Machine Learning
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Saturday 7 November
Embodied intelligence systems rely on fast and reliable inference to enable real-time perception, reasoning, and control. Recent advances in large-scale artificial intelligence (AI) models, including Large Language Models (LLMs) and Vision-Language-Action (VLA) models, have significantly enhanced the capabilities of embodied agents in complex environments. However, deploying these models remains challenging due to the limited computational resources and energy budgets of robotic platforms. A fundamental question is where inference should be executed. Collaborative inference, which partitions models across end devices, edge servers, and the cloud, has emerged as a promising solution. Yet determining the optimal partitioning point remains an open problem, as it depends on wireless network conditions, model structure, and privacy requirements. This article reviews the state of the art in collaborative inference for embodied intelligence and proposes an edge-centric framework in which the edge coordinates model partitioning and execution. Through both simulation and real-world experiments over WiFi and 5G networks, we demonstrate that the optimal partitioning strategy varies with network conditions and model structure. We further identify latency-minimizing partition points for ResNet-101, Qwen2.5-VL-7B, and OpenVLA-7B in collaborative inference between NVIDIA Jetson Nano or AGX Xavier devices and an AGX Thor edge server under varying wireless conditions. Finally, we discuss the future directions for collaborative inference in 6G-empowered embodied intelligence systems.
Analysis of PDEs
A uniform lower bound for logarithmic Trudinger-Moser suprema in dimension $N\ge3$
Let $B$ be the unit ball of $\mathbb{R}^N$, $N\ge2$, and let $\mathrm{MT}_N(\alpha_{N,\beta},\beta)$ denote the critical Trudinger-Moser supremum of Calanchi and Ruf for radial functions in the Sobolev space with logarithmic weight $(-\log|x|)^{\beta(N-1)}$, $0\le\beta<1$. We prove that $\mathrm{MT}_N(\alpha_{N,\beta},\beta)>(2N-\frac25)|B|$ for every $\beta\in[0,1)$ and $N\ge 3$. This lower bound is uniform in $\beta$ and lies strictly above the upper bound $|B|\big(1+e^{1+\frac12+\cdots+\frac1{N-1}}\big)$ for the concentration level obtained by Carleson-Chang for $\beta=0$ and, for $\beta>0$, by Roy for $N=2$ and by Roy and Nguyen for $N\ge 3$. As a consequence, we prove that $\mathrm{MT}_N(\alpha_{N,\beta},\beta)$ is attained for every $\beta\in[0,1)$ and $N\ge3$, which was previously known only for $\beta>0$ small enough. For $N=2$, the existence of extremals for every $\beta\in [0, 1)$ was recently obtained by de Oliveira.
Read the full abstract
Quantum Physics
A Witness for Quantum Correlations Using Nonlinear Spectroscopy
The ability to detect and quantify entanglement in molecular systems is essential for understanding their quantum correlations and to evaluating molecules as platforms for quantum technologies. Here we show how signals from existing spectroscopic techniques can provide sufficient information to witness entanglement. The main ingredient is a spectroscopic map from admissible quantum states to a finite-dimensional space of signal coefficients, which are expectation values of state-independent effective observables. The molecular response model and measurement protocol determine the effective observables for a fixed system bipartition, and the spectroscopic map defines sets of joint expectation values compatible with all admissible quantum states and their separable subset. We illustrate the construction of the effective observables and their expectation value sets using rotational invariants and the spectral structure of optical signals. Fluorescence dynamics and photon coincidences provide a complementary example. Our findings provide a theoretical basis for entanglement certification with established spectroscopic techniques.
Read the full abstract
Machine Learning
Unified Factored Representations Resist Catastrophic Forgetting
This paper explores an unexpected connection between the quality of a neural network's (NN's) internal representation and its vulnerability to catastrophic forgetting. The representational forgetting hypothesis is that networks with unified, factored representation (UFR) can resist catastrophic forgetting without any modification to the underlying fine-tuning algorithm or network architecture. The key insight is that catastrophic forgetting may largely result from fractured entangled representation (FER): fine-tuning on only part of the input domain results in disruptive changes for other input patterns, because of how individual features are globally entangled. In contrast, if a gradient update is compatible with how a UFR network factors its representation, such an update may result in coordinated change across the input domain, alleviating catastrophic forgetting. Experiments in two domains provide evidence for this hypothesis: (1) an illustrative 1D regression task, in which UFR networks constructed by hand are compared against networks trained from scratch with SGD, and (2) an image-regression task with NNs previously shown to exhibit UFR. In both settings, UFR networks forget substantially less than their FER counterparts when repeatedly fine-tuned on UFR-compatible variations. Further experiments show that a training loss that distills UFR representations into fresh NNs reduces forgetting; along with several additional controls, these results help isolate representation as the critical factor. Together these results suggest that representational quality is an underappreciated contributor to catastrophic forgetting, and that the encouragement and maintenance of such quality may require attending to what patterns a particular NN is ready to learn.
