Answers are grounded in this paper and cite their sources.
In this paper, we focus on the weight quantization problem of neural networks, which is essential for practical implementation in control systems. In our previous work, we proposed a weight coefficient quantization method that preserves the intrinsic performance of the neural network as much as possible. However, when a neural network is embedded into a control system, quantization should not only focus on the input-output accuracy but also take into account the system’s dynamic characteristics. To this end, we propose a novel quantization method based on the Markov parameters of the controlled object. Furthermore, we confirm that the proposed approach effectively reduces the network weights while maintaining favorable control performance of the feedforward system through numerical experiments.
No takes yet. Share an insight, caveat, or question.
Tsubone et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: