Alexandre Alahi

Thanks RTS for the interview. Autonomous driving level 3 will be allowed in Switzerland by 2025:

Is trajectory prediction promptable? Yes! Social-Transmotion, a generic Transformer-based model, makes it possible by integrating diverse visual cues as prompts. Joint work with Saeed, Yang, Kaouther. #ICLR2024 Paper: Webpage/code:

🎉 Exciting news!  Our team has achieved top rank on the nuScenes trajectory prediction leaderboard! Our strategy? Unprecedented data scaling with our new open-source framework . Project page: Special thanks to Lan, hossein, kaouther, Eloi, & Matthieu!

RT @epflENAC: Les véhicules neufs seront tous équipés d'un premier niveau d'assistance à la conduite en Suisse dès cette année. La @RadioTeleSuisse fait le point sur ce domaine d'innovation avec le prof. @AlexAlahi @EPFL.

RT @LeTemps: 🚗 #Débatsdété Pour ce 4e épisode, @AlexAlahi (@EPFL) passe en revue les avantages et les défis de l'#IA au service de l'automobile

RT @KathrinGrosse: Dear All, are you working in industry with AI/ML or know someone who does? You are dearly needed! It is time for another survey to threat model ML/AI in practice. Please take 10 minutes to click through our questions or share with someone who can help:

ChatGPT has shaken the world with its advancements in NLP. However, AI has still been found lacking in certain tasks, such as autonomous driving. Why? Click below to start the conversation: #AI #SelfDrivingCars #sociallyawareAI #tedx

RT @epflENAC: Des scientifiques de l'ENAC collaborent avec @Dartfish et le @lausannehc pour repousser les limites de l’analyse vidéo sportive. Avec la contribution du prof. @AlexAlahi. L'objectif? La détection et le suivi automatisé des athlètes.

Benchmarking trajectory forecasting models: Case of the missing test set! by Parth Kothari

RT @bendingspoons: @AlexAlahi, winner of the @CVPR Open Source Award and ICDSC Challenge Prize, gives us a glimpse into the future of #visualintelligence in #Transportation—#SelfDrivingCars powered by socially and ethically aware #AI.

RT @liu_yuejiang: Distribution shifts are grand challenges for machine learning models. When it comes to motion, what are the unique problems and opportunities? We provide an answer in our #CVPR2022 paper: “Towards Robust and Adaptive Motion Forecasting: A Causal Representation Perspective” 1/6

RT @SSaadatnejad: "Vehicle trajectory prediction works but not everywhere" #CVPR2022 Come and check our poster next week (June 24th morning, Session: Poster 4.1) Project website and code: Paper: Joint work with @MHBahari @smoosavid @AlexAlahi

Thank you Andrej @karpathy (Director of AI at Tesla) for the great interactive Q&A during our “CIVIL 459: Deep Learning for Autonomous Vehicles" class. We all loved the interactions (including our robots!).

Do you want to generate photorealistic images? We show that you can improve GAN-based methods by making their Discriminator semantically aware. Code: Paper (ITS): Project website:

Several postdoc positions are available in our VITA lab. Feel free to apply here:

RT @trekkinglemon: At @EPFL_EN we believe Machine Learning is poised to fundamentally change the way we do Science and engineer solutions to our most pressing challenges. If you’re a young researcher sharing this passion, apply to our new #AI4Science fellowship:

RT @liu_yuejiang: Key takeaway * both take-specific info (e.g. SSL) and model-specific info (e.g. feature moments) are crucial * we should rethink what to store, in addition to model parameters, for robust deployment Link: w/ Parth, Bastien, Baptiste, Taylor, @AlexAlahi

RT @liu_yuejiang: Test-time training (TTT) is an emerging approach to tackle distributional shifts. Yet, it's still not robust for practical use. Have you ever wondered when TTT fails or thrives? Our paper TTT++ #NeurIPS2021 provides an answer. To know more, plz join us at Session 1 this Tue :)

RT @EPFL_Imaging: Stay informed on Imaging @EPFL_en : newsletter#2 is out. Don’t miss @AlexAlahi Lab presentation on his fascinating work on visual intelligence for transportation

RT @liu_yuejiang: Can we learn a robust motion representation (e.g., forecasting/navigation) from ill-distributed data? Our #ICCV2021 paper 'Social-NCE' provides an answer! ( Join us at Session 12 this Fri :) w/ @qi_yan98, @AlexAlahi via @YouTube