Research & Education

torch-light is an open-source deep learning resource built on PyTorch, designed to help users learn and implement deep learning concepts. It provides fundamental neural network implementations, including Logistic, CNN, and RNN models, making it accessible for those starting in the field. Beyond basic examples, torch-light also includes advanced implementations using more complex models, catering to users looking to deepen their understanding and practical skills. This tool is ideal for developers and students who want to explore deep learning with a hands-on approach, offering a practical way to understand how various neural network architectures function and can be applied.

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.

VirtualHome is an interactive platform and API designed for simulating complex household activities using programs. It enables multi-agent simulations where agents can interact with environments by picking up objects, switching appliances, and more. The platform offers two simulators: the Unity Simulator for generating videos of activities and the Evolving Graph simulator for tracking environmental changes. It supports various ground-truth streams like time-stamped actions, segmentation, and optical flow, making it suitable for training agents in embodied AI tasks. VirtualHome also features procedural generation for unique environments, enhanced physics, time management, and improved lighting, with a Python API for easy integration and control.

wevi, the Word Embedding Visual Inspector, is an open-source tool designed for visualizing and exploring word embeddings directly within a web browser. This tool is invaluable for data scientists and developers working with natural language processing, as it provides a clear visual representation of the relationships between words. By offering an intuitive interface, wevi aids in debugging NLP models and gaining deeper insights into how word embeddings capture semantic similarities. It is compatible with popular browsers like Chrome and Firefox, making it easily accessible for a wide range of users. The project encourages community contributions to further enhance its features and capabilities.

WebDataset is a Python-based I/O system specifically engineered for both large and small-scale deep learning tasks, providing robust integration with PyTorch. It streamlines data handling by organizing training samples and datasets within tar files, adhering to specific conventions for efficient access. This approach is particularly beneficial for high-performance data loading, reducing I/O bottlenecks during model training. The tool's design focuses on optimizing data pipelines, making it a valuable asset for developers and data scientists working with extensive datasets in machine learning projects. Its emphasis on structured data organization within tar files facilitates scalable and reproducible research.

Harvard University is a world-renowned institution dedicated to excellence in teaching, learning, and research. It offers a wide array of undergraduate and graduate degree programs across its many schools, including Harvard College, Harvard Business School, Harvard Law School, and Harvard Medical School. Beyond academics, the university provides extensive resources such as libraries, museums, and athletic facilities. Harvard is committed to developing leaders who contribute globally and actively engages in research across various fields, including artificial intelligence and mental health. The institution also highlights its rich history, including its role in the American Revolution, through various exhibits and events.

Mode Maison is an independent research lab dedicated to exploring the intersection of physics-based simulation, artificial intelligence, material science, design, and retail. The company is actively developing multimodal Large World Models (LWMs) designed to simulate and generate diverse experiences. Their work aims to introduce a new dimension of intelligence, creativity, and reality, leveraging AI to push boundaries in these fields. While specific features are not detailed, the focus is on advanced AI research and development, suggesting a highly technical and innovative approach to complex problems in scientific computing and related domains.

NOARKTECH is a nature-tech company specializing in AI-powered environmental intelligence. It develops sensor networks and analytics platforms designed to foresee, uncover, and mitigate ecological threats and climate hazards. The platform helps predict disasters, protect lives, and preserve ecosystems by building climate resilience. NOARKTECH's technology is applied across various environments, including forests, farms, and other frontline ecosystems, providing crucial data and insights for climate action. The company aims to unlock the 'secret language of nature' through DePIN Enabled Edge AI Sensor Networks to support a better global environment.

CrossPrism Nature Id is a mobile application designed for outdoor enthusiasts to identify natural organisms in real-time. Utilizing its proprietary CrossPrism AI engine, the app can identify plants, fungi, and animals directly from a user's camera, photo library, or even from images pasted from the internet. A key differentiator is its offline functionality, ensuring identification is possible without an internet connection, prioritizing user safety, privacy, and performance. Beyond identification, it includes an image evaluator to guide users in capturing optimal photos for better results. The app supports training and customization of its AI models.

