Research & Education

ResShift is an efficient open-source diffusion model designed for image super-resolution, developed by Zongsheng Yue and others. It addresses the common limitation of slow inference speeds in diffusion-based SR methods by introducing a novel residual shifting technique, which drastically reduces the required sampling steps to as few as 15, or even 4 in its journal version, without compromising output quality. This approach constructs a Markov chain that efficiently transfers between high-resolution and low-resolution images. Beyond super-resolution, ResShift also supports applications like image deblurring, natural and face image inpainting, and blind face restoration. The project has been recognized at NeurIPS 2023 (Spotlight) and published in TPAMI@2024, highlighting its advanced capabilities and efficiency in image enhancement.

Roadmap-To-Learn-Agentic-AI is an open-source GitHub repository offering a comprehensive guide to mastering agentic AI systems. It begins with foundational knowledge in Python programming and essential machine learning concepts, including Natural Language Processing (NLP) techniques like TFIDF and Word2vec. The roadmap then progresses to in-depth Deep Learning for NLP, transformer explanations, and extensive Generative AI tutorials with end-to-end projects. A significant portion is dedicated to Agentic AI tutorials, exploring various frameworks such as Langchain, LangGraph, Agno, Phidata, CrewAI, and Autogen. This resource is ideal for individuals looking to build a strong understanding and practical skills in the rapidly evolving field of agentic AI.

RoleLLM-public is a comprehensive framework designed to benchmark, elicit, and enhance the role-playing capabilities of Large Language Models (LLMs). It introduces RoleLLM, a four-stage process encompassing role profile construction, Context-Based Instruction Generation (Context-Instruct) for knowledge extraction, Role Prompting using GPT (RoleGPT) for style imitation, and Role-Conditioned Instruction Tuning (RoCIT) for fine-tuning open-source models. The framework includes RoleBench, a systematic and fine-grained character-level benchmark dataset with over 168,000 samples. RoCIT on RoleBench has led to the development of RoleLLaMA (English) and RoleGLM (Chinese), significantly improving role-playing performance to levels comparable with GPT-4.

spaCy is a powerful, open-source library for advanced Natural Language Processing (NLP) in Python and Cython. Designed for production use, it incorporates the latest research and provides pre-trained pipelines for over 70 languages, enabling tokenization and training. Key features include state-of-the-art speed, neural network models for tasks like tagging, parsing, named entity recognition, and text classification, as well as multi-task learning with transformers like BERT. It boasts a robust training system, easy model packaging, deployment, and workflow management, making it suitable for industrial-strength applications. spaCy is released under the MIT license, offering a comprehensive solution for developers and researchers working with NLP.

Stanford CME 295 Transformers & Large Language Models offers a comprehensive VIP cheatsheet for the Stanford CME 295 course. This resource condenses key concepts related to Transformers and Large Language Models, including self-attention mechanisms, architectural variants, and optimization techniques like sparse attention and flash attention. It also covers LLM-specific topics such as prompting, fine-tuning (SFT, LoRA), preference tuning, and optimization methods like mixture of experts, distillation, and quantization. The cheatsheet extends to practical applications like LLM-as-a-judge, RAG, agents, and reasoning models, making it an invaluable study aid for students and researchers.

Treasure-of-Transformers is an open-source GitHub repository offering a comprehensive collection of Transformer models for Natural Language Processing (NLP). It serves as a valuable resource for anyone interested in NLP, providing links to papers, videos, blogs, official repositories, and Colab notebooks for over 100 different Transformer models, including popular ones like GPT-3, BERT, and T5. The repository is organized as a list of NLP deep learning algorithms with their respective years of introduction, making it easy to navigate and find specific models. It's an excellent resource for students, researchers, and practitioners looking to explore or implement various Transformer architectures.

YuLan-Chat is an open-source large language model developed by researchers at GSAI, Renmin University of China. The model is chat-based, developed through pre-training from scratch and supervised fine-tuning using curriculum learning with high-quality English and Chinese instructions and human preference data. Key technical characteristics include improved language ability due to large-scale pre-training on high-quality English, Chinese, and multilingual data, and enhanced helpfulness, honesty, and harmlessness through curriculum learning for human alignment. It also supports longer Chinese inputs and outputs by expanding the vocabulary with Chinese words and increasing the maximum input length to 4k context. Various versions, including YuLan-Mini and YuLan-Base-12B, have been released, with some based on LLaMA or LLaMA-2 architectures.

ML4DB-paper-list is an open-source repository offering a comprehensive and continuously updated list of research papers related to database systems enhanced by artificial intelligence, including machine learning, deep learning, and reinforcement learning. The collection covers various topics such as system and tutorial papers, training data collection, data access configuration tuning, physical design, learned index structures, workload management, query optimization, and Text-to-SQL. It serves as a valuable resource for researchers and academics interested in the intersection of AI and database systems, providing insights into the latest advancements and foundational theories in the field.

