Projects

### Overview: This repository contains a collection of high-quality deep learning tutorials and repositories. It serves as a comprehensive resource for anyone interested in deep learning, providing both theoretical insights and practical implementations. ### Key Features: - State-of-the-art Deep Learning based Papers with their associated codes. - Deep-Learning-Roadmap - DeepLizard - Sebastian Ruder - Blog - Jeremy Jordan - Blog - Excellent Blog - Torchvision Release Notes - The 6 most useful Machine Learning projects of the past year - ResNet Review - Receptive Field Estimation - An overview of gradient descent optimization algorithms - How to decide on learning rate - Overview of State-of-the-art Machine Learning Algorithms per Discipline per Task - Practical Machine Learning - Awesome Machine Learning and AI Courses - UVA Deep Learning II Course - PyTorch Book - Fast.ai Course: Practical Deep Learning for Coders - labmlai: 59 Implementations/tutorials of deep learning papers with side-by-side notes - labml.ai - FightingCV-Paper-Reading - Learn PyTorch for Deep Learning: Zero to Mastery Course - ML Papers Explained - Alpha Signal: Latest Research in Machine Learning - Harvard CS197: AI Research Experiences - The Course Book - Deep learning jupyter notebook book - A Comprehensive Survey on Pretrained Foundation Models: A History from BERT to ChatGPT - interconnects.ai: Great AI Blog Posts & Podcasts - The Fundamental of Modern Deep Learning

Overview: MIScnn is an open-source Python library designed for medical image segmentation using convolutional neural networks and deep learning. It provides a streamlined setup for creating and utilizing state-of-the-art segmentation pipelines in just a few lines of code. Key features: - Support for 2D/3D medical image segmentation - Comprehensive data I/O, preprocessing, and data augmentation - Patch-wise and full image analysis capabilities - Access to a wide range of deep learning models and evaluation metrics - Intuitive model utilization and fast training/prediction - Customizable model, data I/O, preprocessing, and metric settings Tech stack: Python with Keras and Tensorflow as backend. Why it matters: MIScnn simplifies the process of medical image segmentation by providing a user-friendly interface and a set of pre-configured models and tools, making it accessible and efficient for researchers and practitioners alike. Its modular design allows for easy integration of new models and data sources, ensuring flexibility and adaptability to various medical imaging tasks.

Overview: This repository provides a comprehensive statistics and accepted paper list for major NLP conferences, including ACL, EMNLP, NAACL, EACL, and AACL, with direct access to their arXiv links. Key features: - Statistics and accepted paper lists for NLP conferences - Direct access to arXiv links - Updated statistics and acceptance rates Tech stack: Jupyter Notebook Why it matters: This project offers valuable insights into the current state of NLP research, helping developers and researchers stay updated on the latest trends and accepted papers. Repository: roomylee/nlp-papers-with-arxiv Primary language: Jupyter Notebook GitHub stars: 423

### Overview: Distributed and Graph-Based Structure from Motion ### Key features: - Distributed and Graph-Based Structure from Motion (SfM) implementation. - Supports parallel processing for large-scale image datasets. - Utilizes graph-based methods for efficient computation. - Integrates with COLMAP for compatibility with existing pipelines. ### Tech stack: C ### Why it matters: This project provides a robust and efficient solution for large-scale Structure from Motion (SfM) tasks, making it suitable for applications requiring high accuracy and scalability.

Overview: This repository contains the implementation of a Relation-Shape Convolutional Neural Network for Point Cloud Analysis, authored by Yongcheng Liu, Bin Fan, Shiming Xiang, and Chunhong Pan. The project was presented at CVPR 2019 and was recognized as an Oral Presentation and Best Paper finalist. Key features: - A Single-Scale-Neighborhood classification model with an accuracy of 92.38%. - A Multi-Scale-Neighborhood part segmentation model with class mIoU and instance mIoU of 84.18% and 85.81% respectively. Tech stack: Python Why it matters: This project showcases advancements in point cloud analysis and provides a valuable tool for researchers and practitioners in the field of computer vision and machine learning.

### Overview: Resource Bank of Computer Vision, Natural Language Processing, and MLOps is a comprehensive repository designed for students of computer vision, natural language processing, and machine learning operations. It offers a wide range of educational resources, including courses, specializations, and additional resources, making it an invaluable tool for learners seeking to enhance their skills in these fields. ### Key Features: - **Courses**: Includes specialized courses in Machine Learning, Deep Learning, Natural Language Processing, TensorFlow, MLOps, Generative Adversarial Networks, and Mathematics of Data Science. - **Specializations**: Offers specializations in TensorFlow, Data and Deployment, Machine Learning Engineering for Production (MLOps), and Mathematics of Data Science. - **Resources**: Provides additional resources such as tutorials, use cases, and links to online learning platforms. ### Tech Stack: - Jupyter Notebook ### Why It Matters: This initiative compiles educational resources on a daily basis, making it possible for users to get all the data they want in one place, enhancing their learning experience and providing a wealth of resources for students and professionals alike.

