Deep Reinforcement Learning with Python: RLHF for Chatbots and Large Language Models

© The Author(s), under exclusive license to APress Media, LLC, part of Springer Nature 2024

N. SanghiDeep Reinforcement Learning with Pythonhttps://doi.org/10.1007/979-8-8688-0273-7_7

7. Improvements to DQN**

Nimish Sanghi¹  

(1)

Bangalore, India

 

This chapter looks at various enhancements and variations to DQN. Specifically, it looks at Prioritized Replay, DDQN (Double Q-Learning), Dueling DQN, NoisyNets DQN, C-51 (Categorical 51-Atom DQN), Quantile Regression DQN, and Hindsight Experience Replay. All the examples in this chapter are coded using PyTorch. This is an optional chapter with each variant of DQN as a standalone topic. You can skip this chapter in the first pass and come back to it when you want to explore specific variants of DQN.

The first ...