Research & Education

Practical Machine Learning with Python offers a structured and comprehensive three-tiered approach to learning machine learning and deep learning. This resource, based on a book, is packed with over 500 pages of useful information, helping readers master essential skills to recognize and solve complex problems with a data-driven mindset. It uses real-world case studies and leverages the popular Python Machine Learning ecosystem, including frameworks like scikit-learn, pandas, statsmodels, spaCy, nltk, gensim, tensorflow, and keras. The content covers machine learning concepts, the Python ecosystem, standard pipelines, and real-world case studies across diverse domains like retail, finance, and computer vision, making it ideal for practitioners.

PointLLM is a multi-modal large language model designed to understand colored point clouds of objects. It excels at perceiving object types, geometric structures, and appearance, effectively bypassing common issues like ambiguous depth, occlusion, or viewpoint dependency. The tool leverages a novel dataset comprising 660K simple and 70K complex point-text instruction pairs, enabling a robust two-stage training strategy. PointLLM also establishes two benchmarks, Generative 3D Object Classification and 3D Object Captioning, for rigorous evaluation. It offers capabilities for inferencing, chatting with 3D models, and evaluation using traditional metrics or GPT-4, making it a powerful resource for advanced 3D data analysis and robotics applications.

PointMamba is an open-source state space model (SSM) specifically designed for point cloud analysis, leveraging the success of Mamba from natural language processing. Unlike traditional Transformers, PointMamba employs a linear complexity algorithm, enabling global modeling while substantially reducing computational costs and GPU memory usage. This tool utilizes space-filling curves for efficient point tokenization and features a simple, non-hierarchical Mamba encoder as its backbone. Comprehensive evaluations demonstrate its superior performance across various datasets, making it a valuable resource for researchers and developers in 3D vision. PointMamba underscores the potential of SSMs in 3D vision-related tasks and provides a robust baseline for future research.

SparkNet is an open-source framework designed for building and training distributed neural networks using Apache Spark. It allows users to leverage the power of Spark for scalable AI model development, particularly beneficial for handling large datasets. The framework provides functionalities for quick cluster setup on EC2, training models like Cifar and ImageNet, and installing SparkNet on existing Spark clusters. It supports GPU acceleration with CUDA and offers pre-built JavaCPP binaries for various platforms, making it a robust solution for data scientists and machine learning engineers working with distributed computing environments.

stanford_dl_ex is a repository offering programming exercises for the Stanford Unsupervised Feature Learning and Deep Learning Tutorial. It provides starter code designed to help users engage with and practice the concepts taught in the official Stanford tutorial, available at ufldl.stanford.edu/tutorial. This resource is particularly useful for individuals looking to deepen their understanding and practical application of deep learning principles through hands-on coding. The repository includes various modules covering different aspects of deep learning, such as convolutional neural networks (CNN), principal component analysis (PCA), and sparse autoencoders (STL). It serves as a valuable, free educational tool for students and researchers alike.

TrajectoryCrafter is an advanced Content & Design tool designed to redirect camera trajectories in monocular videos using sophisticated diffusion models. This tool, presented at ICCV 2025, allows users to generate high-fidelity novel views from standard monocular video footage, offering precise control over camera pose. It is particularly useful for researchers and developers working with video manipulation and synthesis. The system requires a GPU with at least 28GB VRAM for optimal performance and can be set up using standard Python environments. While powerful, its capabilities are rooted in a pretrained video diffusion model, meaning it performs best with well-defined objects and clear motion, and may face limitations with highly complex scenarios beyond its base model's generation capacity. It provides both command-line inference and a local Gradio demo for ease of use.

TimeSeries_Seq2Seq is a GitHub repository offering a valuable collection of notebooks and code designed to facilitate the understanding and implementation of sequence-to-sequence (seq2seq) neural networks specifically for time series forecasting. The networks within this repository are built using popular deep learning frameworks, Keras and TensorFlow. It serves as a practical resource for data scientists and researchers looking to apply advanced neural network architectures to predict future values based on historical time-dependent data. The repository includes instructions for setting up the environment and working with the provided notebooks, making it accessible for those interested in hands-on learning and application of seq2seq models in time series analysis.

XuanCe (玄策) is an open-source, comprehensive, and unified deep reinforcement learning (DRL) library designed to provide high-quality and easy-to-understand implementations of DRL algorithms. It aims to address the sensitivity of DRL algorithms to hyper-parameter tuning and unstable training processes by offering a robust and flexible framework. XuanCe is highly modularized, easy to install and use, and supports flexible model combinations. It includes abundant algorithms for various tasks, supporting both DRL and Multi-Agent Reinforcement Learning (MARL) tasks. The library boasts high compatibility across different deep learning backends (PyTorch, TensorFlow2, MindSpore), operating systems (Linux, Windows, MacOS), and hardware (CPU, GPU). Key features include fast running speed with parallel environments, distributed training with multi-GPUs, automatic hyperparameter tuning, and good visualization effects with TensorBoard or Weights & Biases.

