Learnable Sparse Coding and Its Application to Inverse Problems

可学习的稀疏编码及其在反问题中的应用

• 批准号: 21K17766
• 负责人: 李 想
• 金额: $ 2.41万
• 依托单位: The University of Aizu
• 依托单位国家: 日本
• 项目类别: Grant-in-Aid for Early-Career Scientists
• 财政年份: 2021
• 资助国家: 日本
• 起止时间: 2021-04-01 至 2024-03-31
• 项目状态: 已结题
• 来源: https://kaken.nii.ac.jp/en/grant/KAKENHI-PROJECT-21K17766/
• 关键词: sparse representation; deep learning; optimization; inverse problem; signal processing; learnable sparse coding; neural network; regularization

Beyond requiring the ground-truth solution to complete the iteration procedure, we have achieved the second year’s goal to develop the unsupervised learning algorithms which extend the ability of the Learnable Sparse Coding algorithms which are mainly built with L1-norm iterative shrinkage-thresholding algorithms (ISTA) and the proposed non-convex constraint algorithms. The real-time training and implementation for the inverse problem experiments are performed. Our constraints incorporate both the convex and non-convex ones employed during the first year, while also exploring the integration of recent constraints derived from the latest research efforts.

除了要求基真解完成迭代过程之外，我们已经实现了第二年开发无监督学习算法的目标，该算法扩展了主要由l1范数迭代收缩阈值算法（ISTA）和所提出的非凸约束算法构建的可学习稀疏编码算法的能力。进行了反问题实验的实时训练和实现。我们的约束包括第一年使用的凸约束和非凸约束，同时也探索了来自最新研究成果的近期约束的集成。

项目成果

Exchange Rate Forecasting Using Machine learning: Explore Gains From External Information

• DOI: 10.1145/3490700.3490704
• 发表时间: 2021-09
• 期刊: Proceedings of the 5th International Conference on Algorithms, Computing and Systems
• 作者: Zhenlin Liang;Xiang Li

Staging Emotions in Stock Market Data Using Natural Language Processing with Transformers

使用 Transformers 的自然语言处理在股票市场数据中呈现情绪

• 发表时间: 2022
• 作者: Hiroki Takenaga;Xiang Li
• 通讯作者: Xiang Li