# Dueling Network Architectures for Deep Reinforcement Learning

**Authors:** Ziyu Wang, Tom Schaul, Matteo Hessel, Hado van Hasselt, Marc Lanctot, Nando de Freitas  
**arXiv:** [1511.06581](https://arxiv.org/abs/1511.06581)  
**Date:** 2015-11-20  
**Summary page:** https://www.summarizepaper.com/en/paper/1511.06581v3/

## Key points

- Paper titled "Dueling Network Architectures for Deep Reinforcement Learning" by Ziyu Wang, Tom Schaul, Matteo Hessel, Hado van Hasselt, Marc Lanctot, and Nando de Freitas
- Introduces a novel neural network architecture for model-free reinforcement learning
- Dueling network factors two separate estimators: one for state value function and another for state-dependent action advantage function
- Improves policy evaluation performance with numerous similar-valued actions
- Experimental results show dueling architecture surpasses existing methods in Atari 2600 domain

## Summary

In their paper titled "Dueling Network Architectures for Deep Reinforcement Learning," authors Ziyu Wang, Tom Schaul, Matteo Hessel, Hado van Hasselt, Marc Lanctot, and Nando de Freitas discuss the advancements in using deep representations in reinforcement learning. The authors introduce a novel neural network architecture specifically designed for model-free reinforcement learning. This dueling network effectively generalizes learning across actions by factoring two separate estimators: one for the state value function and another for the state-dependent action advantage function. This approach results in improved policy evaluation performance when faced with numerous similar-valued actions. The experimental results demonstrate that this dueling architecture surpasses existing state-of-the-art methods in the challenging Atari 2600 domain. Overall, this research contributes valuable insights into leveraging deep neural networks to optimize reinforcement learning processes and achieve superior performance outcomes.