Visualizer is a specialized tool designed to simplify the process of visualizing attention maps within deep learning models, particularly those based on Transformer architectures. It addresses common challenges faced by developers, such as the difficulty of extracting deeply nested attention maps without modifying model code or encountering out-of-memory errors. The tool provides a non-intrusive method using Python decorators and PyTorch hooks, allowing users to precisely retrieve intermediate variables like attention maps. This ensures consistency between training and testing phases, as no code changes are required for visualization. It's particularly useful for analyzing complex models like Vision Transformers, enabling the extraction of all attention maps across multiple layers with minimal effort.

Synnada is an AI infrastructure company dedicated to rethinking how intelligent systems are built. It provides the foundational technology for data science and content understanding, enabling the creation of reliable, scalable, and agent-native systems. Built by Apache DataFusion contributors, Synnada's offerings include Mithril for efficient model compilation, Tenet for multi-cloud AI workload deployment, and Agentia, a runtime for persistent agent systems with first-class code execution. This infrastructure supports the agentic economy, allowing intelligent agents to operate continuously across clouds, datasets, and decision loops, ensuring correctness, efficiency, and long-term operability for production-grade AI.

Arbiter is a Rust-based, event-driven multi-agent framework designed for orchestrating strongly-typed, high-performance simulations and networked systems. It provides foundational types and traits for building actor-based systems with pluggable networking and lifecycle management. Tailored for discrete-event simulation, automated trading, and complex distributed systems, Arbiter's core concepts include Actors for execution units, LifeCycle for actor behavior, Handlers for message processing, Networks for system connections, and Runtimes for managing execution context. The framework is open-source and actively developed by Harnesslabs, offering extensive documentation and examples for in-depth understanding.

Arraymancer is a powerful n-dimensional tensor (ndarray) library implemented in Nim, designed for high performance and ease of use. It provides a robust foundation for scientific computing, machine learning algorithms, and deep learning applications. The library supports various backends including CPU, Cuda, and OpenCL, and can leverage OpenMP for multithreaded compilation. Key features include basic math operations generalized to tensors, matrix algebra primitives, efficient slicing, broadcasting support, and a variety of reshaping operations. Arraymancer can handle tensors up to 6 dimensions and supports reading/writing .csv, Numpy (.npy), and HDF5 files. While its deep learning components are still evolving, it offers functionalities for neural networks, including fully-connected layers and convolutional networks, making it a versatile tool for developers and data scientists working with Nim.

Chinese-Text-Classification-Pytorch is an open-source toolkit designed for Chinese text classification tasks, built on the PyTorch framework. It offers out-of-the-box implementations of several popular text classification models, including TextCNN, TextRNN, FastText, TextRCNN, BiLSTM_Attention, DPCNN, and Transformer. The toolkit is user-friendly and ready for immediate deployment, supporting both character-level input and the integration of pre-trained word vectors, specifically using Sougou News Word+Character 300d. It also includes a pre-processed Chinese dataset (THUCNews) for training and evaluation, making it a comprehensive resource for researchers and developers working on Chinese NLP.

ecg is an open-source AI tool designed for advanced arrhythmia detection and classification in ambulatory electrocardiograms. Leveraging a deep neural network, it aims to achieve cardiologist-level accuracy in analyzing ECG data. The tool is hosted on GitHub, providing a platform for researchers and developers to access, train, and test models. It includes instructions for setting up a Python environment, installing dependencies with or without GPU support, and training/testing models using configuration files. This makes it a valuable resource for medical diagnosis, research, and the development of AI-powered healthcare solutions.

External-Attention-pytorch is a comprehensive GitHub repository offering PyTorch implementations of numerous attention mechanisms, Multi-Layer Perceptrons (MLPs), re-parameterization techniques, and convolution operations. This resource is designed for developers and researchers looking to deepen their understanding of these fundamental components in deep learning models. It includes detailed examples and usage instructions for over 30 different attention mechanisms, such as External Attention, Self Attention, MobileViT Attention, and many more. Additionally, it covers various backbone architectures like ResNet and MobileViT, several MLP types, and re-parameterization methods like RepVGG. The repository serves as a valuable educational and practical toolkit for implementing advanced neural network architectures.

hum.ai is dedicated to building advanced multimodal foundation models designed for practical, real-world applications. Their core focus is on leveraging satellite remote sensing and ground truth data to train these models, aiming to develop Artificial General Intelligence (AGI) for a deeper understanding of the natural world. The technology developed by hum.ai is currently being utilized in critical sectors such as nature conservation, carbon dioxide removal initiatives, and by various government agencies. This positions hum.ai at the forefront of applying AI to solve complex environmental and scientific challenges, providing robust solutions for data analysis and predictive modeling in these domains.

