Recordings of the NeSy 2025 keynotes are now available! 🎥
Check out insightful talks from @guyvdb.bsky.social, @tkipf.bsky.social and D McGuinness on our new Youtube channel www.youtube.com/@NeSyconfere...
Topics include using symbolic reasoning for LLM, and object-centric representations!
Yes don’t try this at home
Working on Veo's ingredients to video feature has been a blast. Check it out on flow.google
Two life updates:
1) About a year ago I decided to join the Veo team to work on capabilities. It’s been a fun ride! Excited for what’s still to come.
2) I've been busy caring for a newborn the past couple of days 🥰 Excited for the incredible world he will grow up in. Veo's impression below:
Check out @tkipf.bsky.social's post on MooG, the latest in our line of research on self-supervised neural scene representations learned from raw pixels:
SRT: srt-paper.github.io
OSRT: osrt-paper.github.io
RUST: rust-paper.github.io
DyST: dyst-paper.github.io
MooG: moog-paper.github.io
I'm excited to announce that I have no idea what day of the week it is and I'm hoping I can keep this up for the rest of the year
Congrats!! Lots to think about
I gave a talk on Compositional World Models at NeurIPS last week 🌐
The recording is now online: neurips.cc/virtual/2024... (for registered attendees; starts at 6:06:00)
Workshop: compositional-learning.github.io
Welcome to Google!
That’s a great recommendation, thanks!
Yet our first two days looked like this 😄
Thanks, Durk!
Blue skies over Joshua Tree 🌌
Sending reminders really shouldn’t be something we have to deal with manually. Clearly there’s headroom in designing better incentive structures.
I think there is still *a lot* of headroom for automation while ultimately reducing potential for human error (or just laziness on the AC part).
I think that depends on the conference. ICLR pretty much already automated the reviewer assignment part using a new bidding system that seemed to work pretty well. Manual AC assignments were heavily discouraged, only minor adjustments were needed.
Yeah, it'll have to be a tightly kept secret among people who enter the exclusive AC circle 🙃
Hot take: 90% of what ACs/SACs do could in principle already be automated (with the remaining 10% being process oversight and borderline decision making).
At least right now, it seems like reviewers have the more important job for the most part.
Agreed, important to find the right balance. Deeply caring about something doesn’t mean one should neglect other aspects of life (especially health, sleep, nutrition, social connection, downtime, …).
Waymo deserves to be the number one tourist attraction in San Francisco right now, and it's not even close
For like ~$11 you get to ride in a genuine self-driving car with up to four people!
Wildly entertaining
Being totally obsessed with your work really helps with motivation and with getting things done. Exciting times.
🙋♂️
Let’s welcome @ellis.eu to Bluesky and give them a follow! 🦋
Hello World!
Thrilled to announce Boltz-1, the first open-source and commercially available model to achieve AlphaFold3-level accuracy on biomolecular structure prediction! An exciting collaboration with Jeremy, Saro, and an amazing team at MIT and Genesis Therapeutics. A thread!
We're planning to open source, but no ETA yet. Stay tuned :)
We'll present this work at NeurIPS (Spotlight, yay 🙌) this year - come find us at the poster soon or reach out if you have questions!
This was a fun project with an amazing set of collaborators (and co-leads Sjoerd van Steenkiste and @zdanielz.bsky.social) at Google DeepMind / Google Research.
MooG can provide a strong foundation for different scene-centric downstream vision tasks, including point tracking, monocular depth estimation, and object tracking.
Especially when reading out from frozen representations, MooG is competitive with on-the-grid baselines.
Under the hood, MooG uses two independent cross-attention mechanisms to write to – and read from – a *set* of latent tokens that are consistent over time.
Think of it as a scene memory consisting of a set of tokens that can flexibly bind to individual scene elements.
