Research & Education

Deep-Learning-Roadmap is an open-source project designed to serve as a comprehensive collection of organized resources for deep learning researchers and developers. The project aims to provide a shortcut for finding useful information by categorizing resources into a large number of sections, making it easy for users to locate specific topics. It covers a wide array of subjects, including various deep learning models like Convolutional Networks, Recurrent Networks, Autoencoders, and Generative Models. Additionally, it delves into core optimization techniques, representation learning, understanding and transfer learning, and reinforcement learning. The roadmap also highlights diverse applications such as image recognition, object recognition, natural language processing, and speech technology, alongside an extensive list of relevant datasets.

deepinv is an open-source PyTorch-based library designed for solving imaging inverse problems with deep learning. It is part of the official PyTorch Ecosystem and aims to accelerate deep learning research across various imaging domains. The library enhances research reproducibility through a common modular framework for problems and algorithms, making it easier for new practitioners to get started. Key features include a large framework of predefined imaging operators, state-of-the-art deep neural networks with pretrained models, comprehensive frameworks for plug-and-play restoration, optimization, and unfolded architectures, as well as training losses for inverse problems. It also supports sampling algorithms, diffusion models for uncertainty quantification, and a framework for building datasets.

Naru Healthcare specializes in innovative AI-powered solutions for the healthcare sector, particularly in oncology. The company's core offering, aiatech, leverages advanced AI and proprietary clinical outcome generation algorithms to transform Real-World Data (RWD) from hospitals and patients into Real-World Evidence (RWE). This technology aims to maximize treatment effectiveness, learn from each patient, and improve clinical practices and research. Naru's Step Oncology solution provides a comprehensive system for patient monitoring, predictive modeling, and analytical visualization, supporting clinical decision-making and optimizing resource allocation in oncology.

unlock-deepseek is an open-source learning project dedicated to systematically interpreting and reproducing the DeepSeek series of AI models. It covers DeepSeek's advancements in large language models, mathematical reasoning, code generation, multimodal AI, inference models (like DeepSeek-R1), MoE architecture, and training infrastructure. The project aims to break down DeepSeek's cutting-edge technologies into understandable and reproducible learning content for a wide range of AI researchers and learners. Key features include in-depth paper analysis, hands-on tutorials for reproduction, technical breakdowns of core components, and comparative analysis with similar works.

RecFM offers a comprehensive suite of tools and frameworks specifically designed for building foundation models in recommendation systems. Developed by the USTCLLM group at USTC, it provides modular libraries and technologies that streamline the development process. The platform aims to facilitate the creation of robust and efficient recommendation systems, enabling researchers and developers to leverage advanced AI models for personalized content delivery and user experience optimization. Its focus on foundation models suggests capabilities for handling large datasets and complex recommendation logic, making it suitable for advanced AI research and application development.

seqeval is a Python framework designed for the evaluation of sequence labeling tasks, including named-entity recognition (NER), part-of-speech (POS) tagging, and semantic role labeling. It provides robust evaluation capabilities, tested against the industry-standard Perl script `conlleval` for compatibility with CoNLL-2000 shared task data. The framework supports multiple common annotation schemes such as IOB1, IOB2, IOE1, IOE2, IOBES, and BILOU, with strict mode evaluation available for IOBES and BILOU. Users can compute standard metrics like accuracy, precision, recall, and F1 score, and generate comprehensive classification reports to assess model performance effectively. Its flexibility makes it a valuable tool for researchers and developers working on natural language processing tasks.

ReActGPT is an AI research tool designed to enhance the accuracy and reliability of AI models by implementing the ReAct (Reasoning and Acting) paradigm. This framework helps in reducing errors in AI outputs, making it a valuable asset for those working on advanced AI systems. The tool is specifically tailored for researchers and developers who are focused on pushing the boundaries of AI technology and building more robust and dependable AI applications. By integrating the ReAct approach, ReActGPT provides a structured method for AI agents to reason about their actions and observations, leading to more intelligent and less error-prone behavior. This makes it an essential component for anyone looking to develop cutting-edge AI solutions with improved performance and trustworthiness.

DeepMesh is an open-source project providing the official code for an auto-regressive artist-mesh creation system powered by reinforcement learning. Developed by researchers from Tsinghua University, ShengShu, and S-Lab at Nanyang Technological University, DeepMesh can generate high-quality 3D meshes based on given point cloud data using an auto-regressive transformer. The project has released a pretrained weight of DeepMesh (0.5 B) and optimized its inference code for a 50% reduction in generation time. It is designed for researchers and developers in the field of generative AI and 3D modeling, offering a framework for advanced mesh generation.