BenchSci is a leading AI solution for preclinical R&D, leveraging a neuro-symbolic AI platform called ASCEND to decode complex disease biology. This platform integrates a rigorously vetted Biological Evidence Knowledge Graph (BEKG) with advanced foundation models, offering AI copilots and co-scientists to accelerate discovery. Unlike platforms limited to public data, BenchSci's BEKG combines tens of millions of scientific publications, including closed-access journals, with clients' proprietary internal data. This comprehensive knowledge base, supported by a proprietary multimodal AI and human-in-the-loop curation by over a hundred scientists, ensures data quality and mitigates AI hallucinations. BenchSci aims to transform drug discovery by providing a unified view of disease mechanisms, from genetic modifiers to outcomes, helping scientists go from hypothesis to successful experiment in days, not years.

ml-surveys is an open-source GitHub repository that provides a curated collection of survey papers. It aims to help researchers and practitioners keep pace with the rapid advancements in various machine learning domains, including deep learning, natural language processing (NLP), computer vision (CV), graph embeddings, reinforcement learning, and recommendations. The repository organizes papers by topic, offering a structured way to explore the state-of-the-art in each area. This resource is particularly valuable for those looking to understand the foundational and recent developments in specific ML subfields without having to sift through countless individual research papers.

Osmosis is an AI search engine designed to offer insights into the future of technology and innovation. It allows users to explore a wide range of topics, including the next frontiers of AI, the future of hybrid vehicles, cybersecurity, generative AI, digital healthcare, and fintech innovation. The platform provides a curated selection of episodes and content to help users stay informed about emerging trends and developments in these fields. To access Osmosis, users are required to enter their email address, suggesting a gated content model for its offerings.

nlp-tutorial is an open-source GitHub repository offering a comprehensive tutorial for Natural Language Processing (NLP) specifically tailored for deep learning researchers and students using PyTorch. The project focuses on providing concise implementations of various NLP models, with most examples written in under 100 lines of code, excluding comments and blank lines. It covers fundamental concepts from basic embedding models like NNLM and Word2Vec, to more advanced topics such as CNNs (TextCNN), RNNs (TextRNN, TextLSTM, Bi-LSTM), Attention Mechanisms (Seq2Seq, Bi-LSTM with Attention), and Transformer-based models like BERT. The tutorial includes Colab notebooks for easy experimentation and supports PyTorch version 1.0 or higher. An archive of older TensorFlow v1 code is also available.

The LF AI & Data Foundation is dedicated to advancing open-source innovation in artificial intelligence, machine learning, deep learning, and data technologies. As a neutral host, it facilitates collaboration among developers, organizations, and users, providing a trusted infrastructure for the AI and data community. The foundation supports projects from inception to fruition, offering resources and a collaborative environment to accelerate development and adoption. It aims to create scalable, interoperable solutions that drive impact across various industries worldwide, with a strong emphasis on community-driven development and open standards.

The University of British Columbia (UBC) is a global center for research and teaching, consistently ranked among the top public universities. Its Computer Science department is dedicated to advancing knowledge through education and research in various fields, including artificial intelligence, machine learning, and human-computer interaction. UBC actively engages in innovative research, such as developing AI responsibly in healthcare to speed up treatment paths and exploring the impact of AI on obesity research. The institution also focuses on community engagement, training the next generation in health care innovation across British Columbia and supporting initiatives like the UBC Innocence Project to reshape legal practice.

Lorka AI is an all-in-one AI platform designed to streamline workflows by integrating multiple leading AI chat models such as GPT, Claude, Gemini, Grok, Qwen, and DeepSeek into a single subscription. Users can switch between different AI engines within the same conversation without losing context, allowing for dynamic brainstorming, refinement, and verification. Beyond chat, Lorka AI offers a suite of dynamic features including AI Web Search for quick information retrieval, an AI Image Editor for visual content creation, and advanced tools like PDF chat, AI Translator, and AI Humanizer. It also supports a voice mode for hands-free interaction. This platform aims to provide significant savings and flexibility compared to subscribing to individual AI services, catering to professionals and students across various fields.

Process Intelligence Research is a research group at Delft University of Technology dedicated to transforming chemical engineering through Artificial Intelligence. Their vision centers on integrating AI and chemical engineering to create next-generation intelligent knowledge and decision-making platforms. The group conducts fundamental and applied research in machine learning, data science, and process systems engineering, aiming to enhance model accuracy, efficiency, and predictive capabilities within chemical engineering. They are open to collaborations with industry and academia to advance their vision, focusing on areas like Process Systems Engineering, machine learning, computer vision, natural language processing, and hybrid modeling.

