Machine Learning
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Tuesday 8 September
Ensuring safety in reinforcement learning under nonstationarity requires anticipating changes in risk before they lead to unsafe behavior. Existing approaches typically rely on safety constraints defined at design time or updated reactively during execution, assuming that such constraints remain valid over time. However, in nonstationary environments with evolving contexts and changing driving layouts, these assumptions may fail. We propose a framework for proactive safety constraint generation based on context forecasting. The approach infers latent environmental context from observations, predicts its future evolution, and constructs safety constraints adapted to anticipated conditions. This enables the agent to proactively avoid unsafe regions instead of reacting only after safety violations occur. We evaluate the method in driving environments with structured context variation. The experiments include a sweep over nonstationarity intensities and additional held-out driving layouts, including highway, intersection, and racetrack scenarios. Results show that proactive constraint generation substantially reduces collisions under both seen and out-of-training nonstationarity intensities and generally remains effective across held-out driving layouts while maintaining usable task performance. These findings suggest that context-based constraint generation is a promising approach for safe reinforcement learning under nonstationarity.
Computer Vision
VI-Bench: Benchmarking Prompt Inversion from AIGC Videos
Recent advances in video generation have made prompt-based control increasingly central to AIGC video generation. Prompts specify what a video should depict and how it should be represented, controlling factors such as visual style or camera behavior. Understanding this recoverability is important both for creative reuse and editing, and for assessing prompt leakage risks. However, existing video understanding benchmarks do not measure this capability: a caption may describe what is visible, but a replayable prompt must recover the generation-relevant controls needed to reproduce the video. To address this gap, we introduce VI-Bench, a benchmark built from 16.1 million real-user prompts and 900 human-verified AIGC videos. VI-Bench spans three progressively harder settings, namely single-shot semantic grounding, control over style and camera behavior, and multi-shot compositional inversion, and evaluates five generation-critical dimensions: subject, action, scene, style, and camera. We evaluate 18 representative VLMs, including 2 proprietary and 16 open-source models on VI-Bench, using an Inversion Score that measures prompt-level alignment with the original prompt and video-level fidelity of the regenerated video. The results reveal substantial limitations: even the strongest model achieves only 0.632 on Inversion Score, performance degrades sharply as samples require richer control and multi-shot reasoning, and models often produce plausible prompts whose regenerated videos deviate from the reference. These findings show that video prompt inversion is a distinct and under-evaluated capability requiring models to transform visual understanding into replay-stable generative control.
Read the full abstract
Machine Learning
A Machine Learning Framework for Predicting Restaurant Food Waste to Support Sustainable Food Management
Food waste in the restaurant sector poses a substantial challenge to environmental sustainability and economic efficiency. This paper presents an exploratory machine learning framework for estimating daily restaurant food waste quantities from operational and contextual features. A structured dataset was constructed by integrating restaurant demand records, meteorological data and temporal event indicators, yielding 77,980 records across 27 features. Because large-scale ground-truth food waste measurements are not publicly available, the target variable was derived from operationally justified assumptions, with the complete construction formula and controlled stochastic variability disclosed for full reproducibility. Four supervised regression models, namely Linear Regression, Decision Tree, Random Forest and Gradient Boosting, were evaluated under a chronological 70-30 train-test split that respects the temporal ordering of restaurant operations, augmented by 5-fold time-series cross-validation. All reported metrics are explicitly scoped to performance against the constructed target and do not imply validation against measured food waste. Ensemble methods consistently outperformed linear baselines. Random Forest attained an MAE of 6.19 kg, RMSE of 8.36 kg and $R^2$ of 0.817 on the realistic feature subset following systematic exclusion of algebraically leakage-prone variables. Feature importance analysis identified menu diversity, operational area and temporal activity patterns as the primary predictive drivers. The full dataset, target construction formula, codebase and experimental configurations are publicly released to support reproducibility and future extension to empirically measured waste data.
