Projects

### Overview: A project that uses Keras and Theano to generate jazz music through deep learning, developed in 36 hours at a hackathon. ### Key features: - Generates jazz music using deep learning techniques. - Built with Keras and Theano libraries. - Uses a two-layer LSTM model trained on MIDI files. - Available on SoundCloud for listening. ### Tech stack: Python ### Why it matters: Demonstrates the potential of deep learning in generating human-like music, showcasing AI's capability to create complex art forms.

### Overview: This repository is dedicated to providing deep learning-based articles, papers, and repositories for recommendation systems. It covers various techniques such as relational stacked denoising autoencoders, collaborative deep learning, recurrent autoencoders, convolutional matrix factorization, and more. The repository aims to enhance understanding and practical application of deep learning in recommendation systems. ### Key Features: - **Relational Stacked Denoising Autoencoder for Tag Recommendation** by Hao Wang, Xingjian Shi, and Dit-Yan Yeung (AAAI 2015) - **Collaborative Deep Learning for Recommender Systems** by Hao Wang, Naiyan Wang, and Dit-Yan Yeung (KDD 2015) - **Collaborative Recurrent Autoencoder: Recommend while Learning to Fill in the Blanks** by Hao Wang, Xingjian Shi, and Dit-Yan Yeung (NIPS 2016) - **Convolutional Matrix Factorization for Document Context-Aware Recommendation** by Donghyun Kim, Chanyoung Park, Jinoh Oh, Seungyong Lee, Hwanjo Yu, RecSys 2016 - **A Neural Autoregressive Approach to Collaborative Filtering** by Yin Zheng et al. (Proceedings of the 22nd International Conference on Machine Learning (ICML) 2015) - **Hybrid Recommender System based on Autoencoders** by Florian Strub (2016) - **Deep content-based music recommendation** by Aaron

Overview: NeuralCoref is a fast, production-ready coreference resolution pipeline for spaCy, utilizing a neural network to enhance spaCy's coreference resolution capabilities. Key features: - NeuralCoref integrates spaCy's NLP pipeline for coreference resolution. - It uses a neural network to improve coreference resolution accuracy. - It is written in Python/Cython and comes with a pre-trained statistical model for English only. Tech stack: Python/Cython Why it matters: NeuralCoref offers a significant improvement in coreference resolution accuracy by leveraging neural networks, making it a valuable tool for developers working on NLP tasks requiring coreference resolution.

Overview: Thinc is a lightweight deep learning library that offers a functional programming API for composing models, compatible with popular frameworks like PyTorch, TensorFlow, and MXNet. It supports type checking, wrapping models, and a flexible configuration system. Key features: - Type-checking with custom types and `mypy` plugin - Support for PyTorch, TensorFlow, and MXNet models - Functional programming approach to model definition - Optional custom infix notation via operator overloading - Extensible backends - Config system for describing trees of objects and hyperparameters - Choice of extensible backends Tech stack: Python Why it matters: Thinc simplifies deep learning by providing a type-checked, functional programming API that allows developers to easily compose models using their preferred framework, making it ideal for integrating with existing projects and reducing the learning curve for new models.

Overview: This tutorial provides a comprehensive guide to implementing the _Show, Attend, and Tell_ model using PyTorch for image captioning. The objective is to generate descriptive captions for images, leveraging the attention mechanism to focus on relevant parts of the image. Key features: - Implementing the _Show, Attend, and Tell_ model - Using PyTorch for neural network implementation - Visual examples of generated captions on test images - Transfer learning from a pretrained model Tech stack: PyTorch Why it matters: This tutorial offers a practical introduction to image captioning using PyTorch, making it accessible for developers interested in applying deep learning to visual tasks.

### Overview: 本项目汇集了大模型(LLMs)、计算机视觉与感知算法、深度学习基础与框架、自动驾驶、智慧医疗等行业专题，以及手撕项目代码、优异开源资源推荐等模块，旨在为面试和学术研究提供高质量的题目和解析。 ### Key features: - 大模型(LLMs)专题：涵盖模型微调、模型结构、模型优化、模型架构、数据准备、模型推理等多方面内容。 - 计算机视觉与感知算法专题：涉及图像处理、深度学习基础、卷积神经网络、目标检测、图像分类等。 - 深度学习基础与框架专题：介绍深度学习框架、模型结构、模型优化、数据准备、模型推理等。 - 自动驾驶、智慧医疗等行业专题：包括自动驾驶系统、医疗影像分析、智能机器人、自然语言处理等领域的最新问题和解决方案。 ### Tech stack: Python, Jupyter, Markdown, GitHub, Git, LaTeX. ### Why it matters: 本项目提供了最新的面试题和解析，涵盖了深度学习、计算机视觉、大模型等多个领域，旨在帮助读者提升面试技巧和学术研究能力。同时，通过分享开源资源和项目代码，为读者提供实际操作和学习的指导。

