arXiv subjects

Enric Boix-Adserà

Publications and source records attributed to Enric Boix-Adserà.

7 recordsLinked to original sources

arXiv2026

MoRE: Scaling mixture of experts with hardware-aware low-rank routing

Mixture-of-Experts (MoE) layers are central to frontier language models, and recent architectures push toward more and smaller experts. In this regime, the standard linear router becomes a bottleneck: with $M$ experts and hidden dimension $h$, its per-token cost $Θ(Mh)$ dominates the MoE layer once $M$ is large. We introduce MoRE (Mixture of Rank-reduced-routed Experts), which factorizes the router weight matrix at rank $r$ and reduces the routing cost to $O((h + M)r)$. We prove that rank logarithmic in $M$ suffices for routing expressivity when the number of active experts is fixed, and is necessary up to precision factors. We also prove that logarithmic rank preserves load balance in a Gaussian memorization model, and training on a synthetic phonebook task shows that low rank does not hurt memorization. At matched active FLOPs, the factorization allows a factor of $Θ(h/r)$ more experts. To realize this gain in wall-clock time, we design a fused Triton kernel at inference that avoids expensive memory operations on HBM. Empirically, MoRE improves memorization on the phonebook task and performance on knowledge-intensive Q\&A benchmarks after pretraining, while matching reasoning ability. Code available at https://github.com/Matheart/MoRE_code.

cs.LG↗

arXiv2026

There Will Be a Scientific Theory of Deep Learning

In this paper, we make the case that a scientific theory of deep learning is emerging. By this we mean a theory which characterizes important properties and statistics of the training process, hidden representations, final weights, and performance of neural networks. We pull together major strands of ongoing research in deep learning theory and identify five growing bodies of work that point toward such a theory: (a) solvable idealized settings that provide intuition for learning dynamics in realistic systems; (b) tractable limits that reveal insights into fundamental learning phenomena; (c) simple mathematical laws that capture important macroscopic observables; (d) theories of hyperparameters that disentangle them from the rest of the training process, leaving simpler systems behind; and (e) universal behaviors shared across systems and settings which clarify which phenomena call for explanation. Taken together, these bodies of work share certain broad traits: they are concerned with the dynamics of the training process; they primarily seek to describe coarse aggregate statistics; and they emphasize falsifiable quantitative predictions. We argue that the emerging theory is best thought of as a mechanics of the learning process, and suggest the name learning mechanics. We discuss the relationship between this mechanics perspective and other approaches for building a theory of deep learning, including the statistical and information-theoretic perspectives. In particular, we anticipate a symbiotic relationship between learning mechanics and mechanistic interpretability. We also review and address common arguments that fundamental theory will not be possible or is not important. We conclude with a portrait of important open directions in learning mechanics and advice for beginners. We host further introductory materials, perspectives, and open questions at learningmechanics.pub.

stat.ML↗

arXiv2025

Toward universal steering and monitoring of AI models

Modern AI models contain much of human knowledge, yet understanding of their internal representation of this knowledge remains elusive. Characterizing the structure and properties of this representation will lead to improvements in model capabilities and development of effective safeguards. Building on recent advances in feature learning, we develop an effective, scalable approach for extracting linear representations of general concepts in large-scale AI models (language models, vision-language models, and reasoning models). We show how these representations enable model steering, through which we expose vulnerabilities, mitigate misaligned behaviors, and improve model capabilities. Additionally, we demonstrate that concept representations are remarkably transferable across human languages and combinable to enable multi-concept steering. Through quantitative analysis across hundreds of concepts, we find that newer, larger models are more steerable and steering can improve model capabilities beyond standard prompting. We show how concept representations are effective for monitoring misaligned content (hallucinations, toxic content). We demonstrate that predictive models built using concept representations are more accurate for monitoring misaligned content than using models that judge outputs directly. Together, our results illustrate the power of using internal representations to map the knowledge in AI models, advance AI safety, and improve model capabilities.