SymbyAI is an AI-powered platform designed to accelerate scientific research by automating critical stages of the review process. It allows authors to submit papers and associated data for rapid AI analysis, including supplementary data, code, and images. A key feature is its ability to attempt replication of study methods, generating necessary code, proofs, and mathematical calculations. Within 15 minutes, SymbyAI provides detailed analysis, identifying potential issues, inconsistencies, and offering suggestions for improvement. This dramatically reduces the traditional review time from months to minutes, enhancing research quality, validating findings, and improving overall integrity. It aims to combat research waste by providing a thorough and efficient review process for individual researchers and large organizations alike.

Urban AI Lab is a pioneering research hub at the University of Florida dedicated to transforming future cities through the application of artificial intelligence and computational science. The lab's mission is to advance the fundamental theory of urban science by leveraging AI-driven approaches across various interdisciplinary domains, including urban planning, civil engineering, and the built environment. Their research is structured around three key themes: Resilient Urban AI, Ethical Urban AI, and Generative Urban AI. The lab aims to push the boundaries of urban science to create smarter and more equitable cities, offering opportunities for collaboration and research.

Leman Biotech is a clinical-stage company specializing in the development of innovative cancer immunotherapies. Founded by a team from EPFL and XtalPi, the company leverages a unique combination of immune metabolic reprogramming and advanced artificial intelligence. Their core technologies include a metabolic reprogramming macromolecule platform (Meta 10) to reactivate exhausted T-cells, a metabolic-enhanced cell therapy platform, and an AI super-factor platform for efficient computational design of biomacromolecules. Leman Biotech focuses on researching, producing, and commercializing new oncology treatments, with recent successes including FDA IND approval for their META 10-19 injection and positive clinical trial results for their metabolic-enhanced CAR-T therapy.

Idler is a platform dedicated to providing reinforcement learning environments, offering essential tools and resources for the development and testing of AI models. It is specifically designed to support researchers and engineers in training AI agents within simulated environments. The platform aims to facilitate the exploration and implementation of reinforcement learning techniques, enabling users to build and refine intelligent systems. By offering a dedicated space for experimentation, Idler helps bridge the gap between theoretical AI concepts and practical application, allowing for iterative development and performance optimization of AI agents.

The Knowledge Engineering and Machine Learning group (KEMLg) at UPC Universitat Politècnica de Catalunya is dedicated to advancing AI techniques through rigorous analysis, design, implementation, and application. Their research is geared towards understanding and improving the operation and behavior of complex real-world systems. KEMLg's work spans critical domains such as health, environmental processes, and the industrial sector, demonstrating a commitment to applying AI solutions to pressing societal and technological challenges. The group actively contributes to the broader AI strategy of UPC, as highlighted by initiatives like LEIA UPC, which aims to consolidate the university's international leadership in AI research and knowledge transfer.

Autonomous-Ai-drone-scripts is a GitHub repository dedicated to developing advanced autonomous navigation and obstacle avoidance systems for multi-rotor vehicles. The core innovation is an end-to-end AI model that processes sensor inputs directly to output drone control commands. The project also features a fully functional autonomous system for tracking moving targets using an AI-based object detection model with a camera and LiDAR, which has been tested in real flights. The repository includes all necessary tooling for gathering, training, and validating AI-based pilots for Pixhawk-based multi-rotors. It also provides scripts for autonomous person following using a live RGB camera feed and a MobileNet AI model, along with a LiDAR for distance measurement. The system is designed to run on a Jetson Nano, enabling fully autonomous operation without external data connections.

G2Q Computing is at the forefront of developing hybrid quantum-classical software solutions, specifically designed for optimization and machine learning. This innovative platform aims to provide superior computing performance for complex challenges across various industries. By combining the power of quantum and classical computing, G2Q Computing offers advanced tools for researchers, scientists, and engineers who require efficient solutions for intricate computational problems. The platform is cloud-based, ensuring accessibility and scalability for its users. It is engineered to transform industries by enabling breakthroughs in areas that demand high-performance computing and sophisticated algorithmic approaches.

CuspAI is a frontier AI company dedicated to solving the challenge of breakthrough materials needed for human progress. By harnessing advanced AI, the platform aims to accelerate materials discovery from millennia to mere months. The founding team comprises world-class researchers in AI, chemistry, and engineering, making it a robust solution for complex material science problems. CuspAI is positioned to usher in an era of on-demand materials, offering significant advancements in various sectors. The tool is designed to support researchers and engineers in their pursuit of innovative material solutions.

