// intelligence · research
AI Research Papers
Preprints and publications from arXiv, Berkeley AI Research, MIT, Google Research, and leading academic institutions worldwide.
500 papers · showing 20
Study reveals data centers’ long-distance damage to air quality
Researchers find air pollution from data center electricity use can affect communities far from the facilities, with some of the highest impacts in the Midwest.
NOTREKS: Marginal Independence in Score-based Structure Learning
arXiv:2611.24015v1 Announce Type: new Abstract: Score-based structure learning estimates causal graphs by optimizing a data-dependent objective over structural equation models. We study how prior knowledge that selected pairs of variables are marginally independent can be incorporated into such causal discovery methods. Under faithfulness, marginally independent pairs cannot be connected by a trek, so the knowledge becomes a graphical no-trek constraint, which we call NOTREKS information. We introduce a power-series trek (PSTrek) penalty for gradient-based optimization and a trek constraint for order-based search, and we analyze why such information helps in finite samples and how it restricts the search space. Our numerical experiments illustrate to what extent NOTREKS information helps in causal discovery. In particular, using synthetic benchmarks with oracle NOTREKS information, we show that these constraints can improve both structural recovery and adjustment-based effect identification.
How Reward Gain Imbalance Shapes Ranking: Rethinking Rubric-Based Reward Aggregation
arXiv:2611.24005v1 Announce Type: new Abstract: Rubric-based reinforcement learning constructs rewards for aligning large language models with human preferences, typically by equally aggregating continuous scores that a judge assigns to individual criteria. However, scoring heterogeneity across criteria can distort candidate rankings. Existing work has focused on improving judges, learning criterion weights, or refining aggregation, whereas the interaction between scoring heterogeneity and ranking remains insufficiently understood. We formalize this heterogeneity as \emph{reward gain imbalance}: each criterion enters the aggregated reward with its own reward gain, the gap between its expected scores on satisfied and unsatisfied responses. Counterintuitively, our experiments show that removing this imbalance, even with ground-truth labels, lowers ranking accuracy. To explain this, we derive an exact decomposition of the reward difference between any two candidates into a base satisfaction signal, a reward gain imbalance term, and a residual term, showing that reward gain imbalance can attenuate or amplify the base signal, and finding that misrankings usually concentrate on pairs where it attenuates. Guided by this analysis, we propose the adaptive \textbf{D}ual-\textbf{C}enter \textbf{C}lipping aggregation (DCC), which bounds each criterion's contribution to the reward difference instead of simply removing the imbalance. Using only the scores of the current candidate group, it determines a clipping interval that imposes a common upper bound on per-criterion score differences between candidates, limiting the excessive influence of extreme scores while preserving continuous judgment information....
ReliFlow: Reliability-Calibrated Residual Transport for Combinatorial Single-Cell Perturbation Prediction
arXiv:2611.23999v1 Announce Type: new Abstract: Predicting the transcriptional response to combined perturbations requires distinguishing interactions between interventions from errors in their individual effect estimates. The latter become more pronounced when single-perturbation evidence is sparse or withheld. We introduce a conditional flow-matching model that estimates the reliability of single-gene effects from training-only cross-fitting and uses those estimates to calibrate the additive source distribution. The velocity is decomposed into a known correction of the additive reference and a learned residual transport field. On Norman genetic perturbations, reduces mean absolute error (MAE) from 0.142 to 0.131 for unseen combinations of observed singles and from 0.198 to 0.178 when constituent singles are withheld. On held-out ComboSciPlex drug combinations, MAE decreases from 0.191 to 0.176. Non-additive residual correlation improves from 0.574 to 0.638, and component ablations identify the contribution of reliability weighting. The results indicate that accounting for uncertain constituent effects improves generalization to unseen genetic and chemical combinations, with a moderate increase in training cost.
Optimal Deep Nonparametric Regression with Uniformly Bounded Lipschitz Constants
arXiv:2611.23979v1 Announce Type: new Abstract: Stability under input perturbations is necessary for deep neural network estimators in many applications involving noisy covariates or estimated features. Weight norm constraints are widely used to control sensitivity to input perturbations by bounding the Lipschitz constant of neural networks. However, existing approximation results under fixed bounds on standard norm products are insufficient to establish optimal statistical rates. In this paper, we establish a sharp approximation theory for deep ReLU networks under a split spectral norm product constraint. This theoretical result shows that sufficiently smooth H\"{o}lder functions can be uniformly approximated to arbitrary accuracy as the network depth increases, with the norm-product bound held fixed. We then propose a penalized least squares estimator for deep nonparametric regression. The resulting estimator has a uniformly bounded Lipschitz constant while still achieving the optimal nonparametric rate up to a logarithmic factor. As applications, we establish optimal prediction rates for two-stage learning and provide robustness guarantees under adversarial input perturbations. Numerical studies show that the penalized estimator has well-controlled sensitivity to input perturbations and prediction accuracy comparable to that of classical deep network methods.
