Today something crazy happened. POT has reached 1000 citations (total) 🤩🚀. Very proud to be part of a scientific community that acknowledges open source research software. Please continue to use, cite and contribute to POT ! Small🧵below for those interested

Thank you for supporting my work 🙂🙏

Last but not least, a new pre-commit scheme to automatically correct common programming mistakes likely to be made by our future contributors. Thanks to our dear @RFlamary. Many thanks to all POT contributors and users ! Road to POT 1.0 📈 7/7

New features to support asymmetric structures and different inner losses in partial GW solvers with also speed improvements thanks to a novel partial conditional gradient solver. 6/7 https://t.co/Q8ML8K8i4V

New unbalanced FGW and Co-optimal transport solvers to promote robustness to outliers in graph matching problems. Thanks to Huy Tran. 5/7 https://t.co/PyoFwvSBwP

A semi-relaxed FGW barycenter solver, coupled with new initialization heuristics for the inner divergence computation. 4/7 https://t.co/mOYYzh7X9L

Generic unbalanced OT solvers now supportting any non-negative reference measure in the regularization, and novel translation invariant UOT solver showcasing a higher convergence speed. Thanks to Huy Tran and @Clement_Bonet_ 3/7 https://t.co/6s5wrVANaG

novel Gaussian Mixture Model Optimal Transport solver to compare GMM while enforcing the transport plan to remain a GMM. Thanks to Eloi Tanguy. 2/7 https://t.co/fovA0Pirl5

Python Optimal Transport (POT) 0.9.5 released: new solvers for Gaussian Mixture Model OT, unbalanced OT, semi-relaxed (F)GW barycenters, unbalanced FGW and COOT, partial GW. more details in 🧵1/7 https://t.co/c0TCTHQTnR

Fine-tuning pre-trained models leads to catastrophic forgetting, gains on one task cause losses on others. These issues worsen in multi-task merging scenarios. Enter LiNeS 📈, a method to solve them with ease. 🔥 🌐: https://t.co/65fMr3nmBb 📜: https://t.co/3jcyYMG69z 🧵 1/11

We are #hiring #PhD students via the #ELLIS PhD Program! Interested to work on (generative) #AI method development to advance personalized medicine? Located at the School of Computer Science and Life Sciences at @EPFL🇨🇭and part of the @EPFL_AI_Center, we are closely collaborating

Super happy to announce the release of TorchDR 0.1 : the Dimensionality Reduction (DR) library using @PyTorch. Joint work with an amazing team : @RFlamary , @mathusmassias, @nicolas_courty, @CedricCuaz, Titouan Vayer, Aurélien Garivier. Thread 🧵. Repo :

Super proud to be part of this new open-source python library! Skada is set to transform domain adaptation, enabling robust learning across datasets with distribution shifts 🚀

Extremely happy to announce the (alpha) release 0.3.0 of SKADA Scikit Adaptation, a python toolbox for domain adaptation fully compatible with @scikit_learn and @PyTorch Code : https://t.co/MHFsi7bPhH Doc: https://t.co/6ujVkwCX4A

Python Optimal Transport (POT) 0.9.4 released : works with numpy 2+, new quantized and low rank Gromov-Wasserstein solvers, and general unbalanced/regularized solvers in ot.solve. More details in 🧵1/7

If you are a student or academic researcher and want to make progress towards human-level AI: >>>DO NOT WORK ON LLMs<<< LLMs are an off ramp. Thousands of engineers are working on LLMs with enormous computing resources. The only way you could possibly contribute is by analyzing

Wouldn't it be great if we could merge the knowledge of 20 specialist models into a single one without losing performance? 💪🏻 Introducing our new ICML paper "Localizing Task Information for Improved Model Merging and Compression". 🎉 📜: https://t.co/JC4mdujKkd 🧵1/9

Gromov-Wasserstein distance is great, but what if its invariances were detrimental sometimes? Our @aistats_conf paper (presented tonight) introduces Augmented GW which blends fine control over GW's invariances and guides its mapping with prior knowledge

Here is a grid of different 1D optimal transport solutions for different regularizations (columns) and unbalanced marginal penalizations (rows). The example is coming to POT shortly and is done with a for loops because we now implement all those solvers in one function (ot.solve)

Here are the slides of my talk about learning on graphs with optimal transport at OTML Workshop @NeurIPSConf https://t.co/lQyIBuLArw