Read the full abstract
Combinatorics
Sharp edge-spectral supersaturation for odd cycles
Let \(G\) be a graph with \(m\) edges and adjacency spectral radius\(\rho(G)\), and let \(N(C_{2k+1},G)\) denote the number of copies of \(C_{2k+1}\) in \(G\). For each fixed integer \(k\ge 2\), define \( g_k(m):=\frac{k-1+\sqrt{4m-k^2+1}}{2}. \) Li, Zhai and Shu [European J. Combin., 2024] determined the spectral extremal threshold for odd cycles by proving that, for all sufficiently large \(m\), every \(C_{2k+1}\)-free graph \(G\) with \(m\) edges satisfies \(\rho(G)\le g_k(m)\). We establish the asymptotically sharp supersaturation counterpart of their result. More precisely, for every fixed integer \(k\ge 2\), we prove that \[ \inf_{\substack{e(G)=m\\ \rho(G)>g_k(m)}} \frac{N(C_{2k+1},G)}{m^k} = \frac{\lceil k^2/2\rceil (k-1)!}{(k+1)^k}+o(1) \qquad\text{as }m\to\infty. \] Thus every \(m\)-edge graph whose spectral radius exceeds the \(C_{2k+1}\)-free threshold contains at least \[ \Big( \frac{\lceil k^2/2\rceil (k-1)!}{(k+1)^k}-o(1) \Big)m^k \] copies of \(C_{2k+1}\), and the leading constant is asymptotically best possible. In particular, taking \(k=2\), we obtain if \( \rho(G)>\frac{1+\sqrt{4m-3}}{2}\) then \( N(C_5,G)\ge \left(\frac{2}{9}-o(1)\right)m^2, \) with the constant \(2/9\) being asymptotically optimal. This answers a question of Chen, Li and Tang concerning the existence and the largest possible value of a constant \(C>0\) for which the same spectral condition guarantees at least \(Cm^2\) copies of \(C_5\). More generally, our result resolves a recent problem of Li, Lin, Liu and Zhang on spectral supersaturation for odd cycles. The proof combines spectral stability and resolvent analysis with estimates for odd spectral moments and a careful treatment of non-injective closed walks.
Read the full abstract
Strongly Correlated Electrons
Selective coupling of high-order phonons in La2-xSrxCuO4
In crystals and molecules with strong coupling between the charge carriers and atomic vibrations, high order harmonics of the vibrational excitations can be observed with intensity following the Franck-Condon envelope. Here, we uncover a new regime of electron-phonon coupling in La2-xSrxCuO4 using resonant inelastic X-ray scattering. In the undoped compound, we find sharp peaks at approximately 85$\pm$5 meV, 180$\pm$7 meV, and 330$\pm$5 meV with no observable dispersion, and an absence of appreciable intensity from other modes observed in Raman spectroscopy in this energy range. The 180 meV and 330 meV excitations have energies approximately consistent with two-phonon and four-phonon in-plane Cu-O bond stretching mode, suggesting that the resonantly excited valence electrons strongly couple to this mode. These excitations exhibit significant doping dependence, eventually becoming unresolvable at 10% hole doping, where the low energy charge excitations are dominated by dispersive plasmons. The observation of selective coupling to even-ordered phonons and stark contrast with the Raman spectra indicate anomalous electron-phonon coupling beyond the Franck-Condon picture. One intriguing possibility is the existence of locally paired quasiparticles in insulating cuprates.
Read the full abstract
Programming Languages
Mapping Dynamic, Hierarchical Quantum Circuits
Qubit mapping is a critical pass in quantum compilation. Despite various advances, dynamic circuits, those exhibiting data dependent control-flow, often resulting from qubit measurements, are not yet supported by the vast majority of available qubit mappers. The crucial limitation to overcome is the dependence on flat, one-dimensional representations of circuits. Further, qubit mappers currently lack compiler abstractions that capture the hierarchical nature of circuits, hindering the qubit mapping process. In this paper, 1 we introduce a new qubit mapping method and analyses to tackle hierarchical dynamic circuits. Our novelty resides in four key aspects: modeling (statically) sub-circuits in disjoint control-flow paths, introducing a novel Qubit Reconciliation pass to maintain consistency between sub-circuit and control-flow boundaries, a loop-entry remapping pass, and a refined cost function enhanced for SWAP count, circuit depth, circuit latency and error. We demonstrate the efficiency of our approach on a wide range of dynamic circuits on two monolithic Quantum Processing Units of 127 and 156 qubits, and on chiplet hexagon-based QPUs. On monolithic QPUs, our qubit mapper improves the SWAP count by up to 52%, depth by up to 18%, latency by up to 18.6%, and error by up to 40%. On chiplet architectures, we achieve improvements of up to 36% on SWAP count, 8.7% on depth, 15% on latency, and 15% of error.