### Overview: Interactive Tools for Machine Learning, Deep Learning, and Math is a project repository that provides a collection of tools for exploring and understanding various aspects of machine learning, deep learning, and mathematical concepts. ### Key features: - **Transformer Explainer**: An interactive visualization tool designed to help anyone learn how Transformer-based models like GPT work. - **exBERT**: A tool to help humans conduct flexible, interactive investigations and formulate hypotheses for the model-internal reasoning process, supporting analysis for a wide variety of Hugging Face Transformer models. - **BertViz**: A tool for visualizing attention in the Transformer model, extending the Tensor2Tensor visualization tool by Llion Jones and the transformers library from HuggingFace. - **CNN Explainer**: An interactive visualization system designed to help non-experts learn about Convolutional Neural Networks (CNNs). - **Interactive Deep Learning Playground**: A tool for exploring neural networks, including initializing neural networks and embedding projectors. - **OpenAI Microscope**: A tool for visualizing activation atlases in neural networks. - **A visual introduction to Machine Learning**: An interactive introduction to machine learning concepts. - **Interactive Deep Learning Playground**: An interactive playground for exploring neural networks. - **Initializing neural networks**: A tool for initializing neural networks. - **Embedding Projector**: A tool for visualizing embeddings. - **OpenAI Microscope**: A tool for visualizing activation atlases in neural networks. - **The Language Interpretability Tool**: A tool for visualizing the meaning of contextual representations and attention in the model. - **

Overview: TensorFlow Internals is an open-source ebook that delves into the kernel and implementation mechanisms of TensorFlow, covering the programming model, computation graph, and distributed training methods for machine learning. Key features: - Comprehensive coverage of TensorFlow's kernel and implementation details. - Detailed explanation of the programming model, computation graph, and distributed training techniques. - Open-source nature, allowing for free access and sharing of knowledge. Tech stack: LaTeX Why it matters: TensorFlow Internals serves as an invaluable resource for developers seeking to understand and enhance their knowledge of TensorFlow's inner workings, providing a practical guide for both beginners and experienced practitioners in the field of machine learning and deep learning.

Overview: OpenAlgo is a free, open-source, self-hosted trading platform that provides a common interface for designing, hosting, and executing strategies through 36 broker plugins, including 35 securities integrations and Delta Exchange for crypto derivatives. Key features: - Unified Broker API - Python Strategy Host - Flow: No-Code Strategy Builder - AI Agent - Options Trading Suite Tech stack: Python Flask + React 19 Why it matters: OpenAlgo offers a comprehensive trading platform that combines four trading surfaces in one self-hosted instance, providing a unified interface for traders of all backgrounds to execute strategies without being tied to a specific broker.

Overview: This repository contains a collection of papers and code related to the 2025 International Conference on Computer Vision (ICCV). The repository features accepted papers with code, allowing researchers and developers to access and utilize the latest advancements in computer vision. Key features: - Papers and code for ICCV 2025 - Papers and code for CVPR 2025 - Papers and code for ECCV 2024 - Various papers and code across different conferences and workshops Tech stack: Python, TensorFlow, PyTorch, OpenCV, and other relevant tools and libraries. Why it matters: This repository serves as a valuable resource for researchers and developers looking to stay updated with the latest advancements in computer vision. It provides access to papers, code, and datasets that can be utilized for research, development, and educational purposes.

