# Playing Atari with Deep Reinforcement Learning

**Authors:** Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Alex Graves, Ioannis Antonoglou, Daan Wierstra, Martin Riedmiller  
**arXiv:** [1312.5602](https://arxiv.org/abs/1312.5602)  
**Date:** 2013-12-19  
**Summary page:** https://www.summarizepaper.com/en/paper/1312.5602v1/

## Key points

- Groundbreaking study "Playing Atari with Deep Reinforcement Learning"
- Introduces novel deep learning model for learning control policies directly from high-dimensional sensory input
- Model is a convolutional neural network trained with variant of Q-learning
- Takes raw pixels as input to output value function estimating future rewards
- Applied to seven Atari 2600 games, outperformed previous approaches on six games and surpassed human expert on three
- Signifies significant advancement in artificial intelligence research
- Showcases potential of deep reinforcement learning in solving complex tasks based solely on visual inputs
- Published at NIPS Deep Learning Workshop 2013
- Highlights capabilities of deep learning models in mastering challenging tasks through autonomous decision-making processes
- Opens up new possibilities for developing intelligent systems without explicit programming or human intervention

## Summary

The groundbreaking study "Playing Atari with Deep Reinforcement Learning" introduces a novel deep learning model that revolutionizes the field by successfully learning control policies directly from high-dimensional sensory input using reinforcement learning. The model is a convolutional neural network trained with a variant of Q-learning and takes raw pixels as input to output a value function that estimates future rewards. The researchers applied this cutting-edge method to seven Atari 2600 games from the Arcade Learning Environment without any adjustments to the architecture or learning algorithm. Remarkably, their model outperformed all previous approaches on six of the games and even surpassed a human expert on three of them. This achievement signifies a significant advancement in artificial intelligence research and showcases the potential of deep reinforcement learning in solving complex tasks based solely on visual inputs. Published at the NIPS Deep Learning Workshop 2013, this study sheds light on the capabilities of deep learning models in mastering challenging tasks through autonomous decision-making processes. The success of this approach opens up new possibilities for developing intelligent systems capable of adapting to diverse environments and tasks without explicit programming or human intervention.