Neurona Lab specializes in leveraging artificial intelligence for the rapid and accurate diagnosis of neurodegenerative diseases. The platform is developing several key diagnostic algorithms, including Neurona PET, which calculates and compares MRI & PET fusion data. Neurona ARIA is designed to detect brain edemas and microhemorrhages, while Neurona VOX focuses on recognizing early signs of Alzheimer's disease through speech analysis. This innovative approach aims to address the challenges in dementia diagnosis, such as diagnostic difficulty, specialist shortages, and late disease detection, ultimately improving patient outcomes and reducing the economic and social burden of these conditions.

CogVideoX Fun 5b is an AI video generation tool hosted on Hugging Face Spaces by alibaba-pai. This application allows users to generate short videos based on textual descriptions of a scene. Additionally, it offers a unique feature where users can upload an existing video with empty or incomplete areas, and the system will intelligently fill them in according to user input. This makes it a versatile tool for experimenting with video generation models and creative video editing. The tool is built with Gradio, indicating an accessible and user-friendly interface for interaction. It is licensed under an open-source license, promoting accessibility and community engagement.

DarkIR is an AI-powered tool designed for robust low-light image restoration, accessible via Hugging Face Spaces. Users can upload images taken in dark environments, and the application will automatically enhance them to significantly improve clarity and reduce visual noise. This makes it an ideal solution for anyone needing to salvage or improve photographs captured under challenging lighting conditions. The tool is free to use, making advanced image enhancement accessible without cost. Its core functionality focuses on a single, powerful capability: transforming underexposed or noisy images into clearer, more usable visuals.

GPT-OSS-120B on AMD MI300X is an AI chatbot hosted on Hugging Face Spaces, designed to run on AMD MI300X GPUs. This tool offers a simple chat interface where users can input questions or requests and receive spoken-language responses from the GPT-OSS-120B model. It provides flexibility by allowing users to adjust the system prompt and temperature, enabling customization of the AI's behavior and output. This makes it suitable for experimentation and research with large language models, offering a platform to explore different conversational AI scenarios and model responses. The tool is open-source, licensed under Apache 2.0, promoting accessibility and collaborative development within the AI community.

Pixacare is a digital solution designed for healthcare professionals to document, measure, and remotely monitor wound healing. It provides a secure medical photo library for organizing patient images and videos, automatically classified by patient and date. The platform enables structured documentation of chronic, post-operative, or dermatological wounds, generating dynamic healing reports. Its medical AI automatically evaluates wound dimensions and evolution, offering objective, reproducible, and standardized tracking. Pixacare also facilitates reliable remote monitoring, allowing patients to send photos securely from home. The platform supports collaborative teamwork, secure messaging between caregivers, and integrates with existing hospital systems like DPI and GAM, ensuring data security with HDS & ISO 27001 certification.

papersgpt-for-zotero is a powerful Zotero plugin that transforms your research workflow by integrating state-of-the-art AI capabilities directly into your Zotero library. It allows users to chat with single or multiple PDFs, quickly gain key insights, generate summaries, and perform literature reviews. The tool supports a wide array of LLM models including ChatGPT, Gemini 3.1, Claude, DeepSeek V4, Grok, OpenRouter, Kimi 2.5, GLM 5, SiliconFlow, GPT-oss, Gemma 4, and Qwen 3.5. A standout feature is 'AutoPilot', an autonomous research assistant that can analyze hundreds of papers overnight, saving key insights directly into Zotero Notes. It also boasts 100% privacy and local data processing for many models, ensuring data safety and offline functionality. The plugin is compatible with Windows, Mac, and Linux, and supports Zotero 8 and 9.

MedINT is an AI-powered platform specifically designed to assist physicians in their clinical practice. It delivers personalized, evidence-based clinical information and recommendations, helping medical professionals make informed decisions across a wide range of patient care scenarios. The tool supports complex case analysis by providing literature-based insights, thereby aiding in the selection of optimal treatment options. By leveraging artificial intelligence, MedINT aims to enhance diagnostic accuracy and treatment efficacy, ultimately contributing to improved patient outcomes. Its focus on evidence-based recommendations ensures that physicians receive reliable and up-to-date medical knowledge tailored to specific patient dilemmas.

Slidebomb Summarizer AI is an online, free AI-powered tool designed to instantly summarize YouTube videos. Users simply paste a YouTube video URL, and the tool generates a clean and accurate summary in seconds, eliminating the need for downloads or prompts. It's ideal for quickly grasping the essence of lengthy videos, enhancing productivity, and improving learning efficiency. The platform offers various paid plans with increasing credit limits, batch summarization capabilities, and support for longer videos, including those without subtitles. Slidebomb emphasizes data security and provides an easy-to-use interface for generating and exporting summaries.