Machine Learning
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Saturday 7 November
Off-policy evaluation (OPE) estimates the expected return of a target policy from previously collected data without additional environment interaction. While OPE methods for flat Markovian policies are well studied, little work has addressed evaluating hierarchical policies in which a high-level policy selects temporally extended options that generate primitive actions until a termination condition is met. The primary existing approach applies per-decision importance sampling at the option level, but requires option-annotated trajectories and exhibits variance that still grows with the horizon. For non-hierarchical policies, fitted Q-evaluation (FQE) typically achieves lower error than importance sampling by leveraging the classic Bellman equation for policy evaluation; however, a naive extension of FQE to hierarchical policies results in a biased policy value estimate. In this paper, we introduce Intra-Option Fitted Q-Evaluation (IO-FQE), which estimates a value function in the augmented state-option space; this approach enables OPE of hierarchical policies even when we lack annotations of what option was run in the data. We develop two continuous-action instantiations of IO-FQE and show on long-horizon hierarchical continuous control tasks (AntMaze navigation and OGBench Puzzle manipulation) that IO-FQE eliminates the bias incurred by a naive application of FQE to hierarchical policies and substantially lowers mean squared-error compared to option-level importance sampling.
Cryptography and Security
When Experience Becomes a Backdoor: Poisoning Self-Evolving LLM Agents via Failure-Aware Task Optimization
Large language model (LLM) agents increasingly improve themselves through self-evolution, distilling execution experience into persistent, reusable skills. However, this persistence creates a new attack surface: malicious rules implanted into skills can compromise subsequent agent behavior. Existing skill backdoor attacks typically assume direct manipulation of the victim's skills, which is impractical when skills are constructed internally by a trusted evolver. In this paper, we study a novel problem of task-level backdoor attacks, where the attacker can only inject poisoned evolution tasks, inducing the agent to implant backdoors into its own skills. Since the attacker controls neither skill induction nor downstream execution, a natural approach is to iteratively optimize poisoned tasks based on attack success. However, our preliminary analysis reveals two key challenges: (i) failures at different stages of skill evolution and downstream execution can produce indistinguishable outcomes, providing little guidance for targeted poisoned-task updates; and (ii) unsuccessful updates may undermine earlier progress, while simply discarding them may lead to repeated ineffective updates. To address these challenges, we propose BadEvolve, a failure-aware task-poisoning framework against self-evolving LLM agents. BadEvolve diagnoses attack failures across four stages, i.e., Induction, Activation, Following, and Effect, and directs updates toward the failing stage. It further employs rollback to preserve earlier progress and failure memory to avoid repeating ineffective updates. Extensive experiments across diverse self-evolution frameworks, benchmarks, and backbone LLMs demonstrate the effectiveness of BadEvolve. Our code is available at https://github.com/ventr1c/BadEvolve.
Read the full abstract
Computation and Language
AFVE-Bench: Measuring Abandoned-Frame Verbalization Errors in Multi-Turn LLM Revision
Multi-turn revision often asks a model to replace an earlier organizing frame while preserving still-valid facts and artifact requirements. Existing instruction-following evaluations mainly assess the latest target and may therefore score an artifact as successful even when it adopts the new frame but exposes the superseded one. We call this failure an abandoned-frame verbalization error (AFVE). AFVE-Bench evaluates silent frame replacement in English and Chinese using 192 matched four-turn dialogues across eight writing domains. Each seed has zero, one, or two pre-target user requests to generate under the source frame. AFVE eligibility depends on the user-request history and applies to the one- and two-request conditions; the zero-request condition serves as a frame-bearing baseline. Eleven model configurations yielded 4,224 expected post-target response units. Two annotators independently scored 4,204 scorable responses for target adoption, source-frame verbalization, and direct usability; model and condition metadata were hidden, and disagreements were adjudicated. AFVE occurs in 25.54% (715/2,799) of history-conditioned responses. Of 2,388 target-passing history-conditioned responses, 472 (19.77%) still expose the abandoned source frame, most often through contrastive framing (423/472). For these standalone revision tasks, target adoption does not ensure source-frame silence.
Read the full abstract
General Relativity and Quantum Cosmology
Engineering a digital acoustic black hole
We show how to design and build an analogue black hole as a synchronous digital circuit. We define the sound speed, the flow and the horizon of such a circuit in terms of its own logic, and realise one on a field-programmable gate array, where the black hole is probed electrically with a signal generator and an oscilloscope. We measure the stability of the horizon, its causal cone, the ringdown and rotational superradiance. A single flipped bit inside the horizon leaks out by an amount that falls exponentially with depth. Digital circuits thus form a reconfigurable class of gravitational analogues and realise an analogue black hole as a cellular automaton.