Overview: BidMaster Pro 是一款全流程智能招投标平台，通过AI技术实现从招标文件解读到投标书自动生成、投标检查及文档排版的自动化流程。 Key features: - 智能标书生成 - 投标生成 - 投标检查 - 文档输出 - MinerU OCR 集成 Tech stack: Python, FastAPI, React, Electron, TypeScript, ChromaDB, RBAC, REST API Why it matters: BidMaster Pro 提供了从招标文件到投标书的全流程自动化处理，显著提高了工作效率并确保了合规性，助力企业提升竞争力。

Overview: This repository contains a survey paper on system-aware KV cache optimization methods for large language model serving, focusing on improving system metrics without retraining or architecture modification. The survey is part of the ACL 2026 Findings and is organized using a taxonomy based on system behavior, including temporal, spatial, and structural aspects. Key features: - System-aware KV cache optimization methods - Improves system metrics without retraining or architecture modification - System behavior-oriented taxonomy for organizing literature Tech stack: Python Why it matters: This repository provides a comprehensive overview of KV cache optimization techniques for large language model serving, contributing to more efficient and effective system design without the need for retraining or architecture changes. Repository: jjiantong/Awesome-KV-Cache-Optimization GitHub stars: 422 GitHub description: [ACL 2026] Towards Efficient Large Language Model Serving: A Survey on System-Aware KV Cache Optimization

### Overview: Ensembling off-the-shelf models for GAN training aims to leverage pretrained computer vision models to enhance the performance of Generative Adversarial Networks (GANs). This method proposes an effective selection mechanism by probing the linear separability between real and fake samples in pretrained model embeddings, choosing the most accurate model, and progressively adding it to the discriminator ensemble. ### Key features: - **Ensemble of Pretrained Models**: Utilizes a large bank of pretrained vision models to improve GAN training. - **Vision-Aided Discriminator**: Incorporates a vision-aided discriminator to enhance the quality of generated images. - **Probing Linear Separability**: Selects the most accurate model based on its ability to separate real and fake samples. - **Improved Training Efficiency**: Improves GAN training in both limited data and large-scale settings. ### Tech stack: Python ### Why it matters: This method significantly improves the performance of GANs in various applications, particularly in limited sample settings, by leveraging the collective knowledge from pretrained vision models.

Overview: This project is a deep reinforcement learning chatbot that uses a sequence-to-sequence model for structured learning and a policy gradient technique for improved response generation. Key features: - Seq2seq model for sequence-based learning - RL technique for policy gradient to enhance chatbot performance - Pre-trained models for generating structured dialogues - Simulate and generate responses by pre-trained models Tech stack: Python Why it matters: This chatbot demonstrates advanced reinforcement learning techniques for improving dialogue generation, showcasing the potential of combining sequence models with policy gradient methods.

### Overview: PreenCut is an AI-powered video retrieval and clipping tool that uses natural language queries to find and extract relevant segments from media files. It leverages speech recognition and large language models to automate the process, making it efficient and user-friendly. ### Key Features: - **Automatic Speech Recognition**: Transcribes audio content accurately. - **AI-Powered Analysis**: Uses large language models to segment and summarize video content. - **Natural Language Querying**: Finds clips using descriptive prompts like "Find all product demo segments." - **Smart Clipping**: Selects and exports segments as individual files or merged video. - **SRT Export**: Generates subtitles with accurate timestamps and custom line length. - **Batch Processing**: Searches across multiple files for specific topics. - **Re-analysis**: Experiment with different prompts without reprocessing audio. ### Tech Stack: - Python - FFmpeg for video processing - Gradio for the web interface - LLM services (e.g., DeepSeek-V3, DouBao) for AI analysis ### Why It Matters: PreenCut simplifies the process of finding and extracting valuable content from media files, making it easier for users to focus on the most important parts of their videos or audio files. The AI-driven approach ensures accurate and efficient results, making it a valuable tool for anyone looking to analyze and summarize video content.

Overview: Nagisa is a Python module designed for Japanese word segmentation and part-of-speech tagging based on recurrent neural networks. It aims to provide a simple and easy-to-use tool for natural language processing tasks. Key features: - Based on recurrent neural networks - Word segmentation model uses character- and word-level features - POS-tagging model uses tag dictionary information Tech stack: Python Why it matters: Nagisa simplifies the process of Japanese word segmentation and part-of-speech tagging by leveraging advanced neural network techniques, making it accessible for developers and researchers alike.

