2min chapter

Towards Data Science cover image

121. Alexei Baevski - data2vec and the future of multimodal learning

Towards Data Science

CHAPTER

Using Latent Representations as a Target for Your Model Training

The traditional task that i've talked about before is a reconstructing the original into data. This works reasonably well, but then you have to makea choice. What we found actually, well for wivh to bacc, is trying to using the latent representations as the target for your model to to preject. So we use this with contrestive learning. And algertings like b el and ina show that you can use a similar approach by reconstructing latent representations as a way to train your model. The problem with at is, if you're training the model t predict what the model is producing, then the model will very, very quickly a collapse into producing the

00:00

Get the Snipd
podcast app

Unlock the knowledge in podcasts with the podcast player of the future.
App store bannerPlay store banner

AI-powered
podcast player

Listen to all your favourite podcasts with AI-powered features

Discover
highlights

Listen to the best highlights from the podcasts you love and dive into the full episode

Save any
moment

Hear something you like? Tap your headphones to save it with AI-generated key takeaways

Share
& Export

Send highlights to Twitter, WhatsApp or export them to Notion, Readwise & more

AI-powered
podcast player

Listen to all your favourite podcasts with AI-powered features

Discover
highlights

Listen to the best highlights from the podcasts you love and dive into the full episode