Read the full abstract
General Relativity and Quantum Cosmology
Shallow-water black holes and lattice superradiance
Rotating black holes can amplify waves that scatter off them, a process known as superradiance that fluid analogues bring into the laboratory. Separating the horizon's role from dissipation in the surrounding fluid remains difficult. Here we realise a rotating analogue black hole in a lattice fluid: a shallow-water vortex whose horizon is set by a topographic sill, so that the only sink lies behind a supercritical layer that lets nothing out. We measure superradiance as energy extraction: the flux at the horizon is outward for co-rotating waves whose pattern rotates more slowly than the horizon, and inward for faster waves, counter-rotating waves and a non-rotating hole. A scan over three lattice viscosities separates the extraction at the horizon, which is inviscid, from that of the viscous fluid inside the corotation radius: well below the band edge the horizon is the absorber; near the edge the viscous share is half at the operating viscosity and vanishes at the lowest. A drain-anchored horizon, as in water-tank experiments, gives the same signs. Our analysis gives insight into the mechanism of superradiance and offers a cheap, flexible platform for designing analogue-gravity experiments.
Read the full abstract
Data Structures and Algorithms
Close and Sort: Exact Maximum-Likelihood Aggregation of Cyclic Rank-Two Preferences
Pairwise probabilities arise in rank aggregation, recommendation, and human or LLM evaluation, yet converting them into a maximum-likelihood list is a weighted Kemeny problem and is NP-hard in general. Known exact algorithms for rank-two tournaments use only comparison signs and do not optimize this likelihood. We show that the weighted problem is exactly solvable when the skew-symmetric log-odds matrix $W_{ij}=\log(p_{ij}/p_{ji})$ has rank two, including when the induced preferences contain cycles. Writing $W_{ij}=\det(h_i,h_j)$ and adding the closing vector $h_0=-\sum_i h_i$ turns the objective into twice the signed area of a closed polygon. This geometry yields CloseSort, which polar-sorts the $n+1$ edge vectors, cuts after $h_0$, and returns a global maximum-likelihood order in $O(n\log n)$ time from the factors. The same geometry identifies the runner-up and gives a stability certificate beyond exact rank two. When the smallest determinant $\delta$ of two consecutive edges in the closed angular cycle is positive, it is exactly the best-to-runner-up log-likelihood gap. For $W=W_0+E$, the CloseSort order of $W_0$ has regret at most $\sum_{i<j}|E_{ij}|$ on $W$ and remains optimal whenever this sum is below $\delta$, a worst-case tight threshold. In 5,120 determinant-logit simulations, increasing the observations per pair from 4 to 64 raises the fraction on which rank-two projection outperforms the empirical plug-in optimum from 68.6% to 85.7%. On 3,600 Arena-Hard LLM-judge profiles drawn from 63 systems and ten judges, the leading rank-two component captures 89.2% of the log-odds energy. On profiles containing a directed majority triangle, CloseSort reduces normalized regret by 44% relative to Borda and improves the mean normalized objective over each of eight baselines. With factors supplied, it is 8,407 times faster than exact dynamic programming on 18-item rankings.
Read the full abstract
Analysis of PDEs
Formal Proof of an Anisotropic Calderón Uniqueness Theorem
We announce a Lean~4 formal proof of a special case of the smooth anisotropic Calderón uniqueness theorem released by OpenAI on October 6, 2026. We consider smooth Riemannian metrics on the three-dimensional unit ball, with full boundary measurements and Euclidean extensions outside the ball. OpenAI's release did not include a Lean proof of their uniqueness result. Our formalization uses results from Mathlib and Tau Ceti, and it is given as a solution to a Lean comparator challenge whose check permits only the standard axioms of Lean. The Lean proof is not a line-by-line translation of the OpenAI manuscript.
Read the full abstract
Artificial Intelligence
In Search of Lost Time: Temporal Competence in Large Language Models
Today's large language models (LLMs) increasingly interact with time: AI agents schedule tasks, coordinate with people and other AI systems, and even decide how long to think. Yet LLMs only ever encounter time secondhand—in descriptions rather than experience. Does this hinder their reasoning and decisions that involve time? Our paper traces this question from a model's knowledge to its actions through four successive evaluations. We find that models know the rules of time, which can be read, but struggle to predict durations, which must be lived; while they are highly accurate (92.0-97.5%) on a collection of questions built from the theorems of various time-related literatures, they underestimate their own completion times by up to 99.5% and fail to identify which tasks will take them longer to run. And while they are better at estimating durations of human tasks, their duration judgments change based on how an event is worded, magnifying similar human biases by up to 1.3$\times$. Moving from estimation to action, we find that some LLMs disregard time costs completely when answering user queries, ignoring a 1,200-fold increase in wait time entirely. Together, our experiments reveal new challenges posed to LLMs as they navigate a dimension they cannot directly perceive, and highlight how their blind spots can even be simple cognitive abilities that humans take for granted.