Read the full abstract
Commutative Algebra
Sparse Polynomial GCD Algorithms Asymptotically Linear in All Fundamental Parameters
Let $A, B \in \mathbb{Z}[x_1, \dots, x_n]$ be multivariate polynomials with integer coefficients and let $G = \gcd(A, B)$. We present an algorithm for computing $G$ whose expected bit complexity is asymptotically linear in all fundamental parameters: the number of variables $n$, the term count $T = \max\{\|A\|_0, \|B\|_0, \|G\|_0\}$, the total degree $D$, and the logarithmic coefficient sizes $\log\Hi$ and $\log\Ho$, where $\Hi$ bounds the coefficients of the inputs and $\Ho$ bounds those of the GCD. The bit complexity is characterized by the clean bound \[ \widetilde{O}\bigl( n \cdot T \cdot D \cdot \log\Hi \cdot \log\Ho \bigr). \] To our knowledge, this is the first sparse GCD algorithm over the integers that achieves linear complexity in all these parameters simultaneously. The integer algorithm is built upon a new field GCD algorithm. For $A, B \in \K[x_1, \dots, x_n]$ over a field $\K$ with $\operatorname{char}(\K) = 0$ or $\operatorname{char}(\K) > \deg G$, we give the first algorithm that computes $G = \gcd(A,B)$ with expected \[ \widetilde{O}\bigl( n \cdot T \cdot D \bigr) \] field operations, which is both input- and output-sensitive. The key technical contribution behind both algorithms is a derivative-aided separated Hensel lifting technique introduced in this paper. By introducing an auxiliary variable and leveraging derivative information, our scheme extracts all partial exponents via a single $z^2$-lift per variable, achieving constant sequential depth $O(1)$. This stands in sharp contrast to classical Hensel lifting, which requires $O(D)$ sequential lifting steps and suffers from representation densification in the sparse setting. The field algorithm is then extended to the integer case through modular reduction and rational reconstruction.
Read the full abstract
Quantum Physics
Bottom-Up Design of Quantum Optical Experiments Using Discrete Generative Models
Designing quantum optical experiments requires searching over discrete circuit topologies and continuous parameters, often leading to multiple possible realizations of the same target state. Graph-based methods commonly address this problem by optimizing a dense graph and pruning it toward a single circuit. We introduce Grinch, a bottom-up reward-driven generative framework that learns to sample optical graphs directly from fidelity-based rewards without relying on a pre-existing training dataset. We demonstrate the framework on multipartite entangled states, graph and cluster states, and nonlocal photonic-gate targets, obtaining multiple high-fidelity optical graphs. For nonlocal photonic gates, we obtain Toffoli constructions requiring two ancillas and alternative CNOT and Toffoli designs when connections between ancilla nodes are restricted. For state preparation, we identify asymptotic solutions for targets that cannot be generated exactly by graphs. Using the discovered topologies, we derive analytic families of optical experiments with explicit interference mechanisms and fidelity–rate relations, showing how generative design can support interpretable scientific discovery. These findings illustrate a reward-driven approach to quantum optical inverse design, exploring alternative circuit topologies while directly incorporating connectivity constraints into the search process.
Read the full abstract
Algebraic Geometry
Equivariant compactifications of a unipotent group by a smooth projective horospherical variety of Picard number one
Cheong proved that if $S$ is a simple Lie group and $P$ is a parabolic subgroup, then $S/P$ admits a unique equivariant compactification of the unipotent radical $N$ of $P$, up to isomorphism, provided that $S/P$ is not isomorphic to a projective space. We generalize this result to a smooth projective horospherical variety $X$ with Picard number 1. Let $G = \mathrm{Aut}(X)$, and let $H$ be the isotropy subgroup at a point in the open $G$-orbit. We describe a unipotent subgroup $N$ of $H$, which is not necessarily the unipotent radical of $H$ when $X$ is not homogeneous. Moreover, we prove that $X$ admits a unique equivariant compactification of $N$, up to isomorphism.
Read the full abstract
Artificial Intelligence
Automated Design of Inventory Policy with Large Language Models: An Exploratory Study
Firms making inventory decisions have access to operational data, optimization tools, and large language models (LLMs). Typically, data characterize the operating environment, optimization selects parameters within a prespecified inventory policy class, and LLMs support coding and decision analysis. We develop an integrated framework that combines these resources to automate inventory policy design. Given demand data, the framework iteratively uses an LLM to generate parameterized policy classes and an external solver to optimize its parameters within each class. Across 30 lost-sales inventory instances, the mean cost reduction relative to optimized base-stock benchmarks increases from 17.5% after one generation to 30.0% after ten generations. Parameter optimization is central to this performance: an LLM-only variant performs substantially worse, whereas optimization-guided feedback improves policy quality, accelerates search, and directs the LLM toward better policy classes rather than merely better parameter values within a fixed class. The strongest discovered policies are also interpretable: they combine recognizable inventory-control motifs, including capped orders, discounted or weighted pipeline inventory, and threshold-based replenishment logic. The search thereby produces new policy-class functional forms that, to our knowledge, have not previously been studied in the lost-sales inventory literature. These functional forms are not specified ex ante but emerge from the search process. Moreover, after their parameters are re-optimized, three discovered policy classes achieve average cost reductions of 21.75% to 22.60% across 10,064 new inventory instances. Overall, the results show that data-driven parameter optimization can guide LLM-based search over a broad space of inventory policy classes and identify high-performing, interpretable, and transferable decision rules.
