1/ When using multimodal models, we often want to infer over modalities unseen together in training. Prior works typically address this by learning a shared contrastive embedding space between modality pairs, but is this approach principled? We prove that it is! (under certain assumptions)
Paper+Code: https://t.co/d2bOm79ie7
one beautiful way to explore the web is using the site: Google search operator on TLDs. for example:
site:computer interaction
shows domains that end in .computer and talk about interaction. you could also use this if you want only .gov results about a topic, etc.
@ronawang It was great - made it to high schools in 15 states observing classes. I went in without any real plan and left with clear ideas about how students can change their school experience. As a bonus I got access to lots of student and teacher beta testers :)
@damonchen been there! if this is for tiktok/shorts/reels, might help to try doing a simple task while filming - makes it more engaging for viewers and less awkward for you. My best video ever was me talking while tying a tie :)
@tobyshorin I hope some day I'll revive it and you'll see an ad where a blue, cookie monster type creature shouts WOOB and throws cheese at a disgruntled Docs user.
@tobyshorin Yeah... they should clean stuff like that up, could be the everything platform. I was so annoyed always writing blog posts in Docs then copy-pasting into my site that I built a now-defunct service called Woobsite, a website builder that auto-syncs to your google docs.