Ron Vainshtein retweeted
Animation ๐ค Robotics
ProtoMotions GTC 2026 release โ bridging the gap between digital humans and real humanoid robots.
Train in simulation. Deploy on hardware. One framework, one codebase.
nvlabs.github.io/ProtoMotionโฆ
Ron Vainshtein retweeted
Just arrived at San Diego for #Neurips2025 โ๏ธ
If you're interested in
- Tactile sensing ๐ซฐ
- Uncertain decision making โ๏ธ
- Medical robotics ๐ฉบ
Reach out or come see if you process touch better than AI at our poster on Fri 16:30-19:30 #2206.
@NeurIPSConf
Can we use tactile sensors for soft object imaging?๐ค
YES!
Toward Artificial Palpation: Representation Learning of Touch on Soft Bodies.
@shafer_eli , @talt80, @Efrat_Shimron, @AvivTamar1.
Paper, videos, data: zoharri.github.io/artificialโฆ
Check us out at @NeurIPSConf #neurips2025!
Ron Vainshtein retweeted
1. Train a behavior foundation model with a rich token-based representation (eg, MaskedMimic).
2. Fine tune a state-based tokenizer using RL, leveraging the BFM for meaningful gradients.
=> Converge 6x faster, 50x less params, dramatically improved robustness.
๐Using BFMs to learn a new task with less than 0.1% additional params while still enabling user input??๐
Task Tokens: A Flexible Approach to Adapting Behavior Foundation Models
๐: arxiv.org/abs/2503.22886
๐ป: github.com/rvainshtein/task-โฆ
w/@ZoharRimon @shiemannor @ChenTessler
๐งต1/10
Ron Vainshtein retweeted
Behavior models are great โ but getting them to do *exactly* what you want takes *tons* of prompt hacking (just like vision-language models) ๐ช
No more, with our new work ๐ฐ -
Task Tokens: A Flexible Approach to Adapting Behavior Foundation Models
๐Using BFMs to learn a new task with less than 0.1% additional params while still enabling user input??๐
Task Tokens: A Flexible Approach to Adapting Behavior Foundation Models
๐: arxiv.org/abs/2503.22886
๐ป: github.com/rvainshtein/task-โฆ
w/@ZoharRimon @shiemannor @ChenTessler
๐งต1/10
Ron Vainshtein retweeted
Parallels between AI and #LK99:
- Both are modern-day alchemy. Just try more recipes until eureka.
- The holy grail is simpler than we expected.
- Lots of hyperparameters to tune.
- Random seeds matter.
- Arxiv is the new battlefield.
- Scaling up is the key.
- Tend to break the internet with every update.
- Fastest growth in Twitter experts.
- Democratization: there will be GPT-4's from big corps, and Alpaca's from grassroots.
Ron Vainshtein retweeted
ืืืืืืืื ืืืฉืืื ืฉื ืกืื ืืืืื ืืืื ' ืชื ืืืื.
ืื ืืืืื, ืื ืื ืืืืชื ืฉื.
ืจืืื ืืช ืืืงืืื ืืืืจืง ืืื ืืช ืื ืืืืฉืืืช ืฉืื ืคืืืก ืชืืืื.
ืื:
ืกื ืืืืื, ืื ื"ื OpenAI ืืืฉื ืฉืืืงืื ืคืืืช ืืืืจืื - ืืืืื ืกืืฆืงืืจ, ืืืืขื ืืจืืฉื ืฉื OpenAI ืฉืืื ืืฉืจืืื ืืฉืขืืจ ืฉืืื ืคื ืืืจืฅ. >>
youtube.com/watch?v=mC-0XqTAโฆ
Ron Vainshtein retweeted
Nice, love the concept ๐ฆ
Some of my early experiments with @runwayml GEN-1
Itโs still early days of AI video - this tech will only get better. A whole new generation of filmmakers is gonna be able to make whatever they want on zero budget
See below for my process
Ron Vainshtein retweeted
Really excited to see this paper released! Great work by the authors!
Hoping to see it integrated into diffusers soon ๐คฉ @huggingface
Attend-and-Excite: Attention-Based Semantic Guidance for Text-to-Image Diffusion Models
abs: arxiv.org/abs/2301.13826
project page: attendandexcite.github.io/Atโฆ
github: github.com/AttendAndExcite/Aโฆ
Ron Vainshtein retweeted
โA Simple Configuration Library is All You Needโ
Happy to share pyrallis ๐, a small project we've been working on. Pyrallis is a simple library, derived from simple-parsing and inspired by Hydra, for automagically creating configs from dataclasses.
eladrich.github.io/pyrallis/
Ron Vainshtein retweeted
What if we thought of a pretrained NN's weights as a coordinate-value system? Then we could learn a neural representation for that system (just like folks have been doing for 3D scenes)!! And yes, it actually works!
Check out our new work #NeRN
Ron Vainshtein retweeted
ืชืืดื ืืืชืืืืืช ืืขืฆืืื [ืืืืืฆืืงืช] ืกืืื Dall-E 2 ืฆืฆื ืื ืฉื ื ืดืืชืืจืืืด. ืืื ืืืืืจื ืืื ืืืื ืงืื ืคืชืื, ืืฉื Midjourney ืฉืืื ืืชืืืื ืืืกืืช ืืืืื ืืขืฆืื - ืืืฉื ื ืืืืงืืคื ืืื Stable Diffusion.
ืืื ืืื ืืืืื?
>>
Ron Vainshtein retweeted
Super excited to share our new work! On a quest to find good regularization for Meta RL with @AvivTamar1 and Gilad Adler.
Paper:ย arxiv.org/abs/2206.10716
Code:ย github.com/zoharri/MBRL2
[1/7]
Ron Vainshtein retweeted
The community should focus more on positive ways to encourage reviewers instead of finding new ways to scare them away ๐
Ron Vainshtein retweeted
Do brains build generative models all the way down to pixel level? I happened to get woken up this morning just as I was scrutinizing a visual detail in the dream, which gave me a strong sense that it does. Previously I've been less sure. Anyone else try to debug?
Ron Vainshtein retweeted
The elegance of ML is the elegance of biology, not the elegance of math or physics.
Simple gradient descent creates mind-boggling structure and behavior, just as evolution creates the awe inspiring complexity of nature.
Ron Vainshtein retweeted
What would a peer review process look like if it was designed today? ๐ค
We strongly believe that peer review can be so much better for both authors and reviewers and we're super excited to share @peerxiv, our vision for a fresh view on peer review๐ก
peerxiv.web.app/about
























