Do you want to get an overview over recent advances in autoencoders without reading 30+ papers? Check out our review paper at arxiv.org/abs/1812.05069. Google AI Mario Lucic ETH Zurich

The projects #GoogleInterns tackle and how they apply their new skills after finishing their internships often amaze us 🤩. Google AI #Zurich Research Intern and PhD student, Tomas Effenberger is no exception … we'll let him share why 👀➡️ goo.gl/eL19NP

It turns out that only a few parameters need to be trained to fine-tune huge text transformer models. Our latest paper is on arXiv; work Google AI Zürich and Kirkland. bit.ly/2DXCgb7 #GoogleZurich #GoogleKirkland

BERT is in #TFHub and here's a runnable notebook! 🎉 More BERT modules at tfhub.dev/s?q=bert

I added a summary of all our talks, posters and demos at #ICML2019 to olivierbachem.ch. Have a look if you are interested in disentanglement, label-efficient GANs or our new RL environment Google Research Football.

Congrats !! Francesco Locatello Mario Lucic Olivier Bachem

The Whova app schedule at #ICML2019 does not appear to be robust to adversarial words like “lunch”...

If you are interested to train NLP models to solve N tasks in a parameter-efficient fashion, Stanisław Jastrzębski will present our results on Thursday at 11:20am in room 104 #ICML2019. Poster #102 at 6:30pm. Joint work w/ Neil Houlsby Andrei, Bruna, andrea gesmundo Mona, Sylvain

You are serving multiple fine-tuned NLP models? We're presenting how to cut-down the RAM needed by few order of magnitude at poster 102! #ICML2019 proceedings.mlr.press/v97/houlsby19a… Stanisław Jastrzębski

Drop by the AutoML workshop tomorrow at #ICML2019 to discuss our work on graph search spaces for NAS!

Story time! Single neurons in the brain can’t be depended on for reliable information. Here are some neurons from our recent study, recorded twice in response to the same visual stimuli. Different neurons are active at different times! biorxiv.org/content/10.110…

Code released to adapt BERT using few parameters. Can be used to adapt one model to many tasks. Catastrophic forgetting not included. github.com/google-researc…

The next challenge for image classification: build a generic representation that can be adapted to any task with very few training examples.

Google AI Residency 2020 applications are open, with positions in many different locations including Europe and Africa. A fantastic program designed to jumpstart a career in Machine Learning. Apply at g.co/airesidency/ap… before Dec. 19, 2019. #AIResidency Google AI

A great overview of recommender systems by andbloch ! andbloch.github.io/An-Overview-of…

1/5 Self-Distillation loop (feeding predictions as new target values & retraining) improves test accuracy. But why? We show it induces a regularization that progressively limits # of basis functions used to represent the solution. bit.ly/2HnOACo w/Mehrdad Farajtabar P.Bartlett

Does knowledge distillation really work? While distillation can improve student generalization, we show it is extremely difficult to achieve good agreement between student and teacher. arxiv.org/abs/2106.05945 With Sam’s Cub, Pavel Izmailov, Polina Kirichenko, Alex Alemi. 1/10

Our friends at Optalysys did a tutorial on using Concrete to create an encrypted version of Conway's Game of Life! Check it out here medium.com/zama-ai/concre… #fhe #cryptography #homomorphicencryption #concrete #privacy

Hello from the #DLIndaba2022 conference with my colleagues Mehadi and @RachDean_Google at the Google booth! If you're interested to learn about our opportunities in the Machine Learning space, we'd love to chat with you! Pro tip: Ask for our QR code for opportunities ;)

Today is the African AI and ML Research Showcase! 🌍 #DLIndaba2022