Awesome-LLM-Robotics is a comprehensive and curated list of academic papers focusing on the application of large language models (LLMs) and multi-modal models in the fields of Robotics and Reinforcement Learning (RL). Hosted on GitHub, this open-source repository serves as a valuable resource for researchers, academics, and students looking to stay updated on the latest advancements. The list is organized into categories such as Surveys, Reasoning, Planning, Manipulation, Instructions and Navigation, Simulation Frameworks, and Safety, Risks, Red Teaming, and Adversarial Testing. Each entry typically includes the paper title, publication details, and links to the paper, code, and related websites, making it easy to access and explore the research. Users are encouraged to contribute by submitting pull requests to keep the list current and comprehensive.

Awesome-video-text-retrieval is a curated list of deep learning resources specifically focused on video-text retrieval. This open-source repository serves as a valuable hub for researchers and engineers working in the fields of AI, computer vision, and natural language processing. It meticulously organizes implementations across various frameworks like PyTorch and TensorFlow, alongside an extensive collection of research papers spanning from 2018 to 2023. The resource is designed to help users explore the latest advancements, find relevant codebases, and understand different approaches to aligning video content with textual descriptions. It's an essential reference for anyone looking to delve into or stay updated on the state-of-the-art in video-text retrieval.

Scholia Awakens is an AI-powered research assistant designed to streamline academic and professional research workflows. The tool supports dynamic source integration, allowing users to incorporate various data sources seamlessly, and offers real-time adaptation to evolving research needs. A core feature is its insights engine, which automatically generates mindmaps from research materials and detects contradictions within the data, enhancing critical analysis. Additionally, Scholia Awakens provides AI templates for efficient report and summary creation, smart collections for organizing research, and a full editor with live AI assistance to support the entire research process.

The APRIL AI Hub is a UK-based initiative dedicated to transforming the electronics industry through advanced AI research. It aims to bridge the gap between the artificial intelligence and electronics communities, fostering collaboration to develop and commercialize AI-based tools that enhance productivity across the entire electronics supply chain. The Hub focuses on world-leading AI research, building an expansive network of academic institutions and industrial members, and enabling technology adoption through new AI tools or startups. It also emphasizes talent development through workshops, training, internships, and outreach programs, ensuring a strong pipeline of expertise in both AI and electronics. The Hub organizes events like annual summits and webinars, and even developed a card game to introduce AI concepts to younger audiences.

The Institute of Applied Artificial Intelligence and Robotics (IAAIR) is dedicated to crafting a fair and accountable AI future by enhancing human capabilities and contributing positively to society. They focus on developing systems that make fair, transparent, and unbiased decisions, building public trust in AI technologies. IAAIR combines advanced research, ethical frameworks, and real-world implementation to lead the development of cutting-edge and trustworthy AI. Their services include AI research and development, AI audit and security, innovation support, corporate AI training, AI productionization and optimization, AI awareness programs for kids, data labeling services, and on-demand AI experts.

NeuralNLP-NeuralClassifier is an open-source toolkit designed for implementing neural models for hierarchical multi-label text classification. This tool addresses the complexities of real-world classification tasks by offering a variety of text encoders, including FastText, TextCNN, TextRNN, RCNN, VDCNN, DPCNN, DRNN, AttentiveConvNet, and Transformer encoder. Beyond hierarchical multi-label classification, it also supports binary-class and multi-class text classification. Built on PyTorch, NeuralNLP-NeuralClassifier demonstrates performance comparable to reported results in academic literature, making it a robust solution for researchers and engineers in natural language processing.

Machine Learning for Science (ML4SCI) is an open-source organization dedicated to integrating modern machine learning techniques with challenging problems across Science, Technology, Engineering, and Mathematics (STEM) fields. The organization fosters collaboration among researchers, students, and developers to advance the application of AI in scientific contexts. ML4SCI actively participates in programs like Google Summer of Code (GSoC), providing opportunities for students to contribute to open-source projects. It also serves as an umbrella organization, welcoming other projects and organizations focused on machine learning for science, and encourages the publication of scientific articles in peer-reviewed journals by its contributors. The initiative aims to push the boundaries of scientific discovery through AI.