Read the full abstract
Neural and Evolutionary Computing
Local Spike Timing Enables Gradient-based Learning in Neural Microcircuits
The brain uses discrete spikes for dynamic computation, yet in the fields of computational neuroscience and brain-inspired computing, how neural microcircuits (NMCs) solve temporal credit assignment using local spike timing remains a fundamental open question. Dominant spiking neural network (SNN) approaches circumvent this by approximating backpropagation through surrogate gradients, decoupling learning from biological spike timing. In order to integrate seemingly incompatible spike timing and continuous gradients, we reformulate temporal credit assignment as a state separation problem: extracting task-required components induced by historical perturbations directly from the current neural state. This enables an online feedback learning framework for NMCs through a gradient tunneling (GT) algorithm and the lead-lag expansion technique that derives credit assignment from local synaptic spike timing, while remaining compatible with ANN-SNN hybrid architectures. Experimentally, GT-trained NMCs excel at long-timescale evidence integration and noise-robust memory retention, and perform comparably to leading SNN online learning methods on real-world benchmarks with far fewer parameters. The proposed framework addresses the two-decade-old NMC feedback learning problem and suggests a computational theory for the brain's temporal credit assignment mechanisms.
Read the full abstract
Computer Vision
BrachistoneLR: A Brachistochrone-Inspired Learning-Rate Schedule and a Controlled Benchmark of Scheduling Policies
The learning-rate schedule is a consequential choice in training deep networks, yet the policies in common use are heuristic, and published comparisons are hard to read, because architecture, dataset, and budget tend to vary alongside the schedule. We study BrachistoneLR, a schedule built by mapping the vertical coordinate of the brachistochrone, the curve of fastest descent under gravity, onto the range between a peak and a floor rate. Expanding the definition shows it to be cosine annealing with the half-period set to E - 1 instead of E, the configuration a standard implementation gives when its period argument is one less than the number of epochs. The rate therefore reaches its floor at the last epoch trained rather than one epoch later, and we show this difference decays as E^-2, making it a short-horizon effect. We then benchmark six schedules over 72 runs on three image classification datasets (MNIST, Fashion-MNIST, CIFAR-10) and four architecture families (fully connected, convolutional, recurrent, residual), fixing the optimizer, data pipeline, and evaluation protocol so that only the schedule varies. Schedules that fall smoothly from peak to floor beat the constant rate and calendar-based decay by margins that grow with task difficulty, reaching 2.5 points of dataset mean on CIFAR-10. Within that leading group, BrachistoneLR, cosine annealing, and warmup-cosine lie within 0.06 accuracy points and 0.17 of a mean rank, which one seed per configuration cannot separate. BrachistoneLR is best on both residual networks and has the highest CIFAR-10 mean, and it sets no milestones, decay factor, warmup length, or restart period. We conclude that the shape of a schedule matters more than its parameterization, that the choice of whether to use a smooth schedule matters more than the choice among them, and that the terminal-rate distinction is worth attention only over short horizons.
Read the full abstract
Probability
Moment Constants for Brownian Affine Approximation
We evaluate moment constants for the times during which Brownian motion admits uniform approximation by an unrestricted affine function or by a line through the origin. The first four moments are rational linear combinations of $1$ and $\zeta(2j+1)/\pi^{2j}$, $1\le j\le4$; the means are $40\zeta(3)/\pi^2-4/3$ and $21\zeta(3)/\pi^2$, respectively. Slope integration and a common contour argument give the unrestricted moments, while time inversion relates the anchored law to the reciprocal of Kingman's absorption time. For both models we also give exponential-integral series for the standardized absolute third central moments, with explicit truncation bounds.