Read the full abstract
Chemical Physics
Toward universal excited-state machine-learned potentials: a local excitation Hamiltonian for size consistency and transferability
Universal machine-learned interatomic potentials could make excited-state simulations across chemical space routine, but excitation energies, unlike ground-state energies, are not sums of atomic contributions. The excited states of noninteracting molecules are the union of their states, which pooled multistate readouts reproduce at best for the lowest state. We introduce LEH-MACE, whose excitation energies are the eigenvalues of a local excitation Hamiltonian assembled from atom-centred blocks, so this union holds exactly. We generated TD-DFT energies and forces for about 300,000 equilibrium, thermal, and excited-state geometries of over 100,000 molecules. LEH-MACE's advantage over direct readouts grew with the distance from the training data and was largest above $S_1$, with about half their $S_2$ and $S_3$ errors for long conjugated oligomers. For molecule pairs, only LEH-MACE kept its single-molecule accuracy, while a softmin readout remained superior for $S_1$ dynamics of larger chromophores. Exact size consistency is thus one step toward such potentials.
Read the full abstract
Data Structures and Algorithms
Solving Large Rubik's Cubes within a Factor of $2.37+o(1)$ of Optimal
We design optimized algorithms to solve the $n \times n \times n$ Rubik's Cube within $30.5 n^2 / \lg n + o(n^2 / \lg n)$ moves. This upper bound is asymptotically within a factor of $2.37$ of the known lower bound, which applies to almost all instances. The constant $30.5$ is also the first explicit constant factor for the known upper bound of $O(n^2 / \lg n)$, where the implicit constant of the previous algorithm was $1.01 \cdot 10^{40}$ (ten duodecillion). Thus we improve the upper-bound constant by a factor of more than $3.30 \cdot 10^{38}$ (330 undecillion). Our algorithm runs in time linear in the input size $N = \Theta(n^2)$, and uses $O(\sqrt{N} \lg N)$ extra space (beyond the state of the cube). Our implementation is the first practical cube solver to generate solutions of length $O(n^2/\lg n)$. Our latest solver outperforms existing solvers for $n$ as small as $840$.
Read the full abstract
Machine Learning
ProtLink: Sequence and Multispecies Graph Context for Protein Interaction Prediction
Predicting protein-protein interactions (PPIs) from sequence alone is difficult, while known interaction networks may provide informative biological context. We introduce ProtLink, a sequence-conditioned graph-learning framework that combines frozen ESM-2 representations with protein interaction topology and orthogroup-mediated connections between species. We compare sequence-only models, GraphSAGE, a homogeneous graph transformer, and a heterogeneous graph transformer (HGT) for human PPI prediction. In an initial edge-level comparison, three-layer multispecies HGT with ESM-2 650M features achieves an average precision (AP) of 0.980 under random negative sampling, compared with 0.951 for the homogeneous transformer; its AP decreases to 0.878 with two-hop-preferred negatives. In controlled three-seed experiments, three-layer full-graph HGT obtains an AP of $0.9701\pm0.0022$, exceeding an orthogroup-only model ($0.9435\pm0.0094$). The corresponding two-layer comparison shows little additional benefit from non-human PPI edges, consistent with the greater message-passing depth needed to transmit their information. For human proteins excluded from training interactions, multispecies HGT achieves $0.8335\pm0.0094$ AP versus $0.6482\pm0.0052$ for a sequence-only pair model. These results identify the value, depth dependence, and limits of evolutionary graph context while distinguishing prediction of held-out edges from generalization to held-out proteins.
Read the full abstract
Information Retrieval
RoutingTrie: Metadata Routing for Constrained Generative Retrieval
Advances in large language models and semantic IDs (SIDs) have brought generative retrieval (GR) into industrial applications. These systems must return items that satisfy request metadata constraints as well as relevance requirements. A catalog trie ensures SID validity, but filtering metadata only after generation wastes computation on rejected candidates and may leave too few eligible results. We propose RoutingTrie to incorporate metadata constraints before beam selection. RoutingTrie maintains one shared SID trie and builds a compact index for each metadata field using Dense Bitset or Sparse List. We derive a storage cost formula to guide the choice between these index representations for different field distributions. During decoding, field probes determine whether requested values occur below candidate edges and produce a combined mask that guides beam selection. Experiments on ShopTrie, our configurable synthetic benchmark, demonstrate compact storage and efficient CPU and GPU probing across diverse metadata distributions. In a production replay, RoutingTrie raises Hit@100 from 67.5% to 96.1%, a gain of 28.6 percentage points, with only 1.1 ms latency increase. We will release the ShopTrie generator, benchmark scripts, and RoutingTrie CPU and Triton GPU implementations to support reproducibility.