Improved Leverage Score Sampling for Constrained Active Linear Regression
arXiv:2611.23801v1 Announce Type: new Abstract: In this paper we study active learning for constrained linear regression. Given a design matrix $A\in\mathbb{R}^{n\times d}$, a constraint set $C\subseteq\mathbb{R}^d$, and query access to $b\in\mathbb{R}^n$, we seek an approximate solution to $\min_{x\in C}\|Ax-b\|_2^2$ using as few queries as possible. This problem arises in many scientific and engineering applications, and randomized leverage score sampling provides a powerful method to solve it. However, existing leverage score methods ignore the constraint or encode it through ridge regularization, suffering unfavorable query complexity for anisotropic constraint sets. We propose ellipsoid-regularized leverage scores, where we approximate the constraint set by an outer ellipsoid and use its shape matrix to bias sampling toward directions that matter under the constraint geometry. We prove that the query complexity is governed by an anisotropic effective dimension $d_{\text{eff}}(C)\le d$. With appropriate parameters, $O(d_{\text{eff}}(C)\log d/\varepsilon^2)$ queries suffice to return a point in $C$ with objective value at most $(1+\varepsilon)$ times the optimum. This bound can substantially improve on the standard leverage-score bound, since anisotropic constraints can make $d_{\text{eff}}(C)$ much smaller than $d$. For convex constraints, we improve the dependence on $\varepsilon$ from $1/\varepsilon^2$ to $1/\varepsilon$. We also prove two lower bounds: first, the $d/\varepsilon^2$ dependence is unavoidable without convexity assumptions on $C$, and second, in some regimes, the effective dimension dependence $d_{\mathrm{eff}}(C)/\varepsilon$ is necessary for ellipsoidal constraints. Experiments on synthetic regression problems and a parametric heat conduction problem show lower regression error than standard baselines for similar query budget.
A Free Lunch: Gradient Averaging Achieves Optimal Statistical Efficiency in Newton Sketch
arXiv:2611.23671v1 Announce Type: new Abstract: We study statistical inference for online sketched Newton methods, in which each Newton system is approximately solved by an inner randomized sketching solver. Existing work reveals a computational--statistical trade-off: the sketching solver (and its accelerated variants) reduces the per-step cost of Newton methods from $O(d^3)$ to $O(d^2)$, but the randomness introduced by sketching also inflates the limiting covariance of both the last and the averaged Newton iterates. In this work, we address this trade-off via gradient averaging. We show that simply replacing the stochastic gradient with a moving average at each step can mitigate, and in some cases eliminate, the trade-off in Newton sketch. Importantly, gradient averaging requires no additional oracle evaluations and leaves the contraction bound of the inner sketching solver unchanged. In particular, we prove that the last sketched Newton iterate with gradient averaging attains a limiting covariance no larger than that without gradient averaging, while the averaged iterate even attains the minimax optimal limiting covariance, matching that of the exact Newton method. To our knowledge, this is the first analysis showing that randomness introduced to improve computational efficiency need not compromise statistical efficiency as long as proper gradient estimates are used. The key analytical challenge is that gradient averaging destroys the crucial martingale difference structure underlying central limit theorem arguments. We overcome this challenge by analyzing the coupled two-timescale recursions governing the iterate and the gradient estimator. Numerical experiments on linear and logistic regression support our theoretical covariance comparisons.
