XinyuanWangCS/xinyuanwangcs.github.io
A beautiful, simple, clean, and responsive Jekyll theme for academics
Repositories
XinyuanWangCS/COMP3361-2026
XinyuanWangCS/XinyuanWangCS
Config files for my GitHub profile.
XinyuanWangCS/OpenCUA-Website
XinyuanWangCS/AgentNetTool-Website
XinyuanWangCS/OSWorld
[NeurIPS 2024] OSWorld: Benchmarking Multimodal Agents for Open-Ended Tasks in Real Computer Environments
XinyuanWangCS/DeepLearning-Concepts-and-Implementation-main
Based on Pytorch, Linear Regression, MLP, CNN, RNN 本项目包括Pytorch基础操作, Linear Regression, MLP, CNN, RNN(keras实现)； 每个部分包括讲解和代码实现； ipynb文件图片和公式显示有问题，请下载后运行阅读。
XinyuanWangCS/LatentKMeansPrototypeSelection
This is a course project to select a subset of data to build an efficient nearest neighbor classifier. Choosing a representative subset of "prototypes" from the training set is crucial for accelerating nearest neighbor classifiers. This project proposes projecting the data into a latent space using a pretrained embedder.
XinyuanWangCS/Average-Image-Analysis
This project aims to provide a new perspective to understand the trained Convolutional Neural Network(CNN) model and features in its filters. Guided Backporpagation is a typical way of CNN filter visualizaiton, which could generate visualization images for every filter according a given image. The general information of datasets could be represented as its average image. Therefore, comparing the distance of average image and filter visualization image could describe the general information extraction power of a filter. In this way, we could quantify the confusing visualization results as several scalars, called as SimValues. After computing the SimValues of all filters in all layers, we could analyse them by their absolute values, variances and changing trends to get more insights about CNN model.