Read the full abstract
Mathematical Physics
Cylinder spectral curve and 1-cut Matrix model
For a random matrix, with eigenvalues in a one-body potential $V(x)$ and a 2-body interaction $R(x,x')$, it is known that the resolvent converges in the large $N$ limit to an analytic function in the complex plane minus cuts, called the spectral curve. In the 1-cut case, a two-sheeted neighborhood of the cut is a cylinder with two simple ramification points. We describe these spectral curve data by boundary integral operators and generalized cycles. Vice-versa, given a 2-branchpoints spectral curve in a cylinder, there exists a formal random matrix model that corresponds to it. The goal of this article is to show the method to recover a formal random matrix model, with its potential and 2-body interaction, from the geometry of the spectral curve.
Read the full abstract
Probability
The stochastic phase field crystal equation with conservative space-time white noise: well-posedness, regularity and existence of a density
We study the phase field crystal equation $\partial_tu=\Delta[(1+\Delta)^2u+u^3-\varepsilon u]+\sigma\partial_x\dot W$ on $[0,\pi]$ with the boundary conditions $\partial_xu=\partial_x^3u=\partial_x^5u=0$ at $x=0,\pi$, perturbed by a conservative space-time white noise. This sixth order equation is the standard model for crystal growth at atomic length scales and diffusive time scales, and the conservative noise is the form of thermal fluctuations used in the physics literature. We prove that for every initial condition in $H^1$ there exists a unique global mild solution, continuous in space and time, which preserves the mean and whose supremum has finite moments of all orders. We determine its Hölder regularity: away from $t=0$, it is Hölder continuous in time with any exponent smaller than $1/4$, and $C^1$ in space with a spatial derivative which is Hölder continuous with any exponent smaller than $1/2$. Finally, using Malliavin calculus and a localization argument, we prove that for $\sigma\neq0$ the law of $u(t,x)$ is absolutely continuous with respect to the Lebesgue measure for every $t>0$ and every $x$. A specific feature of the model is a second neutral mode of the linear operator, besides the constant mode, which requires a separate treatment in the a priori estimates.
Read the full abstract
Information Retrieval
Slim Foundation Models for Latency-Critical Recommendation Serving
Click-through rate (CTR) prediction is a core component of recommendation and advertising systems. Large language models (LLMs) offer rich semantic priors that improve prediction quality, particularly in cold-start and long-tail settings. Injecting a full-scale LLM into a CTR pipeline is, however, prohibitively expensive at inference time: existing LLM-enhanced CTR methods avoid this cost by caching LLM outputs offline rather than serving the LLM online. We propose a resource-efficient pipeline, SlimFM, that makes online LLM-based semantic modeling practical: (1) dependency-aware structured pruning compresses LLaMA3-8B by 44% to a 4.5B-parameter backbone; (2) LoRA-based field-aware fine-tuning trains the pruned model to produce discriminative field-level semantic embeddings from lightweight textual prompts; (3) a semantic-guided fusion module injects these embeddings into the feature interactions of a conventional CTR model. Across four datasets and six CTR backbones, SlimFM retains over 99% of uncompressed-model accuracy while reducing inference latency by up to 3.4x relative to existing LLM-enhanced baselines.
Read the full abstract
High Energy Physics (Phenomenology)
Sensitivity of jet and dijet observables to medium effects in pp collisions at $\sqrt{s}=13$ TeV using X-SCAPE/JETSCAPE
Establishing jet quenching in small collision systems, such as high-multiplicity proton-proton collisions at the LHC, is challenging because the expected modifications are small and can be obscured by event-selection effects, multiparton interactions, and multijet production. We examine the response of several jet and dijet observables to medium-induced changes in proton-proton collisions at $\sqrt{s}=13$ TeV using the JETSCAPE/X-SCAPE framework. First, a simplified Bjorken-expanding brick medium is compared with a vacuum reference for several brick lengths and strengths of transport coefficients. Second, separately generated realistic vacuum and medium samples with fluctuating initial conditions and event-by-event hydrodynamic evolution are compared after selecting the highest-activity 20% of events. Full anti-$k_T$ jets are reconstructed with $R=0.2$ and $R=0.4$. The observables are dijet invariant mass, leading- and subleading-jet transverse momentum, folded azimuthal separation, dijet asymmetry, Soft-Drop groomed jet radius, and groomed mass. In the brick calculation, the dijet-mass spectrum shows the clearest ordered response: departures from the vacuum reference generally increase with brick length and strength of transport coefficient and are more visible for the narrower jet radius, while the jet-$p_T$ and dijet-asymmetry changes are smaller and no clear azimuthal broadening is resolved. In the selected realistic events, the ungroomed medium-to-vacuum ratios remain close to unity but contain localized kinematic differences, together with a modest increase of the mean dijet asymmetry. The groomed-radius and groomed-mass ratios show localized differences but do not exhibit a persistent angular narrowing or mass redistribution. These findings quantify the sensitivity of the studied observables to medium-induced jet modifications in the present pp calculations.
