TripDataset Machine Learning Project This project is a complete implementation of machine learning pipelines applied to the TripDataset, focusing on data preprocessing, classification, and regression tasks, including: 🧹 Data preprocessing and cleaning (handling missing values, outlier detection, normalization, and feature engineering) 🤖 Model training for classification and regression (various ML algorithms for categorical and continuous prediction tasks) 📊 Performance evaluation and metrics (accuracy, F1-score, RMSE, and other evaluation techniques) 🔍 Exploratory data analysis and visualization (insightful plots for feature relationships, distribution, and model performance)

ViT-torch: Vision Transformer on CIFAR-10 (PyTorch) This project is a complete implementation of Vision Transformer (ViT) applied to small-scale datasets (especially CIFAR-10), including: 🎯 Model implementations with various configurations (native ViT, ResNet+ViT hybrid, different patch/heads/blocks setups, Stochastic Depth/DropPath, etc.) 🌹 Training and evaluation scripts (with learning rate schedulers: Warmup/Linear/Cosine/Constant-Cosine/Warmup-Constant-Cosine) 🧩 Data augmentation (RandomCrop+Paste, MixUp, CutMix, RandAugment, and batch random augmentation) 📈 Visualization and analysis (attention maps, attention distance, gradient rollout, feature maps, positional embedding similarity)

🎯 Voice Activity Detection (VAD), or voice endpoint detection, identifies time segments in an audio signal containing speech. This is a critical preprocessing step for automatic speech recognition (ASR) and voice wake-up systems. This project lays the groundwork for my upcoming ASR project 🤭. 📈 Workflow Overview: The VAD pipeline processes a speech signal as follows:Preprocessing, Framing, Windowing, Feature Extraction, Binary Classification, Time-Domain Restoration 🍻 Project Highlights: I conducted extensive experiments comparing frame division methods (frame length and shift) and model performances, with rich visualizations. For details, see the report in ‘vad/latex/’. If you’re interested in voice technologies, let’s connect! 🔗 For more details, please visit my blog VAD

Establishment and solution of mathematical optimization model This project is a lab of the course “Linear Optimization and Convex Optimization”. It discusses a classic optimization problem, the Water filling problem. Please refer to the project description file for details. In this project, I transformed the original problem into a classic optimization problem according to the mathematical derivation in the description file, and implemented two optimization algorithms, the gradient descent method and the Newton method, and proposed a Binary-search algorithm for the original problem. At the same time, I built two linear search modes and did a lot of comparative experiments. Please refer to the report file for details. In this project, I also compared my algorithm with Monkey-search as required. As the saying goes, 1xxxxx monkeys can’t write Shakespeare’s works. I am currently working on model optimization and convergence analysis. If you are interested in this, please come and communicate with me!

In this project, I explored some solutions to the OOD (out-of-distribution) problem with my partner. For an introduction to the OOD problem, please see this blog. In this project, we used the simplest image recognition convolutional neural network LeNet. For the detailed structure, please refer to the LeNet.svg file in the warehouse. The basic data set we used is also very simple, which is the most primitive handwritten digit recognition: MNIST. I am mainly responsible for the construction, parameter adjustment, training and deployment of the neural network, and my partner is mainly responsible for the data augmentation part. At the same time, we referred to some cutting-edge methods for solving OOD problems, such as IRM, for a detailed introduction, please refer to Zhihu article. Moreover, we also referred to the implementation of the IRM algorithm and did some analysis. The implementation of IRM has also been attached to the warehouse. Last but not least, in listening to the collective report of the project, we also gained some good ideas for OOD problems given by many other groups. If you are also interested in this issue, welcome to discuss with me!

This project implements two types of clustering algorithms, K-means and GMM. Data description: Four sets of data are given. The first two sets are simple low-dimensional data that can be directly visualized, and the last two sets are 128-dimensional high-dimensional data. In this project, I discussed and analyzed various situations such as the initialization mode of GMM, whether high-dimensional data needs dimensionality reduction and dimensionality reduction methods, and K-means convergence judgment, and conducted corresponding comparative experiments. The shortcoming of this project is that I did not give a comparative analysis with the results of directly calling the sk-learn library. If I have time later, I will make up for it.