Reinforcement Learning Based Quantization Strategy Optimal Assignment Algorithm for Mixed Precision
Source abstract
Abstract
The quantization algorithm compresses the original network by reducing the numerical bit width of the model, which improves the computation speed. Because different layers have different redundancy and sensitivity to data bit width. Reducing the data bit width will result in a loss of accuracy. Therefore, it is difficult to determine the optimal bit width for different parts of the network with guaranteed accuracy. Mixed precision quantization can effectively reduce the amount of computation while keeping the model accuracy basically unchanged. In this paper, a hardware-aware mixed precision quantization strategy optimal assignment algorithm adapted to low bit width is proposed, and reinforcement learning is used to automatically predict the mixed precision that meets the constraints of hardware resources. In the state-space design, the standard deviation of weights is used to measure the distribution difference of data, the execution speed feedback of simulated neural network accelerator inference is used as the environment to limit the action space of the agent, and the accuracy of the quantization model after retraining is used as the reward function to guide the agent to carry out deep reinforcement learning training. The experimental results show that the proposed method obtains a suitable model layer-by-layer quantization strategy under the condition that the computational resources are satisfied, and the model accuracy is effectively improved. The proposed method has strong intelligence and certain universality and has strong application potential in the field of mixed precision quantization and embedded neural network model deployment.
Source: openalex_inverted_index · Quality: baseline
Research context
OpenAlex topics: Neural Networks and Applications · Advanced Vision and Imaging · Advanced Neural Network Applications
SciVX discovery tags are terminology-based. OpenAlex topics are source-provided classifications.
References (24)
Show source-provided references
1. OPQ: Compressing deep neural networks with one-shot pruning-quantization
2. Design and implementation of convolutional neural network based on FPGA
3. Research on physical layer security performance of massive MIMO relay system based on mixed-ADC
4. Improving neural network quantization without retraining using outlier channel splitting
5. ReActNet: Towards precise binary neural network with generalized activation functions
6. Same, same but different-recovering neural network quantization error through weight factorization
7. Explaining knowledge distillation by quantifying the knowledge
8. Learned step size quantization
9. PPO-based automated quantization for reram-based hardware accelerator
10. A new mixed precision quantization algorithm for neural networks based on reinforcement learning
11. Activation redistribution based hybrid asymmetric quantization method of neural networks
12. Reinforcement learning based satellite attitude stabilization method for non-cooperative target capturing10.3390/s18124331 ↗
13. Post training 4-bit quantization of convolutional networks for rapid-deployment
14. HAQ: Hardware-aware automated quantization with mixed precision
15. Search what you want: Barrier panelty NAS for mixed precision quantization
16. Accelerator design and implementation for automatic searching neural network
17. Sequential characteristics based operators disassembly quantization method for LSTM layers10.3390/app122412744 ↗
18. Addressing irregularity in sparse neural networks through a cooperative software/hardware approach
19. Y. Ma, S. Y. Bi, and F. Jiao, “A CNN accelerator with high bandwidth storage,” Chinese Invention Patent, CN20210921363.9, Aug. 11, 2021.
20. J. Choi, Z. Wang, S. Venkataramani, I. J. Chuang, and K. Gopalakrishnan, “PACT: Parameterized clipping activation for quantized neural networks,” 2018. doi: 10.48550/arXiv.1805.06085.
21. Deep compression: Compressing deep neural networks with pruning, trained quantization and huffman coding
22. ReLeQ: A reinforcement learning approach for deep quantization of neural networks10.1109/MM.2020.3009475 ↗
23. S. C. Zhou, Y. X. Wu, Z. K. Ni, X. Y. Zhou, H. Wen and Y. H. Zou, “DoReFa-Net: Training low bitwidth convolutional neural networks with low bitwidth gradients,” 2016. doi: 10.48550/arXiv.1606.06160.
24. S. Ye et al., “A unified framework of DNN weight pruning and weight clustering/quantization using ADMM,” 2018. doi: 10.48550/arXiv.1811.01907.