Read the full abstract
Machine Learning
Learning Actionable Information Landscapes for Multimodal Active Sensing in Hawkmoths
Animals integrate information across sensory modalities to guide exploration, suggesting that they maintain internal estimates of where actions are expected to reduce uncertainty. However, how such action-conditioned information maps are learned from multimodal sensory experience remains unclear, as previous studies have been largely confined to unimodal sensing frameworks. We introduce a multimodal predictive-coding framework for learning uncertainty-reduction maps from visual and mechanosensory observations in hawkmoth flower interaction tasks. A variational generative model trained on simulated environments recovers modality-specific information landscapes over the action space, capturing how visual patterns, surface geometry, and mechanosensory cues shape the value of exploratory actions. These learned maps are consistent with several features of hawkmoth behavior, including nonzero tracking offsets, probing along visual patterns, and geometry-dependent exploration. Applying the model to behavioral data from flower tracking and nectary search, we find that natural trajectories tend to occupy regions of high uncertainty reduction in these learned maps. Building on these observations, we develop an uncertainty-gated lexicographic algorithm that switches between information gathering and reward exploitation based on perceptual uncertainty. In nectary search, uncertainty gating improves learning relative to exploit-only, $\epsilon$-greedy, fixed-time, and uncertainty-calibrated non-lexicographic controls, with better generalization and robustness to sensory and curvature shifts. These results link information landscapes learned through generative modeling to biological sensing and embodied exploration.
Read the full abstract
Artificial Intelligence
Lightfoot: Graph Retrieval-Augmented Generation with Exact, Model-Free Document Deletion
Graph-based retrieval-augmented generation (GraphRAG) improves multi-hop question answering, but deployed systems pay costs that research comparisons rarely measure: model calls at query time, building the graph, and removing documents correctly. We present Lightfoot, a GraphRAG system in which every entity, link, fact and synonym edge is a deterministic function of per-passage extractions that carry chunk-level provenance. Retrieval makes no language-model call, indexing makes one per passage, and deleting a document is a database transaction with no model calls: 14 seconds for 10% of an 11,402-passage corpus, after which the graph is identical to a fresh index of the surviving documents. On a corpus of near-duplicate names, HippoRAG 2's capped synonym lists leave 565 fewer edges than a rebuild, while Lightfoot remains exact. Under identical extraction, embedding and answer models on held-out questions of MuSiQue, 2WikiMultiHopQA and HotpotQA, Lightfoot is competitive with HippoRAG 2: with Holm-corrected paired tests it is better on MuSiQue at recall@5 and recall@10 (+4.8 and +5.5 points) and on 2WikiMultiHopQA at recall@10 (+3.4), and tied on HotpotQA, while HippoRAG 2 keeps a non-significant 3.0-point edge at 2WikiMultiHopQA recall@2. Seeding the walk from entities named in the question accounts for most of the 2WikiMultiHopQA gain. Answers tie on all three datasets, and the retrieval pattern holds with three larger embedders, up to Qwen3-Embedding-8B. The gain is not the reranker's: with the same cross-encoder, dense retrieval trails Lightfoot by 9.3 and 17.8 points recall@5 on the two multi-hop sets. Lightfoot also builds the cheapest index, 26-31% below HippoRAG 2; LightRAG needed 2,187 model calls and 3.7 hours for the same deletion, and fast-graphrag cannot delete.
Read the full abstract
Machine Learning
PIA: Agentic Physics-Informed Runtime Anomaly Detection
Runtime anomaly detectors for cyber-physical systems are either learned as black boxes from normal data, which miss stealthy manipulations that keep every signal inside its normal range, or derived by hand from physical models, which requires expert effort for every plant. We introduce physicsinformed agents (PIA), in which a large language model agent writes the detector itself as a set of physical residuals, such as pump characteristics, tank mass balances, and controller logic, and refines every candidate against normal operating data through an evaluation workbench before deployment. The result is a compact, human-readable detector that runs in milliseconds and names the component behind each alarm. On the BATADAL water distribution attack benchmark, PIA with Claude Opus 5.5 reaches an F1 of 0.878 on the test attacks against 0.504 for the strongest of six statistical and deep baselines, keeps detecting attacks concealed by replayed and offset sensor readings, and attributes every detected test attack to the attacked component. Ablations show that testing hypotheses against data is decisive: agents that write residuals from the plant description alone lose up to 0.28 F1, whereas agents that see only anonymized signals recover most of the physical structure.
