人生重要的不是所站的位置
而是所朝的方向
Hi，我是Aoyu，一名来自东京工业大学CS专业的硕士研究生。现作为RioYokota Lab的一员，从事于高性能计算 / 计算机图形学 / 计算机视觉等相关领域的研究。简历
工作、学习之余，我是一个游戏爱好者，同时也非常喜欢电影与健身。
News
- [2022-12-16] Proxy-ISA算法获得ACM Multimedia Asia 2022的Best Paper Award
- [2022-9-17] 硕士阶段的一项研究（Proxy-ISA算法）以第一作者身份被ACM Multimedia Asia 2022接收为Oral Paper
- [2021-9-7] 以Semantic Space Aware Proxies for Deep Metric Learning为主题在Denso研究会上进行了发表
- [2021-3-19] 以深度度量学习中的General Pair Weighting为主题在Denso研究会上进行了发表
- [2020-9-16] 以Non-Uniform SGD为主题在Denso研究会上进行了发表
Experience
2020.Jul-now Denso IT Lab（研究员）
- 研究基于语义距离的深度学习相关的模型与算法，涵盖图像检索、人脸识别、Few-shot Learning及Transfer Learning。
2021.Apr-now 实验室小组课题（研究员）
- 在组内负责优化算法的研究与落地。为了减小Vision Transformer(ViT)等大规模深度学习模型的训练开销，利用基于K-FAC的NGD近似算法加速模型收敛。
2020.Dec-2021.Mar AIST OIL（研究助理）
- 复现非均一SGD的相关论文实验结果，加速大规模神经网络的训练。
2020.Jul-2020.Sep Flocking Simulation（开发人员）
- 为行为预测相关研究提供实验数据，使用Unity对鱼群巡游进行精确模拟。
- Unity项目开源：Flocking
Publications
Informative Sample-Aware Proxy for Deep Metric Learning. ACM MMAsia 2022 (Oral).
Yet another Researcher.
Yet another Life-long Student.
Hi there, I am Aoyu Li, you can call me Lyle. I am a masters student (2020-) in RioYokota Lab, Tokyo Institute of Technology. My research interests lie in Computer Vision, Computer Graphics, and High Performance Computing.
I am a Video Game enthusiast, I like fitness, basketball and movie.
News
- [2022-12-16] The paper "Informative Sample-Aware Proxy for Deep Metric Learning" received Best Paper Award at ACM Multimedia Asia 2022
- [2022-9-17] The Proxy-ISA algorithm proposed during masters research has been accpeted for an oral presentation at ACM Multimedia Asia 2022
- [2021-9-7] Presented at Denso Research Conference on the topic of Semantic Space Aware Proxies for Deep Metric Learning
- [2021-3-19] Presented at Denso Research Conference on the topic of General Pair Weighting for Deep Metric Learning
- [2020-9-16] Presented at Denso Research Conference on the topic of Non-Uniform SGD
Experience
2021.Aug-2021.Oct DATAGRID (Machine Learning Engineer)
- Motion Matching based character animation generation.
- Solving the stability degradation and feature retrieval inefficiency of Motion Matching due to the diversity of motion data with the help of metric learning.
2020.Jul-2020.Sep CG Simulation
- Provide experimental data for research related to behavior prediction and accurate simulation of fish flocking using Unity.
- Unity Project: Flocking
Research
2020.Jul-2022.Mar Denso IT Lab (Researcher)
- Research on models and algorithms related to Deep Learning based on semantic distance, including Image Retrieval, Face Recognition, Few-shot Learning and Metirc Learning.
2021.Apr-2021.Oct Second-Order Optimization (Research Assistant)
- Responsible for the research and implementation of optimization algorithms in the group. To reduce the training overhead of large-scale Deep Learning models such as Vision Transformer (ViT), a K-FAC-based NGD approximation algorithm is used to accelerate model convergence.
2020.Dec-2021.Mar AIST OIL (Research Assistant)
- Implementation of Non-Uniform SGD related methods, accelerate training of large-scale Deep Neural Networks.