PYNQ-Classification is an open-source framework designed for the rapid deployment of embedded Convolutional Neural Network (CNN) applications on PYNQ platforms. It leverages Python on Zynq FPGA to accelerate CNN processing. The repository provides instructions for setting up Caffe and Theano dependencies, and includes demos for LeNet and CIFAR-10 models. Users can download a pre-configured SD card image or manually set up dependencies. The framework also guides on regenerating Vivado and Vivado HLS projects for implementing additional CNN models, making it a valuable resource for researchers and developers working with FPGA-based CNN acceleration.

Moodify is an innovative AI tool designed to revolutionize the fragrance industry by digitizing olfaction. It provides an autonomous digital perfumer, powered by Artificial Olfactive Intelligence (AOI), to streamline the formulation process. The platform helps fragrance houses reduce material costs, ensure regulatory compliance, and scale their operations through AI-powered reformulation. Key capabilities include brief mastery to align on olfactive targets, computational intelligence to encode perception into smell vectors for precise tuning, and deployment features that generate ranked formulas considering real-world constraints like cost, IFRA, and sustainability. Moodify also offers solutions for malodor control and fragrance portfolio optimization.

The NSF AI Institute for Artificial Intelligence and Fundamental Interactions (IAIFI) is a leading institute dedicated to pioneering interdisciplinary research at the intersection of AI and physics. It aims to advance fundamental physics knowledge, from the smallest building blocks of nature to the largest structures in the Universe, while simultaneously galvanizing AI research innovation. IAIFI focuses on developing AI approaches that incorporate first principles from physics and tackles challenging problems such as precision calculations and gravitational wave detection. Beyond research, IAIFI is committed to empowering the next generation of AI+Physics talent through various educational programs, including fellowships, summer schools, and workshops, and building a dynamic AI+Physics community through events and collaborations.

PreciseRoIPooling is an open-source implementation of the Precise RoI Pooling (PrRoI Pooling) method, as proposed in the ECCV 2018 paper "Acquisition of Localization Confidence for Accurate Object Detection." This tool is designed to improve object detection accuracy by providing an integration-based average pooling method for RoI Pooling, which avoids quantization and offers a continuous gradient on bounding box coordinates. Unlike traditional RoI Pooling or RoI Align, PrRoI Pooling allows for the optimization of RoI coordinates through continuous gradients. The repository provides implementations for PyTorch (versions 1.0+ and 0.4) and TensorFlow (2.2), primarily supporting CUDA. It is a valuable resource for researchers and developers working on advanced object detection models.

T2F is an open-source deep learning project designed for generating realistic human faces from textual descriptions. It leverages a combination of StackGAN and ProGAN architectures to achieve high-quality image synthesis. The project processes textual descriptions through an LSTM network to create a summary vector, which then informs the GAN's generation process. While the original project is not actively maintained, a T2F 2.0 version is planned to utilize MSG-GAN for improved image generation. The tool is implemented using PyTorch and requires specific dependencies for setup and training, making it suitable for researchers and developers interested in generative AI.

T81-558 is a comprehensive GitHub repository containing teaching materials for the T81-558: Keras - Applications of Deep Neural Networks course offered at Washington University in St. Louis. This resource focuses on the Keras/TensorFlow version of the curriculum, covering a wide array of deep learning topics. Students and enthusiasts can explore modules on Python preliminaries, Pandas for machine learning, TensorFlow and Keras fundamentals, training for tabular data, regularization, CNNs for vision, Generative Adversarial Networks (GANs), Kaggle competitions, transfer learning, time series analysis, reinforcement learning, and deploying models with Flask. The repository includes Jupyter notebooks for practical application and a complete textbook available on GitHub, making it an invaluable resource for learning and applying deep neural network concepts.

Singa is an open-source distributed deep learning platform developed by Apache. It provides a flexible architecture for training deep learning models across various devices and distributed environments. The platform supports a wide range of deep learning models and offers tools for efficient computation and data management. Singa is particularly well-suited for researchers and developers who require a robust and scalable solution for their large-scale AI projects, enabling them to build, train, and deploy complex neural networks. Its open-source nature fosters community contributions and allows for extensive customization to meet specific project requirements.

prml is an open-source GitHub repository dedicated to Christopher Bishop's seminal work, "Pattern Recognition and Machine Learning." It provides a comprehensive collection of Jupyter notebooks and Python code that implement many of the algorithms and replicate numerous graphs presented in the book. This resource is invaluable for students, professors, and researchers looking to understand and apply machine learning concepts through practical examples. The repository covers a wide range of topics, from basic probability distributions and linear models to more advanced subjects like neural networks, Gaussian processes, and hidden Markov models, making it a robust companion for academic study and practical implementation in the field of pattern recognition and machine learning.