Read the full abstract
Statistical Methodology
Cause-specific structural nested cumulative survival time models for competing risks, with application to the Antibody Mediated Prevention trials
The Antibody Mediated Prevention (AMP) trials randomized participants to VRC01, a broadly neutralizing antibody, or placebo; the primary analysis estimated prevention efficacy against overall HIV-1 acquisition and against acquisition of VRC01-sensitive strains. All participants were offered daily oral pre-exposure prophylaxis (PrEP), and PrEP uptake in a trial may differ substantially from uptake in other settings. It is therefore of interest to estimate the effect of oral PrEP on acquisition of VRC01-sensitive and VRC01-insensitive strains, as well as the counterfactual cumulative incidence of such acquisition under no or limited PrEP uptake. To address these questions, we extend the structural nested cumulative survival time model (SNCSTM), a causal survival model for time-varying treatments, to a cause-specific version. We show that the cause-specific SNCSTM is closely related to the Fine–Gray subdistribution hazard model, and we study identification under two frameworks. The first, based on sequential ignorability, applies when the time-varying treatment is distinct from the randomized assignment, as with oral PrEP use in AMP. The second, based on an instrumental variable, applies when the time-varying treatment represents compliance with the randomized assignment, and it accommodates unmeasured confounding between the time-varying treatment and event time. Under both frameworks, we propose consistent and asymptotically normal estimators of the structural parameters. We apply the methods to HVTN 704/HPTN 085, one of the two AMP trials, to estimate the effects of oral PrEP use and of VRC01 infusions administered every 8 weeks on acquisition of VRC01-sensitive and VRC01-insensitive HIV-1 strains. We also estimate the counterfactual strain-specific cumulative incidence curves under no oral PrEP use and under full adherence to the scheduled VRC01 infusions.
Read the full abstract
Computer Vision
Efficient 3D plant segmentation through multiscale graph learning and optimized geometric processing
Plant organ segmentation on compact computing platforms requires accurate predictions and affordable geometric preparation. We develop and evaluate a multiscale edge based graph attention model that combines local shape descriptors, two neighborhood scales and a residual stem and leaf correction in approximately 78 thousand parameters. The engineering application is plant point cloud segmentation for precision agriculture, assessed through grouped accuracy experiments and physical Raspberry Pi 5 inference. Broad comparisons with hierarchical, graph and transformer baselines establish PointNeXt as the strongest external accuracy reference among the tested local recipes. A nested four variant ablation on a local maize collection separates geometric information from the combined multiscale configuration across seven outer groups and three seeds. The complete model achieves the highest mean intersection over union, 86.36%, although uncertainty and boundary errors qualify the contribution of each component. We then optimize full cloud principal component analysis, sampling and neighbor queries without changing learned weights or descriptor definitions. All 1,180 paired desktop comparisons on PlantNet and TomatoWUR preserve exact logits and predictions. Multiscale pipeline latency falls from 330.33 to 96.82 milliseconds on PlantNet and from 1,054.52 to 463.10 milliseconds on TomatoWUR. PointNeXt also benefits from optimized sampling and remains faster over the complete pipeline. These results support compact graph learning with reduced geometric overhead, while physical board measurements establish inference feasibility rather than energy efficiency or readiness for autonomous flight.
Read the full abstract
General Relativity and Quantum Cosmology
Dynamical entropy on local causal horizons: the Einstein equation of state and canonical-energy flux
For a twist-free null hypersurface, the Hollands-Wald-Zhang (HWZ) representative $S_{\rm loc}=(4G)^{-1}\int(1-V\theta)dA$, local on each cross-section (cut), obeys an exact off-shell Raychaudhuri identity with no term linear in the expansion, so no final equilibrium cut is needed. Given an entropy density $1/4G$, the Unruh temperature $\kappa/2\pi$ of an approximate local boost, local equilibrium cuts, and stress-tensor conservation, equating the leading equilibrium change of $S_{\rm loc}$ to matter boost heat yields $R_{kk}=8\pi GT_{kk}$ for every null direction, hence the Einstein equation with undetermined cosmological constant, an equation of state in Jacobson's conditional sense. On a fixed caustic-free generator bundle with $dA=J\,dA_{\rm ref}$, the functional $S_{\rm L}=(4G)^{-1}\int(1+\ln J-V\theta)dA_{\rm ref}$ obeys an exact balance per reference cell with curvature, shear, and expansion terms all positive, so under null convergence $S_{\rm L}$ is nondecreasing for $V\geq0$ until the first caustic. Conversely, requiring this balance for positive affine parameter on every small bundle and null direction is equivalent to the Einstein equation, given conservation and the stated normalization. Cut-locality, positivity, and first-order HWZ normalization select the density $1+\ln J-V\theta$ uniquely within a two-function class. Both functionals depend on the affine origin; we give their shift laws, identify the anchoring-invariant content, and show the origin is physical data fixed by the equilibrium cut. Near equilibrium, $S_{\rm L}$ agrees with the HWZ dynamical entropy at first order, and the temperature-weighted change of its Hessian equals the HWZ modified canonical-energy flux for stationary-vacuum perturbations. At finite amplitude, $S_{\rm L}$ is only a monotone geometric functional; reading it as entropy requires a coarse-graining postulate, stated but not derived.
