TORUS AI

Torus AI is proud to become a member of the French ecosystem Aerospace Valley We will help the other companies in this ecosystem with our AI and Signal Processing capabilities

We are very proud to see Belle.ai (AI for dermatology, co-owned by Torus) appear on the agenda of the World Health Organization today :)

🔍 Advancing Dermatologic AI with Synthetic Data. ✨ The analysis of skin images is essential for driving progress in dermatologic AI and ensuring fair, unbiased healthcare solutions. 🚀 S-SYNTH has developed an innovative framework to generate synthetic skin images using a knowledge-based, multi-layered approach. This technology helps AI models: ✅ Train effectively with fewer real-world data ✅ Reduce biases and improve accuracy ✅ Enhance inclusivity across diverse populations 🎯 Highlights of this innovation: • Anatomically realistic 3D skin models • Simulation of diverse skin tones and lesion shapes • Improved segmentation for underrepresented groups 💡A step forward in making AI-powered healthcare more inclusive and effective. 👇Learn more about this research👇 #AI #DeepLearning #Dermatology #SyntheticData #Innovation #Healthcare

How do mathematics help us at Torus AI ? A colleague, who is a world-famous mathematician, asked us this question, because he wanted to show it to a class of students. Below we try to answer his question. Not long ago, coding algorithms into computer programs was a very hard and time-consuming task. Today, with automation tools and platforms and copilots, coding has become much easier and faster. This is great news for mathematicians, like many of us at Torus, who are good at creating algorithms but not necessarily handy at coding. An AI is basically an algorithm that contains “learnable variables”, i.e. coefficients that can be changed by some adaptation (learning) process that aims to optimize something. The word algorithm itself comes from the name of the great Persian mathematician Al Khwarizmi, and to talk about algorithms means to talk about mathematics. Our mathematical background helps us understand the AI. For us, AI is mathematics and not black magic. For example, the Kolmogorov-Arnold theorem tells us that basically any function (= algorithm), however complicated it is, can be composed from very simple functions, and this is the theoretical reason why artificial neural networks work. The convolutional neural networks (CNN) are very efficient at image processing, because they are related to wavelets, and before the appearance of CNN people already knew that wavelets were very useful for signal processing and image compression. The mathematical theories, ideas and concepts guide us in the creation and the improvement of AI solutions. For example, a few things that we actually do at Torus AI: ideas from topology can be used to enhance the precision of image segmentation and classification; transformation groups and dimension reduction (of spaces and of dynamical systems) lead us to things like reduction of the latent space in neural networks and the LoRA (= low-rank adaptations) in generative AI; variational principles allow us to design loss functions that are better adapted to our problems than the off-the-shelf loss functions; stochastic optimization methods allow us to design high-precision neural adaptive filters in signal processing; etc. If you are a very good mathematician, be happy, AI companies will need you 😀

Our co-founded Belle.ai (BelleTorus), the best AI for dermatology, continues to make the headlines. This time it is one of the five companies (among 300 applicants) selected to present medical AI at the US federal agency CMS (Centers for Medicare&Medicaid Management)

A big thank you to Dr Simon Thorpe for giving us a very inspiring lecture on spiking nets and the brain !

📢 Guest Speaker Announcement 🗣 Did you know that every other thursday we host talks with guest speakers in our "AI Wine" seminar series which is open to everyone who is passionate about AI ? For those people who can come to our place (Torus building in the Parc Technologique in Ramonville) to attend the seminar in person, there are always wine and cheese (and other food and soft drinks) after each seminar :) This Thursday, October 31, at 4pm we will be pleased to welcome Dr. Simon Thorpe, a renowned world leader in the interactions between AI and neuroscience, for a discussion on recent developments in AI. 💡 While AI systems now perform at or above human levels in areas like image and speech analysis, there are still unique features of the brain that AI has yet to replicate. Simon will address two important aspects: 👉Ultra-sparse coding – a potential reason why the brain operates on just 20 Watts. 👉Rapid learning – enabling neurons to recognize patterns with minimal exposure. 🤝 Join us to explore how these insights could guide the next steps in AI development (in person or by Zoom, this Thursday at 4pm CET): https://lnkd.in/eVwiACAF

🚀 We are hiring! 🚀   Are you passionate about artificial intelligence and machine learning? Do you want to be part of an innovative startup that’s growing fast?   We are looking for a Machine Learning Scientist to join our dynamic team and work on high-impact projects.   👇 See below the job description and send us your CV at [email protected]   #Hiring #MachineLearning #AI #Startup #Innovation

Acne AI and Skin Cancer AI at MICCAI 2024 Congratulations to the three Torus scientists that attended MICCAI2024 with a paper on Acne AI that attracted huge interest among participants who asked many questions to our team during two straight hours ! Our team also attended the workshop ISIC2024 at MICCAI2024 as a member of the ERC-funded project ITOBOS on skin cancer detection. Professor Rafael Garcia, also present at ISIC2024, is the coordinator of this projet. https://lnkd.in/dukUmgeZ

This Nature study probably forgot about companies like Torus AI, where scientists will find better conditions for doing science than in academia :)