Bounded-Order Non-Identifiability in Latent Structure Block Models: The Fiber in Closed Form
arXiv:2611.23547v1 Announce Type: new Abstract: A graphon is determined by all of its homomorphism densities, but motif-based methods use finitely many. We compute exactly what that truncation costs in a natural family: latent structure block models whose positions lie on $K$ rays, equivalently degree-corrected block models of rank two. For a given radial law, the configurations that agree on every graph with fewer than $2K-1$ vertices form a one-dimensional fiber, which we give in closed form with explicit endpoints; the friendship graph $F_{K-1}$ is its unique edge-minimal separator, so no fixed motif order serves every $K$. The separating signal is suppressed by a factor of at most $m_1^{2(K-1)}$, where $m_1$ is the mean squared slope, about $10^{-4}$ at $K=4$: detecting it from motif counts would take about $10^{15}$ nodes in a single graph, at any edge density. The information is nonetheless in the graph; the fiber belongs to the statistic, not the model. Common spectral pipelines, which count no motifs, still slide along the fiber by more than a third of its length at ordinary densities, with confidence regions that almost never contain the truth, while model-based fits of the same embedding do not. We give practitioners four checks: whether a fit is at risk, what to run instead, why a confidence interval will not warn, and how wide the remaining ambiguity is.
A Differential Theory of Quality-Resource Trade-offs for AI Model Selection: The Multiresource Frontier and Its Minimax Price
arXiv:2611.23324v1 Announce Type: new Abstract: Comparing AI models across resource budgets yields quality-resource frontiers relating error to latency, parameter count, training compute, and energy. Offset power laws summarize such frontiers, but their fitted parameters reveal neither the price-independent geometry nor the manner in which trade-offs extend across several resource channels. This paper addresses both gaps by developing a differential theory in which this structure is made explicit. Each frontier admits a price-independent plane-curve representation relating the effective number of active resource channels, $N_{\mathrm{eff}} = s^2/\kappa$ (squared slope over curvature), to the remaining log-error depth, $\Delta S = \ln(\varepsilon/\varepsilon_{\infty})$. This representation separates intrinsic shape from the scale and origin of the resource axis. Complete monotonicity and divisibility yield differential inequalities and frontier bounds. When each of $N$ resources contributes one active channel, the theory gives the parameter-free operating rule $N_{\mathrm{eff}} = 1$ in delivered units and derives minimax resource prices $\lambda^{\star}_i = \alpha_i/(N+1)$ with an exact worst-case regret of $\ln(N+1)$ nats. The results make it possible to interpret measured frontiers and compare resource costs without hand-chosen exchange rates. The theory is evaluated on 117 published frontiers and implemented in the open-source package nn-rank, which derives prices from measured quality and cost, ranks candidates, and reports supported choices and minimax regret: https://github.com/ABrain-One/nn-rank
Existence of SRB measures for some R\"ossler systems
arXiv:2611.23279v1 Announce Type: new Abstract: We prove that for the R\"ossler system $\dot x=-(y+z)$, $\dot y=x+ay$, $\dot z=b+z(x-c)$ with $a=b=0.2$, the set of parameters $c$ within $6\cdot10^{-5}$ of the classical value $c=5.7$ for which the flow has a H\'enon-like strange attractor supporting a Sinai--Ruelle--Bowen (SRB) measure has positive Lebesgue measure. The key step, proved with computer assistance, is that at a parameter localised in an explicit interval within $5.5\cdot10^{-5}$ of $5.7$ the period-one saddle of the Poincar\'e map has a quadratic homoclinic tangency that unfolds generically; the conclusion then follows from the theorems of Mora--Viana, Benedicks--Young and Benedicks--Viana. The same method gives two further such tangencies near $5.7$ and one at the band-merging parameter $c\approx4.507$. The proof reduces the generic-tangency conditions to finitely many inequalities, through explicit $C^2$ graph-transform bounds for the local invariant manifolds of a Poincar\'e map whose area contraction is of order $10^{-11}$--$10^{-14}$, and a scalar tangency function. The inequalities are checked with interval arithmetic in CAPD and re-checked with an independent implementation based on the Kv library; the analytic tangency criterion is formalised in Lean~4. Whether an SRB measure exists at $c=5.7$ itself remains open.