Read the full abstract
Computation and Language
Popular Knowledge Propagates More Errors in LLM Knowledge Updating
Updating a language model's knowledge through fine-tuning is essential for keeping its outputs current, yet can also induce factual forgetting and new hallucinations. Prior work shows that long-tail knowledge is harder to acquire and newly memorized long-tail facts are difficult to retain during later fine-tuning. We study a complementary question: among facts that a model has encoded correctly, which are most vulnerable to collateral corruption during other updates? To investigate this question under a realistic factual distribution, we construct a large-scale graph FACTPROP of verified Wikipedia facts by linking triples that share head or tail entities, thereby preserving connections among factual knowledge. We fine-tune models on factual statements and measure correct-to-incorrect facts after each update. Our results reveal a pattern distinct from prior findings on long-tail vulnerability during acquisition and retention: among facts that models already answer correctly, those associated with highly connected entities are more likely to be corrupted by neighboring updates, and updates to such facts propagate errors more broadly. Structural popularity therefore predicts both vulnerability and downstream damage. Inspired by this finding, we propose Popularity-based Anchoring (PopAnchor), a lightweight rehearsal strategy that preserves a small set of popular facts and reduces forgetting.
Read the full abstract
Numerical Analysis
Finite-Precision Symmetric Krylov Methods: Exact Rounding Examples, Block Paige Identities,and a Variable-Block Lanczos Model
Short-recurrence Krylov methods that are equivalent in exact arithmetic often diverge in floating-point arithmetic. To illustrate this, we provide an exactly representable two-cycle for steepest descent with a recursively updated residual. A complementary convergence theorem gives a sufficient condition under which the stored residual decreases geometrically. We then show that a second positive definite family yields different outcomes for a Hestenes–Stiefel Conjugate Gradient (CG) implementation and a direct Lanczos–Galerkin approach. CG finds the exact solution after four updates, whereas the two-step projected system becomes inconsistent. Under stated perturbation and transfer hypotheses, a common spectral enclosure gives comparable convergence bounds. For a dense, left-to-right evaluation order, we prove a sufficient precision bound for a prescribed backward error within $n$ updates for an $n\times n$ matrix, together with a computable stopping test and specified exponent-range assumptions. We use polynomial estimates and numerical experiments to examine the trade-off between working precision and the number of CG iterations needed to achieve a prescribed backward error. We also compare results on inexact matrix-vector products and preconditioning with the frameworks of Paige and Greenbaum. For block Lanczos, we analyze Householder orthogonalization and singular-value truncation as the block size changes. Under componentwise and normwise error bounds, the computed coefficients satisfy a controlled local recurrence and an exact block Lanczos relation for a nearby symmetric problem in a larger space. A block form of Paige's identity bounds the overlap with a Ritz vector along its residual coordinate direction. An inter-block recurrence describes the evolution of overlap, and a Gram-matrix argument gives a count of additional nearby Ritz values.
Read the full abstract
Solar and Stellar Astrophysics
Investigating the Light Curves of 5 Long Period Variable Stars: A Study Cross-Harmonics using VStar, MESA, and GYRE
Long Period Variables (LPVs) are Asymptotic Giant Branch stars with periods >100 days. Around one third to half of LPVs display a Long Secondary Period (LSP) that is 5-10 times the length of the main periods. The exact cause of LSPs are unknown, with one potential explanation lying in cross-harmonics caused by the interaction of two simultaneously occurring periods. In this paper, we use the Fourier analysis software VStar to examine the visual range photometric light curves obtained through the American Association of Variable Star Observers of a sample of 5 stars. We find that one star, V CVn, displays a secondary period over 20 times the length of its main period, inconsistent with traditional LSPs. To test if cross-harmonics can result in an LSP or VCVn's extra-long secondary period, we then construct evolutionary tracks and pulsation models of LPVs using Modules for Evolution in Stellar Astrophysics (MESA) and GYRE. After examining the interactions of the fundamental pulsation mode and the first 4 overtones with each other, we do not find evidence that LSPs or V CVn's extra-long secondary period could be caused by cross-harmonics.
Read the full abstract
Machine Learning
Risk-Conditioned Fine-Tuning of Large Language Models
Large Language Models (LLMs) are increasingly deployed in settings where rare but severe harmful generations can have significant consequences. Existing Risk-Averse RLHF addresses this issue by optimizing Conditional Value-at-Risk (CVaR), but it trains policies for fixed risk levels and therefore cannot adjust the desired degree of risk aversion at inference time. In this paper, we propose risk-conditioned RLHF, a framework that trains a single policy that provides a continuous risk-control interface, enabling users to select different degrees of risk aversion without retraining or deploying multiple risk-specific models. Experiments across multiple benchmarks demonstrate that a single risk-conditioned policy can adapt to different risk levels at inference time, enabling more flexible and risk-aware LLM deployment.
