@Jacoed@arnosolin The LeJEPA paper proves it as the optimal distribution for downstream probes + prevents collapse without teachers, etc.
Worth a read: https://t.co/1vf6L7eqDS
@Jacoed@arnosolin Both are basically batch-level anti-collapse regularizers - but the difference is what they actually constrain.
EBGAN just tries to push embeddings apart in angle, while SIGReg tries to make the whole distribution look like an isotropic Gaussian (in scale/shape)
@Cory29565470@arnosolin Yes - in AudioSet (the pretraining dataset) there are a few thousand helicopter clips, though it never saw this specific one or any labels
๐ฅ #LeAVJEPA: A simple recipe for self-supervised learning from sound and vision ๐ฅ
One shared encoder, trained without class labels. Dropping a modality makes audio & video learn a common representation.
๐งต
Prepare for takeoff. โ๏ธ Flight simulator is now available globally on web to all users. https://t.co/hQP0No142P
We've recently added many our most powerful professional desktop features to web. Elevation profiles, new import types, but there's always been one other feature you've been asking us to add to the web version of Google Earth, just for fun...
Where will you fly? Share your best maneuvers, views, and flyovers with us!