### Overview: MuZero is a state-of-the-art reinforcement learning algorithm designed for board games and Atari games. It learns a model of the environment without any prior knowledge of the environment dynamics, focusing on value prediction networks. MuZero is highly adaptable, supporting multiple platforms, GPUs, and multiple players. ### Key Features: - Residual Network and Fully Connected Network in PyTorch - Multi-threaded/Asynchronous with Ray - Multi-GPU support for training and self-play - TensorBoard real-time monitoring - Model weights automatically saved at checkpoints - Single and two-player mode - Commented and documented - Easily adaptable for new games - Examples of board games, Gym, and Atari games - Pretrained weights available - Experimental Windows support (workaround: use the notebook in Google Colab) ### Why It Matters: MuZero is a groundbreaking algorithm that revolutionizes the field of reinforcement learning by learning a model of the environment without relying on the underlying dynamics. This approach opens up new possibilities for solving complex problems in games and other domains. ### Tech Stack: - Python - PyTorch - Ray - TensorBoard - Pseudocode from Schrittwieser et al., Nov 2019 - Pseudocode from Schrittwieser et al., Nov 2019 ### Further Improvements: - Hyperparameter search - Continuous action space - Tool to understand the learned model ### Demo: All performances are tracked and displayed in real time in TensorBoard. The implementation is designed for educational

### Overview: AI Fairness 360 (AIF360) is a comprehensive toolkit designed to detect and mitigate bias in machine learning models throughout the AI application lifecycle. It includes a set of fairness metrics, explanations, and algorithms to translate algorithmic research into practical solutions for domains such as finance, human capital management, healthcare, and education. ### Key Features: - Comprehensive set of fairness metrics for datasets and models. - Detailed explanations of fairness metrics. - Algorithms to mitigate bias in datasets and models. ### Tech Stack: Python ### Why It Matters: AI Fairness 360 provides a practical solution for detecting and mitigating bias in machine learning models, making it essential for ensuring fair and unbiased AI applications across various industries.

### Overview: QuantHarness is a sophisticated multi-agent trading analysis system that combines technical indicators, pattern recognition, and trend analysis using LangChain and LangGraph. The system provides both a web interface and programmatic access for comprehensive market analysis. ### Key Features: - **Indicator Agent**: Computes five technical indicators (RSI, MACD, Stochastic Oscillator) on each incoming K-line, converting raw OHLC data into precise, signal-ready metrics. - **Pattern Agent**: Identifies familiar patterns in the price chart based on recent high and low points and overall direction, providing a short, plain-language description. - **Trend Agent**: Quantifies market direction, channel slope, and consolidation zones using annotated K-line charts, delivering a concise summary. - **Decision Agent**: Synthesizes outputs from the Indicator, Pattern, Trend, and Risk agents to formulate actionable trade directives, specifying LONG or SHORT positions, entry and exit points, stop-loss thresholds, and rationale. ### Tech Stack: LangChain and LangGraph ### Why it Matters: QuantHarness offers a comprehensive and user-friendly solution for high-frequency trading, providing traders with the tools to analyze market trends, identify patterns, and make informed decisions based on technical indicators and pattern recognition. This system is designed to be both accessible and powerful, making it an essential tool for any trader looking to improve their market analysis capabilities.

### Overview: 计算机视觉实战演练：算法与应用 是一个结合了代码、图示和HTML的在线学习平台，旨在帮助读者快速掌握计算机视觉领域的算法和应用。通过互动式学习，读者可以在平台上运行代码并实时查看结果，从而更好地理解和应用理论知识。 ### Key features: - 代码运行在线，实时反馈 - 图解和代码结合，直观理解算法 - 交互式学习，提高理解和应用能力 - 丰富的算法和应用示例 - 持续更新的在线电子书和源码 ### Tech stack: Jupyter Notebook ### Why it matters: 计算机视觉是人工智能领域的重要分支，掌握其算法和应用对于开发人员和研究人员来说至关重要。通过这个平台，读者可以轻松地学习和实践计算机视觉的算法和应用，从而提高自己的技能和理解。

Overview: aeneas is a Python/C library and set of tools designed to automatically synchronize audio and text, commonly referred to as forced alignment. Key features: - Synchronization map generation between text fragments and audio narration - Output formats for various applications: research (AUD, EAF, TextGrid), digital publishing (SMIL for EPUB 3), closed captioning (SRT, SBV/SUB, TTML, VTT), Web (JSON), further processing (CSV, SSV, TSV, TXT, XML) Tech stack: Python/C, FFmpeg, espeak, BeautifulSoup4, lxml, numpy, Python headers Why it matters: aeneas simplifies the process of synchronizing audio and text, making it easier for researchers, digital publishers, and content creators to manage and present their work effectively.