Read the full abstract
Machine Learning
GradReason: LLMs as Meta-Learners that Replace Training Epochs with Training-Data-Free Weight Updates
Standard training recipes make a fixed number of passes over the training data, even when an epoch mostly adds mini-batch noise. We study whether a large language model (LLM) that reads the weight and gradient history of a run can replace some of these epochs with updates that compute no gradient and read no training image. In GradReason, an LLM agent is called every ten epochs. In a restricted sandbox, it analyses recent weight snapshots, epoch-mean gradients, and validation metrics, then submits a Python program that computes new weights and a plan marking each of the next ten epochs as data or virtual. A virtual epoch is kept only if it passes a forward-only check on 5,000 held-out validation images. In 120 runs on CIFAR-10 and CIFAR-100 with three architectures, Claude Opus 5.5 used 32.8% fewer data epochs than the 80-epoch recipe at 0.31 percentage points lower test accuracy (95% CI -0.72 to +0.09; 0.94 points on ResNet-18), on the recipe's own frontier of data against accuracy. Unprompted, Opus averaged recent weight snapshots in every run. This choice produced the savings: a fixed rule with the same acceptance check but extrapolations from prior work saved 5.4% of the data epochs, and the same rule with averaging added saved 25.0%. The LLM's contribution is to find a working update within a single run; once found, a fixed rule implements it equally well. An Opus run cost 0.68 USD, but at this scale the agent's latency exceeded the GPU time it saved.
Read the full abstract
Statistical Methodology
Choosing the basis to compare multivariate distributions: bases, copulas and optimal transport
We study the choice of an orthonormal basis for comparing two probability distributions on $\mathbb{R}^d$ through their one-dimensional projections. We show that two distributions have a common copula in a given basis if and only if their optimal transport can be performed componentwise along the directions of that basis. This motivates the notion of componentwise related distributions and leads to a structural analysis of the bases in which such a representation is possible. In particular, we establish an essential uniqueness result for the corresponding orthogonal decomposition and characterize elliptical families through equality in Gelbrich's bound within location-scatter families. For general distributions, we consider the maximum, over all orthonormal bases, of the sum of the squared quadratic Wasserstein distances between the corresponding one-dimensional projections. We prove that the resulting quantity is a metric, induces the same topology as the quadratic Wasserstein distance, and lies between the Gelbrich and Wasserstein distances, with equality in the upper bound precisely for component-wise related distributions. We also derive first-order conditions for an optimal basis and establish stability and almost-sure consistency of empirical optimal bases. Simulations and examples illustrate the computational feasibility and competitive performance of the proposed procedures and suggest new applications of copulas to multivariate distributional comparisons.
Read the full abstract
Econometrics
Do Modern Difference-in-Differences Estimators Change Published Findings in Business Research?
Staggered-adoption difference-in-differences (DiD) designs are a workhorse of empirical research in business-school journals. The two-way fixed effects (TWFE) estimator that typically implements them can place negative weights on heterogeneous treatment effects. We ask whether negative weights change published findings by reanalyzing staggered-adoption specifications in business-school journals with journal-wide public replication archives. Following Hull (2026), we use large language model agents to identify eligible specifications, replicate the published TWFE estimate, and re-estimate each specification with the DiD-averaging estimator of Callaway and Sant'Anna (2021) and the imputation estimator of Borusyak et al. (2024). We re-estimate 355 specifications from 40 papers. Across the 283 DiD-averaging and 306 imputation comparisons that pass Hull's leverage screen, from 28 and 29 papers, TWFE and modern estimates have correlations of 0.61 and 0.70, change sign in 25.1 and 18.0 percent of comparisons, and differ significantly with opposite signs in 10.2 and 7.2 percent. Modern standard errors are 1.61 and 1.30 times those of TWFE. Including the comparisons that fail the screen leaves these figures essentially unchanged. Our results suggest that TWFE and heterogeneity-robust estimators disagree in a non-negligible share of the published staggered designs we re-estimate, and more often than in economics.
Read the full abstract
Group Theory
The Herzog–Schönheim Conjecture for Solvable Groups
The Herzog–Schönheim conjecture states that if a group is partitioned into finitely many, and at least two, cosets of subgroups, then two of these subgroups have the same index. We prove the conjecture for all solvable groups, finite or infinite. Sun's bound on unions of cosets, applied in a quotient, is combined with divisibility restrictions on local indices relative to the subgroup generated by the $p$-elements, where $p$ is the largest prime divisor of the order of a minimal counterexample. These restrictions are obtained by counting fixed points in a primitive solvable quotient. This suffices except at the prime $7$, where the counting bound narrowly fails and an additional overlap argument is needed. The proof has been formalized in Lean~4 with mathlib.