Awesome-Deep-Camera-Calibration is a comprehensive open-source repository dedicated to deep learning applications in camera calibration. It serves as a valuable resource for researchers and practitioners in computer vision, offering a curated collection of papers, methods, and benchmarks. The repository details popular calibration objectives, various camera models, and extended applications, including novel calibration representations that show potential to replace traditional objectives for neural networks. It also includes a structural and hierarchical taxonomy of deep learning-based camera calibration, a concise milestone of methods, and statistical analyses. A notable feature is the benchmark dataset, which covers diverse camera models and environments, providing accurate ground truth and labels for evaluation.

smalldiffusion is a lightweight, open-source Python library designed for training and sampling from diffusion and flow models. It prioritizes ease of experimentation, allowing developers and researchers to quickly train new models or develop novel samplers. The library supports a variety of models including MLP, U-Net, and DiT, and multiple parameterizations such as score-, flow-, or data-prediction. It offers dataset support for 2D toy datasets, pixel, and latent-space image datasets, with example training code for FashionMNIST, CIFAR10, and Imagenet. smalldiffusion also provides concise implementations of diffusion transformers and supports conditional training with classifier-free guidance, making it a versatile tool for those working with diffusion models.

FinalSpark is at the forefront of biocomputing, transitioning AI from traditional digital processors to biological neural networks. The company cultivates living neurons in cell cultures to create a new form of computing power, addressing the high energy consumption limitations of current AI technologies. Their research focuses on replicating and surpassing natural biological efficiency, with the goal of developing self-organizing, continuously learning, and highly energy-efficient bioprocessors. These biocomputers are designed to be scalable through natural expansion, offering a simpler and more powerful alternative to silicon-based CPUs and GPUs. FinalSpark's vision is to unlock limitless potential for enhancing life on Earth through this groundbreaking technology.

The UKRI AI Centre for Doctoral Training (CDT) in Decision Making for Complex Systems is a 4-year doctoral program offered jointly by The University of Manchester and the University of Cambridge. This program is designed to educate the next generation of AI researchers, equipping them with the skills to develop and deploy new machine learning models capable of efficiently handling uncertainty in complex systems. The CDT integrates machine learning research with applications in physics, astronomy, engineering, biology, and material science, alongside a cross-cutting theme of using AI to enhance business productivity. The research is ultimately applied to real-world scenarios, fostering innovation and leadership in the AI sector.

DeepRoot Minds is a born-AI company specializing in healthcare intelligence, mental wellness companions, and enterprise ontology platforms. Rooted in truth, their AI systems leverage ontology-driven knowledge graphs to prevent hallucinations, ensuring verifiable and traceable answers. They offer HealthTech solutions like mobile diagnostics for dermatology, ophthalmology, and dental AI, providing clinical insights from smartphone photos. For mental health, they develop advanced AI platforms embodied in social robots for longitudinal care, starting with pediatric applications. Additionally, DeepRoot Minds provides enterprise ontology platforms and knowledge graphs to eliminate AI hallucinations across various products and agents, with a focus on provenance-aware and graph-grounded retrieval.

Datalchemy is a consulting firm specializing in the industrialization of AI and data projects, guiding clients through the entire lifecycle from initial R&D to full production. They offer comprehensive services including data engineering for AI, development of AI projects (POC, MVP), and ongoing management and maintenance with monitoring, drift management, and continuous validation. Datalchemy also provides tailored AI training programs for both technical and business teams. Their unique approach combines cutting-edge scientific research with robust engineering practices, ensuring solutions are robust, ethical, and economically viable across various technical environments like cloud, on-premise, or embedded systems.

DeepSNAP is a Python library designed to facilitate efficient deep learning on graphs. It offers robust support for flexible graph manipulation, integrating with powerful graph libraries like NetworkX and deep learning frameworks such as PyTorch Geometric. The library provides a standard pipeline for tasks like dataset splitting, negative sampling, and defining node/edge/graph-level objectives, ensuring transparency for users. DeepSNAP also efficiently supports flexible and general heterogeneous Graph Neural Networks (GNNs), accommodating both node and edge heterogeneity. Its intuitive API allows users to control message parameterization and passing, making it easy to use for those familiar with PyTorch Geometric.