and disrupt training. We propose a system called Robust Student-DQN (RS-DQN), which permits online robustness training alongside Q networks, while preserving competitive performance. We show that RS-DQN can be combined with (i) state-of-the-art adversarial training and (ii) provably robust training to obtain an agent that is resilient to strong attacks during training and evaluation.

READ FULL TEXT
POST COMMENT

Comments

There are no comments yet.

Authors

page 1

page 2

page 3

page 4

12/11/2017

Robust Deep Reinforcement Learning with Adversarial Attacks

This paper proposes adversarial attacks for Reinforcement Learning (RL) ...
05/28/2021

Objective Robustness in Deep Reinforcement Learning

We study objective robustness failures, a type of out-of-distribution ro...
12/29/2018

Learn to Interpret Atari Agents

Deep Reinforcement Learning (DeepRL) models surpass human-level performa...
12/13/2021

Contextual Exploration Using a Linear Approximation Method Based on Satisficing

Deep reinforcement learning has enabled human-level or even super-human ...
08/14/2020

Defending Adversarial Attacks without Adversarial Attacks in Deep Reinforcement Learning

Many recent studies in deep reinforcement learning (DRL) have proposed t...
07/14/2021

Plan-Based Relaxed Reward Shaping for Goal-Directed Tasks

In high-dimensional state spaces, the usefulness of Reinforcement Learni...
06/28/2019

Learning to Cope with Adversarial Attacks

The security of Deep Reinforcement Learning (Deep RL) algorithms deploye...
This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.