Overview: This project presents an innovative approach to embedding continuous numerical features in tabular deep learning models, significantly improving their performance. Key features: - Embedding continuous features before mixing them in the main backbone of tabular DL models. - Demonstrates that embeddings for continuous features can significantly enhance the performance of tabular DL models. - Applies to any conventional backbone, including simple MLP with embeddings. - Competitively performs with heavy Transformer-based models while being more efficient. - The overhead associated with embeddings is often affordable, especially on large datasets, with many features, or under strict latency requirements. Tech stack: Python Why it matters: This method transforms continuous features into vectors before mixing them into the main backbone of tabular DL models, leading to improved performance and more efficient models compared to traditional approaches. Repository: yandex-research/rtdl-num-embeddings GitHub stars: 422 GitHub description: (NeurIPS 2022) On Embeddings for Numerical Features in Tabular Deep Learning

Overview: This project is an end-to-end SLAM framework that integrates dense end-to-end SLAM with monocular depth estimation priors and dense rendering for dense visual SLAM. It achieves real-time global optimization of poses and 3D reconstruction. Key features: - Monocular depth estimation priors - Dense differentiable rendering with 3D Gaussian Splatting - Real-time global optimization of poses and 3D reconstruction - Integration with existing camera calibration framework Tech stack: Python Why it matters: This framework provides a powerful solution for dense visual SLAM, integrating with existing SLAM and camera calibration tools, making it suitable for real-time applications and challenging outdoor environments. Repository: ChenHoy/DROID-Splat GitHub stars: 421 GitHub description: End-to-End SLAM with camera calibration, monocular prior integration and dense Rendering

Overview: GeoDiff is an implementation of a Geometric Diffusion Model designed for generating molecular conformation data, presented at ICLR 2022. It offers a framework for molecular conformation generation and includes both official and preprocessed datasets. Key features: - Geometric Diffusion Model for Molecular Conformation Generation - Official dataset GEOM - Preprocessed datasets available in Google Drive - Training instructions provided in config files - Model checkpoints and training logs saved in specified directories - Generation capabilities for entire or part of test sets - Pretrained models available in the `pretrain` branch for reproducibility Tech stack: Python Why it matters: GeoDiff provides a practical solution for generating molecular conformation data, which is crucial in drug discovery and materials science research.

Overview: CoLLiE 是一个全面的框架，用于从零开始训练大型语言模型。它集成了数据预处理、模型微调、训练监控和并行策略，旨在提高训练效率和模型质量。 Key features: - 数据并行 (DP) - 流水线并行 (PP) - 张量并行 (TP) - 零冗余优化器 (ZeRO) - 高效微调方法 (LOMO, LoRA, Flash Attention) - 设计优雅的用户界面 - 高度可定制化 Tech stack: Python Why it matters: CoLLiE 提供了一站式的解决方案，帮助用户高效地从零开始训练大型语言模型，同时保持模型质量和训练效率。 Repository: OpenMOSS/CoLLiE GitHub stars: 421 GitHub description: An efficient framework for collaborative training of large language models

Overview: This project distills pre-trained Transformers models like DistilBERT and GPT-2 for on-device inference, leveraging TensorFlow Lite with Android demo apps. It includes fine-tuned models for question answering and text generation, providing enhanced performance and reduced resource usage. Key features: - Fine-tuned DistilBERT for question answering on the SQuAD dataset. - DistilBERT and DistilGPT2 fine-tuned for on-device text generation. - Android demo apps for offline inference. Tech stack: Java Why it matters: This project simplifies the integration of Transformers models on Android devices, making them accessible for applications requiring on-device inference capabilities, such as language understanding and text generation tasks. Repository: huggingface/tflite-android-transformers Primary language: Java GitHub stars: 421 GitHub description: DistilBERT / GPT-2 for on-device inference thanks to TensorFlow Lite with Android demo apps.

Overview: PyCantonese is a Python library designed for Cantonese linguistics and natural language processing. Key features: - Accessing and searching corpus data - Parsing and conversion tools for Jyutping romanization - Parsing Cantonese text - Stop words - Word segmentation - Part-of-speech tagging Tech stack: Python Why it matters: PyCantonese provides a comprehensive solution for Cantonese linguistics and NLP tasks, making it a valuable tool for researchers, language enthusiasts, and developers working in the field of Chinese language processing.

Overview: This repository contains the code for the paper "Learning to Model the World with Language," presented at ICML 2024. Dynalang is an agent that uses language to predict the future through a multimodal world model. Key features: - Dynalang: An agent that leverages language to predict the future in a multimodal world model. - HomeGrid: An environment for training Dynalang using the HomeGrid task. - Messenger: An environment for training Dynalang using the Messenger task. - VLN: A vision-language model that predicts future states of the environment. - LangRoom: An environment that allows for simultaneous movement and talking. Tech stack: Python. Why it matters: This project demonstrates the application of language models to predict and model the future states of the world, which has potential applications in various fields such as robotics, artificial intelligence, and natural language processing.

### Overview: This repository is a curated collection of resources for Azure OpenAI, Large Language Models (LLMs), and their applications. It includes concise summaries, chronological organization, and monthly updates to keep the list relevant and accessible. ### Key Features: - Concise Summaries - Chronological Organization - Monthly Updates - Applications archived or inactive for over 12 months are listed in `applications.old.md` ### Tech Stack: Python ### Why It Matters: This repository provides a comprehensive and curated collection of resources for Azure OpenAI, Large Language Models, and their applications, focusing on concise summaries, chronological organization, and monthly updates. It is useful for developers and researchers looking to stay informed about the latest developments in the field. ### GitHub Stars: 419 ### Repository: kimtth/azure-openai-llm-notes