Molecular biology 🧬 wet lab iteration for cell therapies implies intense R&D resources. By leveraging AI, computational biology algorithms and multi-omics assay data @StammBio presents MoNA: a cell representation atlas designed to accelerate bio-innovation cycles. Take a look👇

Is this a circular RNA reference ? 😅

AI ppl: we basically solved bio with the thinking sands Bio ppl: after decades of lab toil, we solved how to stably express GFP in stem cells

We’re moving biology from the wet lab to the gpu cluster.

I find it fascinating that momentum in standard convex optimization is just about making convergence faster, but in nonconvex problems, it's sometimes the only way a method can work at all. Just saw a new example of this phenomenon in the case of difference-of-convex functions.

Graphons are mathematical objects that model the structure of massive networks. In machine learning, they provide a powerful framework for analyzing and generating large graphs. They are used to estimate the underlying structure of a network, predict missing links, and understand

Theorem: The maximum possible duration of the computational singularity is 470 years. Proof: The FLOPs capacity of all computers which existed in the year 1986 is estimated to be at most 4.5e14 (Hilbert et al. 2011). Based on public Nvidia revenue and GPU specs, this capacity

How does neural feature geometry evolve during training? Modeling feature spaces as geometric graphs, we show that nonlinear activations drive transformations resembling discrete Ricci flow - revealing how class structure emerges and suggesting geometry-informed training

The simplex method is an algorithm that turns an optimization problem, like setting up an investment portfolio, into a geometry problem. Recently, the scientists Sophie Huiberts (left) and Eleon Bach reduced the runtime of the simplex method. https://t.co/52maQuGbA4

Do more math.

🚨 We wrote a new AI textbook "Learning Deep Representations of Data Distributions"! TL;DR: We develop principles for representation learning in large scale deep neural networks, show that they underpin existing methods, and build new principled methods.

Two sources, two frequencies #mathart

A 7 million parameter model from Samsung just outperformed DeepSeek-R1, Gemini 2.5 Pro, and o3-mini on reasoning benchmarks like ARC-AGI. Let that sink in. It’s 10,000x smaller yet smarter. The secret is recursion. Instead of brute-forcing answers like giant LLMs, it drafts a

*The Origins of Representation Manifolds in LLMs* by Modell et al. They study the presence of "interpretable features" in LLMs embedded as manifolds and how their geometry connects to the internal representations of the models. https://t.co/htskNJCllc

This new research showed empirically that KL divergence is the most accurate predictor of catastrophic forgetting in LLMs! And since forgetting tracks KL drift from the base, methods that keep KL small tend to forget less too.

HOW INFORMATION FLOWS THROUGH TRANSFORMERS Because I've looked at those "transformers explained" pages and they really suck at explaining. There are two distinct information highways in the transformer architecture: - The residual stream (black arrows): Flows vertically through

💡 Me encantaron las sesiones de pósters en el Simposio Científico de IA y Aplicaciones #SCIAA2025! Había muchos trabajos interesantes. En particular quiero destacar dos proyectos que me llamaron la atención: 🔹 Impacto del Prompt Engineering en el rendimiento de LLMs: estudio

Que capo

The most complex biological system that we meaningfully understand is a single virion. Everything more complex than that (including bacteria, single human cells, animals, and of course, humans) is totally beyond our comprehension.

Llego el dia de decidir entre: - Bosch - Makita - DeWalt - Black & Decker

Saying that deep learning is "just a bunch of matrix multiplications" is about as informative as saying that computers are "just a bunch of transistors" or that a library is "just a lot of paper and ink." It's true, but the encoding substrate is the least important part here.