BROWSE
Detail
Conference Paper
강화된 보상함수를 사용하는 양자화된 확률 경사 Langevin 동역학 학습 방정식
Cited - time in 
Share 
Share
- Issue Date
- 2023-06
- Citation
- 대한전자공학회 학술 대회 (하계) 2023, pp.1-4
- Publisher
- 대한전자공학회
- Language
- Korean
- Type
- Conference Paper
- Abstract
- In this paper, we propose the learning equation based on a quantized stochastic gradient Langevin dynamics (QSGLD) with an enhanced compensation function. Naturally, the quantized optimization technique serves stochastic optimization property allowing to improve optimization performance. However, the quantization process does not apply to optimization caused by early paralysis, which occurs when a quantization parameter is insufficient to operate. The proposed enhancement function, the quantized learning equation, can escape the early paralysis and represent improved optimization performance. Experimental results show that the proposed methodology is effective.
- KSP Keywords
- Enhancement function, Equation based, Langevin dynamics, Optimization performance, Optimization techniques, Quantization Parameter, Quantization process, Quantized optimization, Stochastic gradient, Stochastic optimization, compensation function