Tootfinder

Maximally Useful and Minimally Redundant: The Key to Self Supervised Learning for Imbalanced Data
Yash Kumar Sharma, Vineet Nair, Wilson Naik
https://arxiv.org/abs/2509.08469 ht…

Maximally Useful and Minimally Redundant: The Key to Self Supervised Learning for Imbalanced Data
The robustness of contrastive self-supervised learning (CSSL) for imbalanced datasets is largely unexplored. CSSL usually makes use of \emph{multi-view} assumptions to learn discriminatory features via similar and dissimilar data samples. CSSL works well on balanced datasets, but does not generalize well for imbalanced datasets. In a very recent paper, as part of future work, Yann LeCun pointed out that the self-supervised multiview framework can be extended to cases involving \emph{more than t…

At #burningman this year, I brought a small 1200 lumen projector that I could mount on my bike. Would it be feasible to do some quick on the fly projection mapping? It seems that yes, for somewhat small (about 6m x 3m max for reasonable brightness) and not very bright pieces, it works.
I may experiment with this more in the future.

Imagine ChatGPT but instead of predicting text it just linked you to the to 3 documents most-influential on the probabilities that would have been used to predict that text.
Could even generate some info about which parts of each would have been combined how.
There would still be issues with how training data is sourced and filtered, but these could be solved by crawling normally respecting robots.txt and by paying filterers a fair wage with a more relaxed work schedule and mental health support.
The energy issues are mainly about wild future investment and wasteful query spam, not optimized present-day per-query usage.
Is this "just search?"
Yes, but it would have some advantages for a lot of use cases, mainly in synthesizing results across multiple documents and in leveraging a language model more fully to find relevant stuff.
When we talk about the harms of current corporate LLMs, the opportunity cost of NOT building things like this is part of that.
The equivalent for art would have been so amazing too! "Here are some artists that can do what you want, with examples pulled from their portfolios."
It would be a really cool coding assistant that I'd actually encourage my students to use (with some guidelines).
#AI #GenAI #LLMs

Towards Adaptive ML Benchmarks: Web-Agent-Driven Construction, Domain Expansion, and Metric Optimization
Hangyi Jia, Yuxi Qian, Hanwen Tong, Xinhui Wu, Lin Chen, Feng Wei
https://arxiv.org/abs/2509.09321

Towards Adaptive ML Benchmarks: Web-Agent-Driven Construction, Domain Expansion, and Metric Optimization
Recent advances in large language models (LLMs) have enabled the emergence of general-purpose agents for automating end-to-end machine learning (ML) workflows, including data analysis, feature engineering, model training, and competition solving. However, existing benchmarks remain limited in task coverage, domain diversity, difficulty modeling, and evaluation rigor, failing to capture the full capabilities of such agents in realistic settings. We present TAM Bench, a diverse, realistic, and st…

So yes, BlueSky finally has bookmarks.
Before this feature came out, people had simply been replying to posts with a 📌 emoji and there was a dedicated feed where you could view your 📌’ed posts.
In an example of how to roll stuff out, someone made an importing tool that either moved OR copied your 📌 posts to bookmarks (saved posts)
VERY well done.

Bluesky (@bsky.app)
v1.108 is rolling out today 🚚 Now live, at long last: Bookmarks, aka Saved Posts. For all those posts you'll definitely plan to come back to! Update the app and give it a try. The button is right down there 👇

Thinking about #DiscoElysium succesors(?), including ones with heavy ethical concerns. A lot of them have gone for female protagonists: Good! There could have been a couple more leading women in DE. But the reveal trailers are mainly voiced by men... men berating or mocking the central character.
Longdue's Hopetown has a father figure voice over, berating the protagonist: "Who do you think you are? [...] You don't understand the forces at play" Infantalised, she doesn't get to respond.
ZA/UM's Zero Parades VO is also a man, initially flattering the spy protagonist "brilliant, relentless" ... then saying she came after him and "everything she touched, written into failure". Again, no chance for the female spy to tell her side of the story ... And there's an objectifying/threatening section about how her fair hair is like a dandelion. Tangeant: what's with the heavy/bad pseudo-German VO accent?
Do I bother mentioning DarkMath's XXX Nightshift? Yes, for once a woman's VO, but (as you'll have guessed from the name) a porn-like, male-gaze view of what a woman would say and think.
So, don't be fooled that a female protagonist means we're actually getting a woman's POV. It's just a cosmetic change. Dora Klindžić's https://bsky.app/profile/did:plc:rx3zlqzbjylvtyg6siikaouq experiences as a writer at ZA/UM show how women struggle to have a voice and stay employed in these types of studios. Many of the senior men she mentions are still at ZA/UM.

VER: Vision Expert Transformer for Robot Learning via Foundation Distillation and Dynamic Routing
Yixiao Wang, Mingxiao Huo, Zhixuan Liang, Yushi Du, Lingfeng Sun, Haotian Lin, Jinghuan Shang, Chensheng Peng, Mohit Bansal, Mingyu Ding, Masayoshi Tomizuka
https://arxiv.org/abs/2510.05213

VER: Vision Expert Transformer for Robot Learning via Foundation Distillation and Dynamic Routing
Pretrained vision foundation models (VFMs) advance robotic learning via rich visual representations, yet individual VFMs typically excel only in specific domains, limiting generality across tasks. Distilling multiple VFMs into a unified representation for policy can mitigate this limitation but often yields inflexible task-specific feature selection and requires costly full re-training to incorporate robot-domain knowledge. We propose VER, a Vision Expert transformer for Robot learning. During …

AUREXA-SE: Audio-Visual Unified Representation Exchange Architecture with Cross-Attention and Squeezeformer for Speech Enhancement
M. Sajid, Deepanshu Gupta, Yash Modi, Sanskriti Jain, Harshith Jai Surya Ganji, A. Rahaman, Harshvardhan Choudhary, Nasir Saleem, Amir Hussain, M. Tanveer
https://arxiv.org/abs/2510.05295

AUREXA-SE: Audio-Visual Unified Representation Exchange Architecture with Cross-Attention and Squeezeformer for Speech Enhancement
In this paper, we propose AUREXA-SE (Audio-Visual Unified Representation Exchange Architecture with Cross-Attention and Squeezeformer for Speech Enhancement), a progressive bimodal framework tailored for audio-visual speech enhancement (AVSE). AUREXA-SE jointly leverages raw audio waveforms and visual cues by employing a U-Net-based 1D convolutional encoder for audio and a Swin Transformer V2 for efficient and expressive visual feature extraction. Central to the architecture is a novel bidirect…

Exploring Scalar Leptoquarks at Muon Collider via Indirect Signatures and Right-Handed Neutrino-Assisted Decays
Subham Saha, Arvind Bhaskar, P. S. Bhupal Dev, Manimala Mitra
https://arxiv.org/abs/2509.04579

Exploring Scalar Leptoquarks at Muon Collider via Indirect Signatures and Right-Handed Neutrino-Assisted Decays
We explore the discovery prospects of the scalar leptoquark doublet $\widetilde{R}_{2}({3},{2},1/6)$ at a future high-energy muon collider, considering both direct and indirect search strategies. For indirect search, we probe the relevant Yukawa couplings by analyzing the dijet final state via virtual exchange of the leptoquark. Even with a sub-$\mathcal{O}(1)$ Yukawa coupling, we achieve a 5$σ$ sensitivity up to leptoquark mass $ \sim 4$ TeV ($7$ TeV) at $\sqrt{s}=5, (10)$ TeV center-of-mass …