Tomás Fernandes

computer scientist and software engineer.
interested in swe, geometric deep learning, distributed systems enjoyer

[Tomás Fernandes @tomasff02]

@karpathy FastMail is a solid alternative to Proton and has integrations with 1Password/Bitwarden Masked Emails

[Daniel Johnson @_ddjohnson]

By popular demand, the Treescope pretty-printer from the Penzai neural net library can now be installed separately, and supports both JAX and PyTorch!

And that's not all: Penzai itself now has less boilerplate and includes more pretrained Transformer models!

[Karolis Jucys (@ICML 2026) @Karolis_Ram]

Breaking: @OpenAI’s Minecraft agent VPT kills villagers!

In other news: our paper on dissecting VPT with various interpretability techniques was accepted at the Mechanistic Interpretability workshop at ICML.
1/7

[Sasha Rush @srush_nlp]

A Tour of Differentiable Rasterization (https://t.co/SSdRzWRNNu). Tutorial about writing an SVG-lite in Jax.

This is only loosely about graphics, more about how I stil want a "Differentiable Everything" future 😃.

[Two Minute Papers @twominutepapers]

AlphaFold 3 is here! I was super lucky to have had access to it for a bit now, and it is absolutely amazing. This might be the very first video out there that you see on it - enjoy! https://t.co/EctsUOiBmP #alphafold

[Accepted papers at TMLR @TmlrPub]

Inference from Real-World Sparse Measurements

Arnaud Pannatier, Kyle Matoba, François Fleuret.

Action editor: Ekin Cubuk.

https://t.co/bCJRonrg3F

#encoder #attention #forecasting

[Anton Tsitsulin @graph_]

𝐋𝐞𝐭 𝐘𝐨𝐮𝐫 𝐆𝐫𝐚𝐩𝐡 𝐃𝐨 𝐭𝐡𝐞 𝐓𝐚𝐥𝐤𝐢𝐧𝐠: 𝐄𝐧𝐜𝐨𝐝𝐢𝐧𝐠 𝐒𝐭𝐫𝐮𝐜𝐭𝐮𝐫𝐞𝐝 𝐃𝐚𝐭𝐚 𝐟𝐨𝐫 𝐋𝐋𝐌𝐬

Don't know what to do with your graphs in 2024? Shove them to an LLM, of course, and let LLM figure out what to do!
https://t.co/lqYpOqeXha
Short thread (1/5):

[Bo Wang @BoWang87]

State Spaces models (SSM) such as mamba have started to revolutionize languages, vision, and genomics, as a promising alternative to transformer architecture. How about graph data?

❤️���🔥 Introducing Graph-Mamba, our latest innovation in Graph Learning for enhanced long-range graph context modeling!

🐍 Graph-Mamba: Towards Long-Range Graph Sequence Modeling with Selective State Spaces

AriXv: https://t.co/aNqA8vx1es
Code: https://t.co/qDKuaHM1R4

TL,DR 👇:
Graph Attention mechanisms face challenges in scaling for large graphs. Enter Graph-Mamba – integrating state space models with input-dependent node selection for efficient long-range context reasoning. Featuring special graph-centric adaptations to effectively employ SSM for non-sequential graph data.

🔑Key Contributions:
--Innovative Design: Graph-Mamba pioneers a novel graph network integrating selective SSM, capturing long-range dependencies with adaptive node selection.
--Adaptation for Graphs: Employing elegant node prioritization strategies and permutation-based training to mitigate sequence-induced biases and boost modeling power.
--Performance & Efficiency: Graph-Mamba outperforms baselines with linear-time complexity and up to 74% reduction in GPU memory consumption on large graphs.

❤️‍🔥 Looking Ahead: Propelling SSMs into an era of efficient pre-training on graph data with Graph-Mamba!
#GraphMamba #SSM #GraphTransformers #AIResearch #DataScience

Shoutout to Chloe Wang (@ChloeXWang1 ) for her leadership in this project and also to Oleksii Tsepa(@AlexTsepa ) and Jun Ma (@JunMa_11 ) for their invaluable contributions!