Riesz and Lusin type theorems in Choquet integral spaces: quasicontinuity and density of continuous functions for monotone measures
arXiv:2611.23277v1 Announce Type: new Abstract: We study the approximation of measurable functions by continuous functions, in measure and with respect to the Choquet, Sugeno and Shilkret integrals, for monotone measures on metric spaces. For every monotone measure, quasicontinuous functions are limits of continuous functions in measure and in the Sugeno integral; for unbounded Choquet integrable functions this remains true under a small-set subadditivity condition or under null-additivity. Conversely, under property [S1] and an outer smallness condition, every such limit is quasicontinuous, and examples show that neither condition can be omitted. This extends to monotone measures a classical description of the closure of continuous functions in capacitary $L^1$-spaces. We characterize the indicator functions in this closure by a symmetric approximation property of sets, and we give criteria, involving regularity and order continuity, for the density of bounded continuous functions. As applications, we show that two approximation theorems of Cao with compactly supported functions fail on spaces which are not locally compact, we prove that, for a given finite, continuous, weakly null-additive measure on a separable metric space, their conclusions for convergence in measure, for the Sugeno integral and for the Choquet integral of bounded functions hold if and only if the measure is concentrated on the set of points having a compact neighbourhood (and, under null-additivity, also for unbounded Choquet integrable functions), and we prove corrected versions on locally compact spaces; we describe the closure explicitly for possibility measures; and we give a Lebesgue-null Cantor dust which shows, for all $n\ge1$ and $0<\delta<n$, that continuous functions are not dense in the Choquet space of the dyadic Hausdorff content.
Cohomology Rings of Fiber Products of Lefschetz Fibrations
arXiv:2611.23272v1 Announce Type: new Abstract: We prove a Lefschetz-type theorem for the cohomology rings of fiber products of two Lefschetz fibrations with disjoint critical loci. As an application, we compute the Hodge numbers of several examples of higher-dimensional Schoen's Calabi-Yau manifolds.
Octahedrally and Icosahedrally Symmetric Solutions of the N-Vortex Problem on the Sphere
arXiv:2611.23270v1 Announce Type: new Abstract: We investigate the octahedrally and icosahedrally symmetric solutions of the N-vortex problem on the sphere, for N = 24 and N = 60 respectively, in the case of equal vorticities. Exploiting the symmetries arising from rotations and permutations of the vortices, the problem is reduced to the analysis of a one-degree-of-freedom Hamiltonian system on the sphere. The original contributions of this article concern the analysis of this reduced system yielding in particular a classification of the equilibrium and collision configurations with the prescribed symmetries. The computational complexity of the analysis is overcome by exploiting the associated discrete symmetries through the invariant theory of reflection groups and Palais' principle of symmetric criticality, and, in some cases, by resorting to numerical approximations.
A counterexample to the Harper-norm silver-mean conjecture
arXiv:2611.23265v1 Announce Type: new Abstract: The silver-mean conjecture formulated in Gerhold and Shalit 2022 states that the norm $E(\alpha)$ of the Harper operator attains its minimum at the silver mean $\alpha = \sqrt{2}-1$. We show that this conjecture is wrong by constructing an explicit rational point with a smaller Harper-norm value: $E(181/437)<2.591046794<E(\sqrt2-1)$. The separating decimal is an exact rational number. All acceptance checks use exact integer arithmetic with rigorously proved coefficient enclosures. No floating-point eigenvalue estimates or external witness data enter the certificate.
A Topologist's Proof of Krasnosel'skii's Bifurcation Theorem
arXiv:2611.23262v1 Announce Type: new Abstract: This note offers a proof of Krasnosel'skii's bifurcation theorem that bypasses the traditional machinery of nonlinear analysis: neither the Leray-Schauder degree nor the Lyapunov-Schmidt reduction is used. The argument rests only on elementary operator theory and straightforward vector-bundle topology, providing a topologist's perspective on a classical result in nonlinear functional analysis.
Spectral dimension reduction of dynamical systems on graphons
arXiv:2611.23238v1 Announce Type: new Abstract: Spectral dimension reduction approximates a dynamical system on a large network by a low-dimensional system whose variables are weighted averages of the node states, with weights given by dominant eigenvectors of the network. Its effectiveness is closely related to the low-rank structure of the interaction matrix. We develop a graphon analogue of this reduction for dynamical systems on dense graph limits. For graphon dynamical systems, we formulate the reduction in terms of the compact self-adjoint integral operator associated with the graphon. This operator-theoretic setting gives a continuum analogue of the spectral/SVD-based framework used for finite networks: the eigenfunctions provide the reduction basis, and the decay of the corresponding singular values quantifies the effect of discarded modes. For finite-rank graphons, we prove exact reduction results for two classes of graphon dynamical systems. We also show that, for finite-rank graphons, the reduced system preserves equilibria, linear stability, and local bifurcations. For infinite-rank graphons, we derive error bounds for the spectral reduction. Numerical examples using finite-rank and infinite-rank graphons illustrate the exact and approximate regimes of the theory.
