arXiv cs.LG· E. Riveros (Institute of Computing, State University of Campinas, Campinas, Brazil), D. Vega-Oliveros (Institute of Science and Technology, Federal University of Sao Paulo, Sao Jose dos Campos, Brazil), A. Soriano-Vargas (Universidad de Ingenieria y Tecnologia, Lima, Peru), A. Rocha (Institute of Computing, State University of Campinas, Campinas, Brazil)·· 1 天前AI 评分22
基于扩散模型合成数据预训练提升活动识别性能
Diffusion-Based Synthetic Data Pretraining for Enhancing Activity Recognition
AI 导读
研究者提出用扩散模型生成合成传感器数据窗口，对 CABiGRU 采用两阶段训练：先在合成数据上预训练，再在真实数据上微调。在 DEO（drinking/eating/other）数据集上，该流程取得 90.6% 的平衡准确率，优于强监督基线，缓解了少数类欠拟合问题。
来源：arXiv cs.LG · arxiv.org