cs.CL↗

arXiv2021

The Average-Case Complexity of Counting Cliques in Erdos-Renyi Hypergraphs

We consider the problem of counting $k$-cliques in $s$-uniform Erdos-Renyi hypergraphs $G(n,c,s)$ with edge density $c$, and show that its fine-grained average-case complexity can be based on its worst-case complexity. We prove the following: 1. Dense Erdos-Renyi graphs and hypergraphs: Counting $k$-cliques on $G(n,c,s)$ with $k$ and $c$ constant matches its worst-case time complexity up to a $\mathrm{polylog}(n)$ factor. Assuming randomized ETH, it takes $n^{Ω(k)}$ time to count $k$-cliques in $G(n,c,s)$ if $k$ and $c$ are constant. 2. Sparse Erdos-Renyi graphs and hypergraphs: When $c = Θ(n^{-α})$, we give several algorithms exploiting the sparsity of $G(n, c, s)$ that are faster than the best known worst-case algorithms. Complementing this, based on a fine-grained worst-case assumption, our results imply a different average-case phase diagram for each fixed $α$ depicting a tradeoff between a runtime lower bound and $k$. Surprisingly, in the hypergraph case ($s \ge 3$), these lower bounds are tight against our algorithms exactly when $c$ is above the Erdős-Rényi $k$-clique percolation threshold. This is the first worst-case-to-average-case hardness reduction for a problem on Erdős-Rényi hypergraphs that we are aware of. We also give a variant of our result for computing the parity of the $k$-clique count that tolerates higher error probability.

cs.CC↗

arXiv2021

The Multiplayer Colonel Blotto Game

We initiate the study of the natural multiplayer generalization of the classic continuous Colonel Blotto game. The two-player Blotto game, introduced by Borel as a model of resource competition across $n$ simultaneous fronts, has been studied extensively for a century and seen numerous applications throughout the social sciences. Our work defines the multiplayer Colonel Blotto game and derives Nash equilibria for various settings of $k$ (number of players) and $n$. We also introduce a "Boolean" version of Blotto that becomes interesting in the multiplayer setting. The main technical difficulty of our work, as in the two-player theoretical literature, is the challenge of coupling various marginal distributions into a joint distribution satisfying a strict sum constraint. In contrast to previous works in the continuous setting, we derive our couplings algorithmically in the form of efficient sampling algorithms.

cs.GT↗

arXiv2019

Approximating the Determinant of Well-Conditioned Matrices by Shallow Circuits

The determinant can be computed by classical circuits of depth $O(\log^2 n)$, and therefore it can also be computed in classical space $O(\log^2 n)$. Recent progress by Ta-Shma [Ta13] implies a method to approximate the determinant of Hermitian matrices with condition number $κ$ in quantum space $O(\log n + \log κ)$. However, it is not known how to perform the task in less than $O(\log^2 n)$ space using classical resources only. In this work, we show that the condition number of a matrix implies an upper bound on the depth complexity (and therefore also on the space complexity) for this task: the determinant of Hermitian matrices with condition number $κ$ can be approximated to inverse polynomial relative error with classical circuits of depth $\tilde O(\log n \cdot \log κ)$, and in particular one can approximate the determinant for sufficiently well-conditioned matrices in depth $\tilde{O}(\log n)$. Our algorithm combines Barvinok's recent complex-analytic approach for approximating combinatorial counting problems [Bar16] with the Valiant-Berkowitz-Skyum-Rackoff depth-reduction theorem for low-degree arithmetic circuits [Val83].

cs.DS↗

arXiv2019

Subadditivity Beyond Trees and the Chi-Squared Mutual Information

In 2000, Evans et al. [Eva+00] proved the subadditivity of the mutual information in the broadcasting on tree model with binary vertex labels and symmetric channels. They raised the question of whether such subadditivity extends to loopy graphs in some appropriate way. We recently proposed such an extension that applies to general graphs and binary vertex labels [AB18], using synchronization models and relying on percolation bounds. This extension requires however the edge channels to be symmetric on the product of the adjacent spins. A more general version of such a percolation bound that applies to asymmetric channels is also obtained in [PW18], relying on the SDPI, but the subadditivity property does not follow with such generalizations. In this note, we provide a new result showing that the subadditivity property still holds for arbitrary (asymmetric) channels acting on the product of spins, when the graphs are restricted to be series-parallel. The proof relies on the use of the Chi-squared mutual information rather than the classical mutual information, and various properties of the former are discussed. We also present a generalization of the broadcasting on tree model (the synchronization on tree) where the bound from [PW18] relying on the SPDI can be significantly looser than the bound resulting from the Chi-squared subadditivity property presented here.

math.PR↗