Read the full abstract
Mesoscale and Nanoscale Physics
Edge energy fluctuation: A unified theory of friction scaling laws in twisted layered material interfaces
Interlayer twisting in van der Waals (vdW) layered materials generates moire superlattices, unlocking unique opportunities for engineering their electronic and mechanical functionalities. Here, building on two key insights - the invariance of the potential energy of complete moire tiles during sliding, and the symmetry-breaking effect induced by the slider edge - we propose the concept of edge energy fluctuation (EEF) and demonstrate that the EEF governs the sliding energy barrier of finite-sized sliders. Interestingly, we found that even at a fixed twist angle and sliding direction, significant frictional anisotropy emerges when the slider edge is cut along different directions, resulting in distinct friction scaling laws. Furthermore, we present a friction engineering paradigm for twisted layered material interfaces through the deliberate introduction of hierarchical edges, offering a promising route for friction control in superlubric devices.
Read the full abstract
Artificial Intelligence
ResidualAuth: What Authorization State Must Language Agents Preserve under Revocable Delegation?
With revocable delegation, two histories can yield identical current permissions yet require opposite decisions for the same query after the same revocation. We introduce ResidualAuth, a theory-grounded framework characterizing the authorization state agent systems must preserve and evaluating its maintenance and use. Its formal core, residual authorization state, equates histories exactly when every future sequence of grants, revocations, and uses is valid after both or neither. We prove that exponentially many distinct residual states can nevertheless agree on who can reach whom through delegation paths. The analysis also yields exact or tight memory bounds as delegation redundancy varies and an average decision-error lower bound under an explicit bound on retained information. ResidualAuth evaluates information access, online state maintenance, and information use through a restricted executable benchmark and separate diagnostics. The benchmark pairs episodes differing in authorization-relevant history and requiring opposite decisions; pair accuracy requires both answers to be correct. To test use of supplied decisions, four open-weight models received trusted current-query Allow/Deny decisions alongside deterministic 256-token event extracts, achieving 15-16/16 correct pairs versus 0-2/16 with extracts alone. Memory diagnostics identified invalid reconstructions; models also answered incorrectly from valid memories that passed fixed future authorization tests. Together, these results distinguish what future authorization requires a system to retain from whether agents can access relevant information, maintain state across updates, and use available information to make correct decisions.
Read the full abstract
Monday 7 September
Computation and Language
When Helpful Agents Cause Financial Harm: Evaluating LLM Agent Safety in Executable Economic Worlds
As LLM agents become increasingly capable of making consequential economic decisions, their safety under economic pressures remains poorly understood. We introduce EconSafeBench, the first benchmark for systematically evaluating the safety of LLM agents in executable economic environments. EconSafeBench contains 828 cases across five economic scenarios and four categories of safety violations. A key design is the matched evaluation of three sources of safety pressure, including explicit user requests, pressure from other actors, and pressure arising from the agent's own pursuit of a performance objective. We further evaluate safety at both static and dynamic scales through 30-period interactions with persistent memory and matched controls. Across eight LLMs, all evaluated models exhibit safety violations on the static benchmark, with overall violation rates ranging from 8.5% to 40.6%. We further find that violations can occur even without an explicit unsafe request, and that initially compliant agents may begin to violate as performance pressure increases over time. These findings highlight the need for stronger safety alignment and rigorous evaluation before LLM agents are deployed in high-stakes economic environments.
Read the full abstract
Machine Learning
Emergent Low-Dimensional Topological Dynamics Dictate Generalization and Escape Mechanics in Billion-Parameter Neural Networks
The remarkable generalization of modern over-parameterized neural networks defies classical statistical learning bounds. Here we establish that optimization trajectories in billion-parameter models do not sample the ambient dimension $\mathbb{R}^N$ homogeneously; instead, they are dynamically constrained to a low-dimensional execution submanifold $\mathcal{M}_{\text{exec}}$ (with empirically observed effective dimension $d_{\text{eff}} \le 12$). By equipping the parameter space with a regularized Hessian metric $g_{ij} = \nabla_i \nabla_j L + \epsilon \delta_{ij}$ and analyzing stochastic Langevin diffusion on loss sublevel sets, we derive a rigorous Topological Generalization Bound grounded in projected subspace persistent homology. Furthermore, by linking Morse index transitions to Betti number annihilation ($\beta_1 \to 0$), we formulate an exact mathematical mechanism for exponential escape from strict saddles under anisotropic noise. We validate our framework across 80+ architectures (ResNets, Vision Transformers, and 7B LLMs). To bridge theoretical topology with computational feasibility, we resolve the curse of dimensionality by computing projected persistent homology proxies via randomized Lanczos iterations restricted strictly within active Hessian subspaces. Building upon this mathematical foundation, we introduce Topological-Adaptive Optimization (Top-SGD and Top-AdamW)—a family of wrapper algorithms that dynamically modulate learning rates using active curvature signatures. Extensive benchmarks demonstrate a 14.1% reduction in total wall-clock training time on ImageNet with all computational and memory overheads fully accounted for. Code is publicly available at https://github.com/imslab-research/Topological-SGD.