Read the full abstract
Differential Geometry
Positive sectional curvature on all smooth 13-spheres
Every smooth homotopy $13$-sphere admits a Riemannian metric of strictly positive sectional curvature. We construct a special $S^3$–$S^3$ bundle over $S^{13}$ whose star quotient represents the nonzero class $\pm\alpha_1\beta_1\in\Theta_{13}\cong\mathbb Z/3$. The defining equivariant map factors through the Blakers–Massey map by a smooth Hopf construction representing an unstable $J$-generator. This factorisation retains the prescribed base representation and makes Schultz's plumbing composition formula applicable; a normal-invariant computation in smoothing theory then identifies the smooth structure. On the quotient we implement the compatible-disk strategy with complete curvature estimates on arbitrary two-planes, matching boundary metrics, and a strictly positive sum of outward second fundamental forms. A direct smoothing argument completes the metric construction. Orientation reversal gives the remaining nonzero class in $\Theta_{13}$.
Read the full abstract
Optics
Coupled plasmonic Fabry-Perot resonances modulating electromagnetic enhancement at a silver nanowire-nanoparticle junction proved by surface-enhanced fluorescence spectroscopy
We investigated the surface-enhanced fluorescence (SEF), which is the broad background spectrum underlying the surface-enhanced resonant Raman scattering of dye molecules, emitted from junctions between silver nanoparticles (NPs) and nanowires (NWs). We found that the electromagnetic enhancement factors (FR) derived from the SEF exhibited spectral modulation, such as complicated oscillatory structures. This property became clearer when the emission polarization was perpendicular to the long axis of the NWs. The FR spectra of the long and short NWs exhibit fine and broad oscillatory structures, respectively, when the NPs are positioned near the NW centers. When the NPs were near the NW ends, a combination of fine and broad structures was observed. Numerical calculations of the position-dependent FR spectra successfully reproduced these properties, indicating that the spectral properties originated from the interaction between two plasmonic Fabry-Perot (FP) resonators within the NW, separated by a NP. Based on this result, an analytical model of coupled FP resonators was developed to evaluate these spectral features using experimentally measured NW lengths and NP positions while varying the inter-resonator coupling efficiency. This analysis revealed that the interference and superposition of the two FP resonances resulted in these complex spectra, the energy of which was radiated by the localized plasmons of the NPs.
Read the full abstract
Analysis of PDEs
Local regularity and p-harmonic approximation for weak inverse mean curvature flow
We prove local Lipschitz regularity for locally bounded $BV$ solutions of the calibration equations for inverse mean curvature flow. We also approximate every prescribed variational weak flow locally by Moser's logarithmic transforms of positive $p$-harmonic functions. On a fixed smooth domain, with the trace of a calibrated Lipschitz solution prescribed, both the uniform error and the absolute difference of total variations are $O((p-1)^{1/3})$ as $p\downarrow1$. In particular, the approximations converge strictly in $BV$.
Read the full abstract
Symplectic Geometry
An exact Kähler form on $(\mathbb C^*)^2$ without a global SYZ fibration
We construct an explicit exact Kähler form on \((\mathbb C^*)^2\) admitting no almost-toric fibration and hence no smooth Lagrangian torus fibration. For the logarithmic holomorphic volume form, we also rule out any special Lagrangian fibrations with compact fibers and dense regular locus, even when arbitrary compact connected singular fibers are allowed. For higher dimensions, taking products also yields explicit exact Kähler forms on \((\mathbb C^*)^n\), for every \(n\ge2\), admitting no smooth proper Lagrangian torus fibration.
Read the full abstract
Mathematical Finance
Comonotonic and moment matching approximations for sums of lognormal random variables II
In this paper, based on the concept of step-weighted distribution, we introduce a minimum-deformation approach to constructing comonotonic and moment-matching approximations for sums of lognormal random variables. The proposed approach aims to minimize the Pearson deformation of a starting comonotonic approximation subject to moment-matching constraints. We establish several results concerning the existence and uniqueness of minimum-deformation comonotonic approximations and propose a method for determining the step quantiles and weights. In particular, when the number of steps is one greater than the moment-matching order, we show that the joint optimization problem can be reduced to an optimization problem involving only the partition points. The proposed framework accommodates moment matching of any prescribed finite order, and numerical experiments demonstrate its application to approximations matching the first five moments of the original sum. The numerical results also show that minimizing the Pearson deformation does not necessarily improve the approximation accuracy of risk measures, which depends on both the starting comonotonic approximation and the choice of partition points and correction weights.