@UHNAIHUB @pmcc_ai @UHN @VectorInst @UofTCompSci @UofT_LMP

[Lenka Zdeborova @zdeborova]

Emergence in LLMs is a mystery. Emergence in physics is linked to phase transitions. We identify a phase transition between semantic and positional learning in a toy model of dot-product attention. Very excited about this one! https://t.co/ALb9D8YdfP

[Bruno Neri @neribr]

"Future Directions in Foundations of Graph Machine Learning" by @HaggaiMaron ,@ismaililkanc, @ffabffrasca, @dereklim_lzh, @mmbronstein et al.

Paper: https://t.co/mhgiWxGNlV

#graphneuralnetworks

[Portal @_portal_]

🧬 You may know gene language models, but what about cell language models?

CellPLM effectively encodes inter-cell relations, integrates spatially-resolved transcriptomic data, and applies a well-justified prior distribution.

New blog by @bilibiliWehos:
https://t.co/GpPVsecDVp

[AK @_akhaliq]

Apple presents AIM

Scalable Pre-training of Large Autoregressive Image Models

paper page: https://t.co/2SkcSumeKy

paper introduces AIM, a collection of vision models pre-trained with an autoregressive objective. These models are inspired by their textual counterparts, i.e., Large Language Models (LLMs), and exhibit similar scaling properties. Specifically, we highlight two key findings: (1) the performance of the visual features scale with both the model capacity and the quantity of data, (2) the value of the objective function correlates with the performance of the model on downstream tasks. We illustrate the practical implication of these findings by pre-training a 7 billion parameter AIM on 2 billion images, that achieves 84.0% on ImageNet-1k with a frozen trunk. Interestingly, even at this scale, we observe no sign of saturation in performance, suggesting that AIM potentially represents a new frontier for training large-scale vision models. The pre-training of AIM is similar to the pre-training of LLMs, and does not require any image-specific strategy to stabilize the training at scale.

[François Fleuret @francoisfleuret]

Recap of the non-NLP experiments with my GPT (not the RNN one), code on

https://t.co/mcmFE58NU6

Manipulate elements on a grid: https://t.co/xjw1QxVceQ

Simulate stacks: https://t.co/BQ78BBvjGM

Find a path in a maze: https://t.co/j7Z6r6TS7i

1/2

[Google DeepMind @GoogleDeepMind]

Today with @YouTube, we’re announcing Lyria: our most advanced music generation model to date. 🎶

We’re also releasing 2️⃣ AI experiments in close collaboration with participating artists and creators to bring their ideas to life responsibly. →

https://t.co/i9ve66A5rv

[Google AI @GoogleAI]

Introducing Mirasol, a multimodal model for learning across audio, video, & text that decouples the modeling into separate autoregressive models to process the inputs according to the characteristics of their modalities, for state-of-the-art performance →https://t.co/PjFHFnSyvl

[Mehdi Azabou @ NeurIPS @mehdiazabou]

Is a universal brain decoder possible? Can we train a decoding system that easily transfers to new individuals/tasks?

Check out our #NeurIPS2023 paper where we show that it’s possible to transfer from a large pretrained model to achieve SOTA 🧠!

Link: https://t.co/0Iebjpt4TM 🧵

[Andrea Cini @andreacini1994]

📢 Happy to finally share our paper on graph deep learning for time series forecasting!

This puts together what we've learned in the past few years using GNNs for TS processing, I hope you'll find it useful😃

W/ @IvanMarisca, @dan_zambon and Cesare 🔥

🔗https://t.co/ypeovJoRLI

[Jean-Rémi King @JeanRemiKing]

We’re very happy to share our latest study:

‘Brain decoding: toward real-time reconstruction of visual perception’

led by @BenchetritYoha1 & @HubertBanville

- paper: https://t.co/HaqClxPiLt
- blog: https://t.co/py2AjqDlmQ
- summary: ⬇️

[Bahare Fatemi @BahareFatemi]

New paper alert 🚨: "Talk Like a Graph" studies the problem of encoding graph-structured data, paving the way for AI to better understand and process complex relationships. Surprising results below 🤯! #AI #LLM #GraphData #MachineLearning

[Dan Hendrycks @hendrycks]

AI models are not just black boxes or giant inscrutable matrices.

We discover they have interpretable internal representations, and we control these to influence hallucinations, bias, harmfulness, and whether a LLM lies.

🌐: https://t.co/6hU83vaYUk
📄: https://t.co/xyrzgfUOfn