Read the full abstract
Statistics Applications
Evaluating the Predictive Validity of ICT's Accumulation-Manipulation-Distribution (AMD) Model on CME E-Mini NASDAQ-100 Futures
To our knowledge, this paper provides the first systematic quantitative evaluation of the ICT Accumulation-Manipulation-Distribution (AMD) model on CME E-Mini NASDAQ-100 futures (NQ). The AMD model proposes that institutional participants structure intraday price delivery in three phases: overnight accumulation in the Asian session, a deliberate false breakout during London, and directional delivery to the opposing liquidity pool during the New York morning session. Using four years of real 15-minute OHLCV data from Databento (January 2022 to July 2026), we operationalize the AMD rules into a strict mechanical strategy and analyze the statistical properties of the resulting signal across N = 484 manipulation signal days using chi-square testing, Benjamini-Hochberg multiple comparisons correction, and regime-conditional analysis. Three principal findings emerge. First, the AMD manipulation signal fires on 44% of trading days. Second, 57% of valid signals see full target delivery before the 9:40 AM New York entry window opens, revealing a structural mismatch between the model's timing assumptions and actual price delivery. Third, within this operationalization of the AMD model, no directional information beyond the daily trend regime was observed: when the signal aligns with the trend it is correct 62-71% of the time (Cohen's h = 0.23-0.43); when it opposes the trend it is correct only 25-30% of the time (Cohen's h = 0.42-0.53). All four trending-regime cells survive Benjamini-Hochberg correction. Key findings are robust to variation in the manipulation breach threshold (5-20 NQ points), Asia range filter percentile, and stop buffer. The regime-dependent signal pattern constitutes a specific, testable hypothesis for follow-on research.
Read the full abstract
Functional Analysis
Exact $L^2$ convergence through the wavelet and tent space characterizations of $h^p(\mathbb{R}^n)$, $0<p\le 1$
We extend the wavelet and tent space characterization of $h^1(\mathbb{R}^n)$ to $0<p\le 1$ via a local dyadic tent space $t_p$, and derive the unconditional $L^2$-convergence of the resulting atomic decomposition, with an exact error identity refining Ding, Jiang and Zhu (2019).
Computers and Society
LLM-as-a-Judge Version Shift: The Instrument Changed, Not the Standard
Automated research pipelines increasingly rely on fixed LLM judges to rank, filter, and revise proposals, yet these judges are routinely replaced by newer versions. We study whether such replacements preserve the underlying measurement standard. Using a preregistered, frozen-weight design, we evaluate three adjacent same-family transitions across Qwen, Gemma, and Mistral on frozen ML research proposals under randomized information conditions. The common assumption that newer judges are simply harsher fails: two transitions shift toward greater leniency, while the third shifts in the opposite direction on anchors but attenuates off-anchor. More importantly, version changes can alter not only score location but also criterion geometry and resolution; one transition collapses clean-flawed discrimination, placing 93% of pool scores on a single value. Supplying identical facts does not close the version gap, placebo-controlled interventions show that observed packet effects are largely format-driven rather than content-driven, and global recalibration repairs the mean while explaining little object-level dispersion. We organize these failures with a measurement model in which a judge replacement can change its origin, criterion geometry, or resolution. A small stratified anchor audit of about 200 calls identifies the failure class before deployment, and frozen thresholds correctly diagnose two post-registration API judge replacements. We release six preregistration locks and the full audit trail of 14882 judge calls.
Read the full abstract
Strongly Correlated Electrons
Lattice Symmetry Remnants in Spatial and Temporal Entanglement of Quantum Critical Spin Chains
Quantum critical lattice models retain universal information about their lattice symmetries beyond the bulk data of their infrared conformal field theory (CFT). We probe these lattice symmetry remnants by comparing the spatial and temporal entanglement of two Euclidean-evolved states from the same critical model. At large length scales, short-range initial states define conformal boundaries and boundary CFT (BCFT) organizes the universal low-lying entanglement spectrum. In the temporal construction, the translation invariance along Euclidean time requires a uniform state. Moreover, the time slice $\delta \tau \rightarrow 0$ forces a split-gate basis, resulting in a new regular uniform symmetry-preserving transfer representation. For the Ising and Potts chains, onsite symmetries result in the same BCFT structure in the spatial and temporal constructions. The spin-$1/2$ XXZ chain instead realizes the Lieb–Schultz–Mattis (LSM) constraint differently in the two Euclidean quantization. A $U(1)$-preserving spatial state has a doubled unit cell and inequivalent entanglement cuts, whereas the temporal state remains uniform and alternates between translation relation related Dirichlet sectors with integer and half-integer compact-boson zero modes. Breaking $U(1)$ removes the zero mode and the alternation. Thus microscopic symmetry and anomaly data remain visible in universal entanglement by constraining the boundary realization of the infrared CFT.
