🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Our paper on learning test-time augmentations has been accepted to #UAI2020! Greedy policy search (GPS) is a way to learn a data augmentation policy for a pre-trained model, and it boosts the performance of a test-time augmentation ensemble both in- and out-of-domain! 1/4

Amir Zamir (Amir Zamir) has openings for #PhD students in his group. Find out more about his #research at: vilab.epfl.ch/zamir/, and learn more about our EPFL #EDIC #computerscience PhD program: go.epfl.ch/phd-edic

Happy to share our SimCLR PyTorch implementation with distributed training support! Joint work with Arsenii Ashukha github.com/AndrewAtanov/s…

Алексея из зала «суда»

Happy our work on transfer evaluation is finally out! We highlight the importance of the choice of 1) the transfer dataset sizes, 2) control baselines, and 3) downstream tasks. We also suggest a metric that displays the transfer performance in an accessible and informative way.

I’m at #NeurIPS2022 this week. DM me if you want to chat about deep learning generalization, transfer/meta-/self-/semi-supervised learning.

We are releasing 4M-21 with a permissive license, including its source code and trained models. It's a pretty effective multimodal model that solves 10s of tasks & modalities. See the demo code, sample results, and the tokenizers of diverse modalities on the website. IMO, the

Are you interested in designing the morphology of intelligent agents, or just want to learn something new? Join us today at the #CVPR2024 tutorial at 9:00 AM in Summit 344. comp-design.epfl.ch

Some exploding design at #CVPR2024 ⁦#CVPR2025⁩ , be careful! colab.research.google.com/drive/1h9MD64o…

Scaling up is everywhere (and is effective), but what about scaling down? I am excited to share our new work on solving vision tasks with simple low-resolution vision sensors (1-pixel cameras).

Standard zero-shot inference makes independent predictions for each task’s input. Instead, we propose joint inference, which makes predictions for multiple inputs simultaneously by maximizing the joint probability of their answers. We find that this approach boosts zero-shot