Hessian--Maslov Geometry of Oja--Brockett Flows
arXiv:2611.23235v1 Announce Type: new Abstract: This paper is organized in three parts. Part I develops the theoretical foundations of the Maslov index for caustics of the Legendre map, covering its semiclassical origin, geometric formulation as an intersection number in the Lagrangian Grassmannian, Hormander's local formula, and the dynamical viewpoint of bifurcating fixed points of a gradient flow. The theory is illustrated on the two-dimensional vector-valued quartic potential, whose caustic is a cusp and whose minima and saddles exchange stability when the eigenvalue ratio crosses 1. Part II lifts the setting to tall n x k matrices and introduces the potential f(X) = -1/2 tr(A^2 X B^2 X^T) + 1/4 tr[(A X B X^T)^2]. Numerical evidence confirms that fixed points of the associated negative gradient flow realize Brockett's sorting theorem. When eigenvalues of A or B degenerate, the fixed-point set expands into a flag manifold; crossing a Weyl wall produces Hessian degeneracy and an exchange of sort order. These phenomena are identified with the Maslov cycle, showing that combinatorial inversion numbers (Coxeter length) and the differential-geometric Maslov index are two faces of the same wall-crossing. Part III supplies the rigorous analysis underlying Part II. The ambient Hessian of the general Oja-Brockett flow dot{X} = A X B - X B X^T A X is completely decomposed into normal, vertical and horizontal sectors, and closed spectral formulas depending on both SPD matrices A and B are derived. In particular, the formula mu^V_{ij} = (lambda_i - lambda_j)(b_i - b_j) reveals a second, dual wall system (the B-wall) that is independent of the classical Weyl wall of A. Under this double-wall structure a genuine eigenvalue crossing becomes an avoided crossing; the associated horizontal and vertical coupling constants are given in closed form.
Biases and Limits of AlphaFold Predictions for Non-Globular Proteins
arXiv:2611.24546v1 Announce Type: new Abstract: AlphaFold has transformed protein structure prediction, making structural models accessible across the life sciences. We explore experimentally validated cases of non-globular proteins to uncover where predictions diverge from biologically relevant structural states and how molecular context shapes these discrepancies, guiding a more informed interpretation of AlphaFold models.
The Shape of Individual Fitness Distributions and the Dynamics of Ecological Communities
arXiv:2611.24047v1 Announce Type: new Abstract: Developing a mechanistic understanding of biogeography and biodiversity that predicts species abundances, distributions, and fluctuations is a central goal of community ecology. We develop a model, that can be seen as an extension of Hubbell's neutral theory, in which a focal species competes with the aggregate of all other species in a local community of fixed size. Each individual has its own competitive ability, and the shape of the distribution of these abilities is incorporated in the model, represented by distributions rather than single species-level values. Under complete mixing, symmetric competitive encounters without ties recover Hubbell's neutral model. We further incorporate immigration from a larger metacommunity and derive a diffusion approximation for the focal species' relative abundance. The model shows that competitive ability distributions with the same mean and variance can nevertheless yield different patterns of local abundance. Thus, species that appear equivalent when summarized by trait statistics need not be ecologically equivalent. Our framework links individual-level variability to population-level persistence and abundance and provides a way to examine how competition and immigration jointly shape ecological outcomes.
A Dual-SOM Learner Model of Instructional Efficiency within Cognitive Load Theory
arXiv:2611.23795v1 Announce Type: new Abstract: This paper presents a novel methodological framework, integrating a dual self-organising map (SOM) neural network architecture with Cognitive Load Theory (CLT), for assessing instructional efficiency within learning environments. By representing curriculum content as a directed acyclic graph of topic dependencies, the proposed model captures collective learning behaviour across two core dimensions: academic performance and subjective/introspective mental effort. During SOM training, representative data sets of successful learner trajectories are compressed into two topologically ordered, identically structured neural networks: PSOM (performance space) and ESOM (mental effort space). By pairing these behavioural maps, the model serves as a decision-support tool that enables instructors to simultaneously pinpoint structural instructional design defects, such as hidden topic prerequisites, and evaluate collective cognitive load profiles. The implementation and performance of the PSOM component (and, by extension, its symmetrical twin, ESOM) is linked to an earlier modelling study employing a SOM with identical structure and encoding semantics situated in the same domain (i.e., a learning environment). This paper combines the study of behavioural phenomena with theoretical work on human cognitive architecture via SOM neurocomputational cognitive behavioural modelling.