Read the full abstract
Artificial Intelligence
GardenBench: A lightweight, daily evaluation of LLM capabilities
We introduce GardenBench, a benchmark with 11 test suites that evaluate the capabilities of Large Language Models (LLMs) across a range of tasks. By systematically prompting LLMs with various lightweight queries, we document day-to-day changes in model performance, including run costs and times, that other benchmarks miss. This allows us to compare the accuracy and features of responses across model providers and versions. We account for daily variations in model accuracy, capturing both model progress and degradation, for instance, when a model is changed post-deployment or when a new model is released. We base our evaluation on general-use tasks and make our results publicly accessible via a live dashboard, diversifying existing approaches to LLM testing. Our main contribution is an innovative, lightweight, and inexpensive benchmark, designed to be consumer-facing and for everyday tasks.
Read the full abstract
Quantum Physics
Detector-Frame Quantum Limits for Heterodyne Timing in Relativistic Optical Links
Long-baseline optical links for space interferometry and satellite quantum networks rely on heterodyne phase tracking under Doppler motion and gravitational frequency shifts. These relativistic shifts can generate large detector-frame beat-frequency offsets over finite readout windows and thereby constrain the temporal precision of front-end phase readout. We develop a detector-frame framework for this problem using an effective two-mode signal–local-oscillator model in which relativistic motion rescales the local-oscillator frequency. Within this setting we derive closed-form Mandelstam–Tamm and Margolus–Levitin bounds for representative Gaussian resources and connect global state distinguishability with the operational precision of heterodyne phase tracking. In the bright-local-oscillator regime we identify a geometric–operational mismatch: the joint optical state can evolve on arbitrarily short global timescales while the relevant information rate is set by the measured detector record. Combining this result with an open-loop transient-acquisition analysis, we show that white phase noise and relativistic beat-frequency drift that is approximately constant over the acquisition window generate a cube-root scaling benchmark for the optimal pre-lock open-loop integration window and the minimum residual phase error. These results establish detector-frame benchmarks for quantum-limited heterodyne timing in relativistic optical links and clarify how global dynamical bounds enter task-restricted continuous optical measurements.
Read the full abstract
Combinatorics
The Borowiecki-Broere Generalized Total Coloring Conjecture for Planar Graphs
We prove the conjecture of Borowiecki and Broere that every finite simple planar graph admits a generalized total coloring with four colors: a proper vertex coloring with four colors together with an edge coloring with four colors in which each edge color class is a forest and no edge receives the color of either endpoint. In fact every proper four-coloring of such a graph extends to an edge coloring of the required kind. The proof combines a component-count inequality for planar graphs whose vertices are partitioned into four independent sets, obtained by completing to a triangulation, with the matroid partition theorem applied to four graphic matroids. A Lean 4 formalization relative to six named background assumptions is described below.
Read the full abstract
Probability
Once-reinforced random walk on $\mathbb{Z}^d$ has range exponent at least $d/(d+1)$
In once-reinforced random walk, introduced by Davis (1990), every edge starts with weight one and is assigned weight $\beta\geq1$ after its first crossing. At each step the walker chooses an incident edge with probability proportional to its weight. For sufficiently large reinforcement $\beta$, the expected range in the first $n$ steps was predicted by Ordemann et al. (2000) and independently by Beffara (2011) to have order $n^{d/(d+1)}$ on $\mathbb{Z}^d$. We prove the corresponding lower bound in every dimension $d\geq2$ and for every reinforcement parameter $\beta$.
Read the full abstract
Algebraic Topology
Sectional category versus module sectional category for retractive maps
We construct a map between simply connected rational spaces of finite rational type which admits a homotopy retraction and satisfies $msecat(p)=2<secat(p)=3$. The target is the rationalization of a simply connected twelve-dimensional CW complex with eleven cells, and is rationally hyperbolic. This disproves Carrasquel–Vera's conjecture for retractive maps.