### Overview: This repository contains a comprehensive collection of research papers on Large Language Model (LLM) agents. We organize papers across key categories including agent construction, collaboration mechanisms, evolution, tools, security, benchmarks, and applications. Our taxonomy provides a structured framework for understanding the rapidly evolving field of LLM agents, from architectural foundations to practical implementations. ### Key Features: - 🏗️ Agent Construction: Methodologies and architectures for building LLM agents - 👥 Agent Collaboration: Frameworks for multi-agent interaction and cooperation - 🌱 Agent Evolution: Self-improvement and learning capabilities of agents - 🔧 Tools: Integration of external tools and APIs with LLM agents - 🛡️ Security: Security concerns and protections for LLM agent systems - 📊 Benchmarks: Evaluation frameworks and datasets for testing agent capabilities - 💡 Applications: Real-world implementations and use cases ### Tech Stack: Unknown ### Why it Matters: This repository serves as a valuable resource for researchers, developers, and practitioners interested in the field of Large Language Model agents. It provides a structured overview of the latest research in the field, including agent construction, collaboration, evolution, security, benchmarks, and applications.

Overview: Lingvo is a framework designed for building neural networks in TensorFlow, particularly sequence models. Key features: - Automatic Speech Recognition - Car - Image - Language Modelling - Machine Translation - 3D Object Detection Tech stack: Python Why it matters: Lingvo simplifies the process of creating neural network models in TensorFlow, making it easier to develop and deploy sequence models efficiently.

### Overview: POT is a Python library that provides various solvers for optimal transport problems, including exact linear OT, entropic and quadratic regularized OT, Gromov-Wasserstein (GW) distances, Fused GW distances, and more. It supports solvers for different divergences and includes OT barycenters for fixed and free support, fast OT solvers in 1D, on the circle, and between Gaussian Mixture Models (GMMs), and many ML related solvers. ### Key features: - A large set of differentiable solvers for optimal transport problems. - OT barycenters for Wasserstein and GW distances for fixed and free support. - Fast OT solvers in 1D, on the circle, and between GMMs. - Many ML related solvers, including domain adaptation, optimal transport mapping estimation, subspace learning, and GNN layers. - Implemented generic OT solvers: OT Network Simplex solver, Conditional gradient, Generalized conditional gradient, Entropic regularization OT, Sinkhorn Knopp Algorithm, stabilized version, greedy Sinkhorn, and Screening Sinkhorn. - Smooth optimal transport solvers for KL and squared L2 regularizations. - Debiased Sinkhorn barycenters. - Smooth optimal transport solvers for KL and squared L2 regularizations. - Non regularized Wasserstein barycenters with LP solver. - Gromov-Wasserstein distances and GW barycenters, exact and regularized. - Fused-Gromov-Wasserstein distances solver and FGW barycenters, exact and regular

### Overview: The Tons of Skills marketplace is a model-agnostic platform that offers a wide range of agent-skills, verified adapters, and a ccpi package manager. It is designed to be accessible and user-friendly, making it a valuable resource for developers looking to enhance their skills. ### Key Features: - **Model-agnostic Skills Catalog**: A comprehensive catalog of agent-skills that can be used across different models. - **Harness-Free Canonical Layer**: A harness-free canonical layer that is currently verified by Claude Code. - **Verified Adapters**: Adapters that have been verified for compatibility with the platform. - **ccpi Package Manager**: A package manager that allows for easy installation and management of skills and plugins. ### Tech Stack: Node.js `>=22.12.0` and pnpm `>=9.15.9` are required for development and verification. ### Why It Matters: The marketplace is designed to be accessible and user-friendly, making it a valuable resource for developers looking to enhance their skills. It offers a wide range of skills and a verified package manager, ensuring that users can easily find and use the skills they need.

Overview: This repository provides a Python implementation of Whisper, a state-of-the-art automatic speech recognition model, with the ability to predict word-level timestamps and confidence scores. This extension aims to improve the accuracy and efficiency of speech transcription by leveraging Dynamic Time Warping and cross-attention weights. Key features: - Predicts word-level timestamps - Provides confidence scores for each word and segment - Optimizes memory usage by predicting timestamps on-the-fly - Supports Whisper models trained on multilingual data Tech stack: Python Why it matters: This extension enhances the accuracy and efficiency of speech transcription, making it suitable for various applications requiring precise timing and confidence analysis in multilingual environments. Repository: linto-ai/whisper-timestamped GitHub stars: 2855