Summary of Compressing Deep Reinforcement Learning Networks with a Dynamic Structured Pruning Method For Autonomous Driving, by Wensheng Su et al.
Compressing Deep Reinforcement Learning Networks with a Dynamic Structured Pruning Method for Autonomous Driving
by Wensheng Su, Zhenni Li, Minrui Xu, Jiawen Kang, Dusit Niyato, Shengli Xie
First submitted to arxiv on: 7 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Robotics (cs.RO)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary Deep reinforcement learning has achieved impressive results in complex autonomous driving scenarios. However, these models often consume high memory and computation resources, limiting their deployment in resource-constrained devices. To address this challenge, researchers have developed methods to compress and accelerate deep reinforcement learning (DRL) models, including structured pruning. This approach involves estimating the contribution of each parameter (neuron) to the model’s performance. In this paper, a novel dynamic structured pruning method is introduced, which gradually removes unimportant neurons during training. The proposed approach consists of two steps: training with a group sparse regularizer and removing unimportant neurons using a dynamic pruning threshold. Experimental results show that the proposed method is competitive with existing DRL pruning methods on various environments. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Deep reinforcement learning has helped autonomous vehicles navigate complex scenarios, but these models require a lot of memory and computing power to work well. Researchers are working on ways to make these models smaller and faster so they can be used in devices that don’t have as much power. One way to do this is by getting rid of parts of the model that aren’t important. This paper proposes a new way to do just that, by slowly removing unimportant parts of the model during training. The method uses two steps: first, it trains the model with a special type of regularizer that encourages the model to use only the most important parts, and then it removes those unimportant parts using a dynamic threshold. This approach has shown promise in various environments. |
Keywords
* Artificial intelligence * Pruning * Reinforcement learning
