Disentangling words from images in CLIP and SOTA video self-supervised learning | Your Daily AI Research tl;dr - 2022-06-19 ??

Welcome to your official daily AI research tl;dr (often with code and news) for AI enthusiasts.

In this newsletter, I share the most exciting papers I find daily, along with a one-liner summary to help you quickly determine if the article (and code) is worth investigating. I will also take this opportunity to share daily exciting news in the field.

Now, let's get started with this iteration!

1?? iBoot: Image-bootstrapped Self-Supervised Video Representation Learning

Improved video self-supervised learning beating SOTA and with a more efficient training scheme (i.e. in less epochs and with a smaller batch) using a strong image-based model, pre-trained with self- or language supervision, enabling the model to learn strong spatial and temporal information without relying on the video labeled data.

Link to the paper: https://arxiv.org/abs/2206.08339

2?? Disentangling visual and written concepts in CLIP

This work investigates the entanglement of the representation of word images and natural images in CLIP, confirming this existing entanglement and they "find that [their] methods are able to cleanly separate spelling capabilities of CLIP from the visual processing of natural images."

Link to the paper: https://arxiv.org/abs/2206.07835

Code: https://github.com/joaanna/disentangling_spelling_in_clip

?? A Graphcore processor?outperformed?an Nvidia A100 GPU in Twitter ML testing

“The choice of hardware for implementing Graph ML models is a crucial, yet often overlooked problem. [...] Graph neural networks offer a means of finding order in complex systems, and are commonly used in social networks and recommender systems. However, the dynamic nature of these environments make these models particularly challenging to train, the trio explained.”

Read more: https://www.theregister.com/2022/06/16/graphcore_nvidia_gpus/

And we are already at the end of this iteration!

I hope you liked the format. Please subscribe and share it with your techy friends if you've enjoyed it or if it was helpful. I would be glad to hear from you to make it even better with time.

Thank you for reading, a fellow AI enthusiast and researcher.