SynthText is an open-source tool designed for generating synthetic text images, primarily for use in computer vision research. It enables the creation of extensive datasets of scene-text images, which are crucial for training and evaluating models focused on text localization in natural images. The tool provides scripts for generating samples, including options for visualizing the output, and supports adding new background images with segmentation and depth-maps. It also offers flexibility for generating text in various non-Latin scripts, with several community adaptations available for languages like Chinese, Arabic, Japanese, Korean, Vietnamese, and German. SynthText is a valuable resource for researchers and developers working on text detection and recognition tasks.

text-embeddings-inference (TEI) is a toolkit designed for deploying and serving open-source text embeddings and sequence classification models with blazing fast performance. It enables high-performance extraction for a wide range of popular models, including FlagEmbedding, Ember, GTE, and E5. TEI incorporates several advanced features such as token-based dynamic batching, optimized transformer code utilizing Flash Attention, Candle, and cuBLASLt, and support for Safetensors and ONNX weight loading. It also offers production-ready capabilities like distributed tracing with Open Telemetry and Prometheus metrics. The solution supports various model types, including Nomic, BERT, CamemBERT, XLM-RoBERTa, JinaBERT, Mistral, Alibaba GTE, Qwen2, MPNet, ModernBERT, Qwen3, and Gemma3, making it a versatile choice for developers and researchers working with text embeddings.

Wiseone acts as a gateway for users to safely access the W88 online betting platform. In an environment rife with fraudulent links, Wiseone provides updated, official links to ensure a secure and uninterrupted betting experience. It highlights W88's robust legal framework, including licensing from First Cagayan Leisure and Resort Corporation, and its adherence to international safety and transparency standards. The platform also emphasizes advanced security measures like SSL 128-bit encryption, diverse payment methods, and 24/7 customer support. Wiseone positions W88 as a top-tier online casino by showcasing its stable access links, a wide array of games including sports betting and live casino, and dedicated customer service.

Verbos Podcast is Denmark's only podcast specifically tailored for AI and software engineers. It offers in-depth technical discussions on the rapidly evolving field of artificial intelligence and its practical applications within software development. The podcast delves into the latest advancements in AI technology, exploring how these innovations can be effectively integrated into various software solutions. Hosted by experienced AI engineers, Verbos Podcast provides valuable insights and perspectives for professionals looking to stay updated on the intersection of AI and software engineering. It serves as a dedicated platform for the Danish tech community to engage with cutting-edge topics and foster knowledge sharing.

ufldl_tutorial is the Stanford Unsupervised Feature Learning and Deep Learning Tutorial, offering comprehensive educational resources for individuals interested in these advanced AI topics. The tutorial provides detailed explanations and practical code examples, making complex concepts accessible. It specifically covers areas such as sparse autoencoders and includes hands-on implementations for visualizing features on MNIST data, a common dataset in machine learning. This resource is designed to help users understand the theoretical underpinnings and practical applications of unsupervised feature learning and deep learning, fostering a deeper comprehension of AI methodologies.

wav2letter is an open-source automatic speech recognition (ASR) toolkit developed by Facebook AI Research. It is specifically designed for AI researchers and speech recognition developers, offering a flexible framework for building and experimenting with ASR models. The toolkit has been consolidated into Flashlight in the ASR application, indicating its integration into a broader machine learning library. While the provided website content is a GitHub pricing page, the context from the tool's description suggests its primary function is to provide foundational tools for advanced speech recognition development, rather than being a consumer-facing application. Users can leverage wav2letter for tasks such as training custom speech models and conducting research in the field of automatic speech recognition.

PDF to LaTeX is an online tool leveraging AI to convert PDF documents into LaTeX code. This tool is designed for ease of use, allowing users to simply upload a PDF file and receive the corresponding LaTeX output. It employs a sophisticated multimodal LLM model, which first converts the PDF into images and then processes these images to generate LaTeX code. The model is trained on an extensive dataset of PDFs and their associated LaTeX code, ensuring accuracy in conversion. This functionality is particularly beneficial for academics, researchers, and students who frequently work with scientific and mathematical documents, enabling them to easily edit and format content. The tool also offers the flexibility to purchase pages for conversion, catering to various user needs.

Awesome-LLM-Eval is a comprehensive, curated list designed for the evaluation of Large Language Models (LLMs) and the exploration of Generative AI's capabilities and limitations. This open-source GitHub project compiles a wide array of resources, including evaluation tools, diverse datasets and benchmarks, practical demos, competitive leaderboards, relevant academic papers, and various LLM models. It serves as an official project for the survey "Beyond Benchmark: LLMs Evaluation with an Anthropomorphic and Value-oriented Roadmap," offering continuous updates that may not be reflected in the arXiv paper. The repository is actively maintained, welcoming community contributions through pull requests and issues, ensuring it remains a dynamic and up-to-date resource for researchers and developers in the LLM evaluation space.

Awesome-RL-for-LRMs is an open-source project offering a comprehensive survey of reinforcement learning (RL) techniques specifically applied to large reasoning models (LRMs). This resource is invaluable for researchers and engineers looking to understand and implement RL in the context of large language models (LLMs) and other reasoning models. It compiles relevant papers, resources, and insights, making it easier to navigate this complex and rapidly evolving field. The project aims to provide a foundational understanding and practical guidance for those involved in AI model training and development, particularly in areas requiring advanced reasoning capabilities.