Read the full abstract
Machine Learning
Elastic Representations via Hyperbolic Geometry
Learned representations (or embeddings) are the foundation of modern machine learning systems, yet they are typically trained as fixed-size embeddings, without accounting for varying downstream resource or task constraints. Matryoshka Representation Learning (MRL) addresses this limitation by learning nested representations, but requires costly end-to-end retraining and does not support post hoc expansion, while Contrastive Sparse Representations (CSR) reduce training overhead via lightweight adaptors but operate at a 4x larger dimensionality. Additionally, all prior works restrict their framework towards compression only. In this work, we introduce Hyperbolic Elastic Representation Learning (HERL), a framework for learning dimension-adaptive embeddings (compression and expansion) through a lightweight adaptor without retraining the base encoder. Our key idea is to exploit the exponential capacity of hyperbolic geometry by learning a non-linear mapping from the encoder space to hyperbolic space. We train a simple MLP to downsample the representations before projecting them onto a Lorentzian and learn adaptive representations that respect the Lorentzian geometry. Our approach is model-agnostic and effective in both supervised and unsupervised settings. Through extensive experiments across image and text modalities, we show that HERL preserves downstream performance across a wide range of embedding sizes, exhibits smooth interpolation between dimensions, and achieves these benefits at a fraction of the computational cost. Code: https://github.com/arjunrkaushik/herl.
Read the full abstract
Materials Science
Efficient Plane-to-Point Energy Transfer from Supramolecular Excitonic Systems to Molecular Acceptors
Collective excitonic interactions in supramolecular assemblies offer opportunities for efficient light harvesting and directed energy transport. Here, we demonstrate efficient energy transfer from two-dimensional silica-encapsulated TDBC J-aggregate sheets to covalently attached ATTO 590 molecular acceptors. Silica encapsulation preserves the J-aggregate excitonic properties while enabling acceptor functionalization at a defined separation from the donor plane. Steady-state donor quenching and acceptor sensitization yield energy-transfer efficiencies of 75-81% in solution, while time-resolved photoluminescence measurements reveal donor lifetime shortening. We develop a generalized Forster resonance energy-transfer model for an extended planar donor coupled to localized molecular acceptors, predicting an inverse fourth-power dependence of the transfer rate on donor-acceptor separation. Using experimental spectroscopic data and estimated structural parameters, the model predicts an energy-transfer rate of 4.77 x 10^10 s^-1, approximately ten times faster than the intrinsic donor decay rate, corresponding to a theoretical energy-transfer efficiency of approximately 91%. These findings provide a framework for developing hybrid excitonic architectures for light harvesting.
Read the full abstract
Machine Learning
Differentiable Agent-Based Modeling for Multi-Echelon Supply Chains: Gradient-Based Optimization of Heterogeneous Policies at Scale
Global supply chains rely on the interactions of thousands of heterogeneous agents, posing significant challenges for optimization. While traditional optimization often assumes homogeneity to ensure tractability, Reinforcement Learning struggles with sample inefficiency when scaling to high-dimensional state spaces. We propose a Differentiable Agent-Based Model for multi-echelon inventory management. By implementing stochastic dynamics via differentiable tensor operations, we enable direct gradient propagation through the simulation. We validate the framework by recovering theoretical Newsvendor optima without supervision. Crucially, we demonstrate “Mass Personalization” by optimizing 10,000 distinct agent policies simultaneously in seconds, achieving a 53% cost reduction compared to scalar baselines. Furthermore, we utilize gradient sensitivity analysis to generate causal heatmaps, accurately recovering inherent lead-time delays from the computation graph. This approach bridges the gap between simulation and optimization, enabling the efficient training of self-correcting Digital Twins.
Read the full abstract
High Energy Physics (Theory)
Spectral function, OPE, and QNMs detect different singular points of black branes with unequal R-charges
We consider black brane solutions in the STU model, with three R-charges labeled by the parameters $q_1,q_2,q_3$. For equal R-charges $q_1=q_2=q_3$, the solution becomes the Reissner-Nordström black brane. For unequal R-charges $q_1>q_2>q_3\geq0$, there are three curvature singular points at $R\equiv r^2=-q_i$. Only $R=-q_3$ is the physical singularity, and the other two lie beyond it. Their optical times $T_i$, the complex times of null rays from the boundary to these singular points and back, form a right triangle in the complex $T$-plane. We probe these singular points using three observables, all built from the retarded correlator of $T_{xy}$ at zero spatial momentum: spectral function, OPE coefficients, and QNMs. The exponentially small correction to the spectral function at large real frequency is dominated by $T_1$ and $T_2$, which have the smallest ${\rm Im}\,T_i$. The OPE convergence radius is determined by $T_1$, which has the smallest $|T_i|$. A line of QNMs lies along the direction of $1/T_2$, since $T_2$ has the smallest phase $\arg T_i$. Furthermore, there is a second line of QNMs along the negative imaginary axis, which carries information about $T_3$ corresponding to the physical singularity. We derive these results from a WKB analysis and check them numerically. The three observables are therefore sensitive to different singular points of the geometry.
Read the full abstract
Computation and Language
Open Problems in Chatbots and Mental Health
General-purpose chatbots are increasingly used for emotional and mental health support. They can be useful, but issues such as sycophancy, reinforcement of delusional beliefs, and poor crisis response create risks that are difficult for researchers to study and evaluate. In this paper, we review the most critical open problems for researchers and policymakers in the use of general-purpose chatbots and their effects on mental health. First, there are tractable problems in the controlled evaluation of chatbots: establishing the construct and ecological validity of evaluations, evaluating multi-turn interactions, and connecting evaluations to improvements in chatbot behavior. For example, conventional benchmarks that primarily rely on static datasets cannot establish chatbot safety, and novel simulation methods need validation. Second, researchers must connect their studies of chatbots and mental health to real-world evidence. This includes analyzing the chatbot products that AI companies and mental health practitioners have deployed, including their user interfaces and human review and escalation practices for user welfare concerns. However, clinical analysis opportunities afforded by real chat log data sharing are complicated by significant privacy concerns. Finally, we argue that researchers should take an active role in the construction and refinement of evidence-based policy through corporate and governmental decisions such as user protections, standards of care, protected access to user chat logs, data sharing between institutions, incident reporting, and post-deployment monitoring. Such behavioral evaluations, studies of real-world outcomes, and data governance depend on each other and must be addressed in tandem. We believe a coordinated research effort would have substantial real-world impact.
Read the full abstract
Machine Learning
Toward a theory of Evaluability
Machine learning models, particularly large foundation models, are increasingly evaluated along multiple axes that are relevant for their usage. In addition to accuracy measures appropriate for the specific task of interest, measures capturing safety, calibration and fairness properties are often of interest. The classical approach to these problems assumes that the test set has been held out from the training process, an assumption that is increasingly harder to justify for large foundation models. In this work, we formalize a notion of evaluability without this assumption, and ask if it is feasible to evaluate black-box models in settings where our test samples may have been used during training. We study this problem under different settings, capturing different levels of access to the model. We first study the natural setting, where the evaluator only looks at the output of the model on the test set. We fully characterize the class of distributions where accurate evaluation is possible, and show tight upper and lower bounds on the sample complexity of evaluation. Our results show that black-box evaluation in this setup is feasible if and only if the distribution is close to being small support. We then consider evaluation algorithms that can query the model on additional inputs and demonstrate connections to self-correctors studied in program testing. Using these, we show that a small amount of query access can increase the power of the evaluator for natural distributions and concept classes.
Read the full abstract
Dynamical Systems
Frequency shifts: non-soficity, absence of specification and intrinsic ergodicity
For $p_c\in(0,1)$, the frequency shift $\Sigma(p_c)\subset\{0,1\}^{\N}$, introduced by Becker and Baraviera, is the closure of the shift orbits of the set of sequences in which a symbol $1$ is followed by $1$ exactly when the density of ones up to that position does not exceed $p_c$. We show that $\Sigma(p_c)$ is never sofic and never has the (weak) specification property. For $p_c=\frac{k}{k+1}$, with $k\ge1$, we describe $\Sigma(p_c)$ completely: $\Sigma(\frac12)$ is the coded system generated by $\{0^m1^m\}_{m\ge1}$, while for $k\ge2$ the shift $\Sigma(\frac{k}{k+1})$ is not transitive, but every invariant probability measure is supported on the coded system generated by $\{0^m1^{km}\}_{m\ge1}$. For every $k \ge 1$, the topological entropy of $\Sigma(\frac{k}{k+1})$ is $\frac{\log2}{k+1}$, and, using the Climenhaga–Thompson criterion, we prove that $\Sigma(\frac{k}{k+1})$ is intrinsically ergodic.
Read the full abstract
Robotics
RoboMoCap: A Large-Scale Whole-Body Motion Capture Dataset with Egocentric Vision for Robot Learning
Humanoid robots must coordinate locomotion and manipulation to navigate complex environments and perform everyday tasks. Learning such behaviours from human demonstrations requires accurate whole-body motion paired with egocentric vision, particularly for lower-body movements and foot placements that are frequently invisible from a first-person perspective. However, existing egocentric datasets often lack high-fidelity optical motion capture or systematic coverage of diverse loco-manipulation behaviours. To address this gap, we introduce RoboMoCap, a large-scale multimodal dataset combining egocentric vision with marker-based optical motion capture, specifically designed for locomotion-centric humanoid learning. RoboMoCap comprises over 200 hours of recordings from 14 participants, featuring six head-mounted cameras, wrist-mounted visual and inertial sensing, and optical full-body motion capture at 120 Hz. The dataset systematically covers diverse locomotion primitives, dynamic whole-body movements, and loco-manipulation tasks, including variable-speed walking, turning, constrained navigation, terrain traversal, object transportation, and manipulation during locomotion. It provides temporally aligned multimodal observations and metric whole-body motion representations, enabling the study of human movement beyond directly observable body regions. We systematically evaluate existing methods on three benchmark tasks: egocentric full-body motion reconstruction, foot contact and placement estimation, and human-to-humanoid motion retargeting. By pairing egocentric vision with high-fidelity human motion across diverse locomotion and interaction scenarios, RoboMoCap provides a foundation for egocentric motion understanding and humanoid learning from human demonstrations.