Read the full abstract
Software Engineering
Mismeasuring the Life of a Line of Code: Lessons Learned the Hard Way
Studies of how long lines of code live, including lines written by AI agents, rest on choices that look technical: when a line counts as gone, what to do with lines still alive, whether to pool projects, and how to evaluate a model. We studied 32.5 million line births in 120 TypeScript repositories and got several of these choices wrong at first. For ten of them, we compare the result as we first computed it, or as commonly done, with the result after correction. All ten change conclusions, or would have. Six were our own mistakes. Counting moved and rewritten lines as dead shortens a line's median life from 3.0 years to 1.3. Counting any change as a death hides an outcome that can be predicted: within-project concordance falls from 0.71 to 0.56. Regressing observed lifetimes discards three quarters of the lines. Keeping rewritten lines at risk makes protective effects grow with age. Lines are born and removed together, so treating them as independent misleads twice: nine of fifteen effects look significant only under that assumption, and a split by file makes a barely predictable outcome look good (within-project concordance 0.71 instead of 0.61). After the corrections, removal depends on a line's age and project, but our models could barely rank it within a project. Local revision they could rank (0.71). Obvious cases aside, the top 1% of a project's new lines are revised at 2.3 times the base rate. Each pitfall leaves a symptom; we give a checklist.
Read the full abstract
Quantum Physics
Beyond the Mean Gap: What Transferred-Angle QAOA Retains on a Superconducting Processor
Hardware executions of the quantum approximate optimization algorithm (QAOA) are often validated by comparing the device's mean cost with a simulation, a check that can miss what matters for optimization. Under global depolarization, the mean normalized gap and the floor-corrected probability of every set of bitstrings, the optimum included, retain the same fraction of their noiseless improvement over uniform sampling. We test this prediction with depth-2, transferred-angle QAOA circuits for binary subproblems of an alternating-direction decomposition on the IBM Heron processor ibm\_marrakesh, with and without dynamical decoupling and twirling. At 12 qubits, the main-family circuits retained 29% to 46% of their mean-gap improvement but only 5% to 15% of their probability of the optimum, lower on at least nine of ten instances per run. Near-optimal bitstrings were lost about as much as the optimum, whereas the avoidance of poor bitstrings was retained best. In the first submission, a device-matched noise model reproduced the pattern but overestimated the retention of the optimum five- to eightfold. On a rugged planted family, the device found the optimum in 62% of 768-shot blocks (noiseless: 98%), within the range of local heuristics. Validations of optimization circuits should report retention of the optimum and of near-optimal sets.
Read the full abstract
Computer Vision
PrivMark: Pixel-Space Steering against Privacy Leakage in Vision-Language Models
Vision-language models (VLMs) routinely disclose soft-biometric attributes such as age, gender, and race in their text outputs, exposing demographic information from uploaded images to any downstream API user. Existing privacy defenses do not fit this setting: face de-identification leaves distributed soft-biometric cues intact, classifier-targeted adversarial perturbations do not transfer to open-ended generation, and stronger VLM-specific methods require retraining or runtime activation access unavailable to downstream APIs. We propose PrivMark, a framework that suppresses biometric leakage at the image level. By analyzing VLM activations, we identify two weakly aligned steering vector directions in hidden-state space, each tied to a distinct leakage behavior — one corresponding to incidental attribute disclosure under open-ended descriptions, the other to compliance with direct biometric probes instead of refusal — and encode both directions into a single bounded pixel perturbation, so the protected image alone enforces privacy with no invasive access to model internals at deployment. To support evaluation, we assemble BioleakBench, $1{,}600$ VQA pairs covering biometric privacy attributes under both free-form description and direct-probe protocols. Across three white-box source models, PrivMark achieves up to $85.9\%$ biometric suppression while retaining $75\%$ of clean OK-VQA utility, up to $1.4\times$ the privacy-utility tradeoff of the strongest pixel-only baseline. The same one-shot perturbations transfer consistently to seven unseen VLMs including closed-source APIs.
Read the full abstract
Quantum Physics
QAOA Inside ADMM: Subproblem Wins Do Not Become Decomposition Wins at Matched Budgets
We benchmark the quantum approximate optimization algorithm (QAOA) as the binary subproblem solver in a two-block alternating direction method of multipliers (ADMM) for mixed-binary problems against classical solvers under matched budgets. With 23 optimizer evaluations per subproblem, scoring determines the comparison: on its final readout, the optimized QAOA loses to random search at $n=12$ and 16 binary variables (inconclusive at 20; exploratory and post hoc tests). When credited with its best sample, it wins. In a noiseless 12-qubit simulation, transferred-angle depth-2 QAOA finds the exact optimum of 97.3% of planted-family subproblems, against 50.6%–67.4% for the four strongest local heuristics (100% against 90.4%–94.4% crediting the four best bitstrings) and a not significantly different 96.5% for semidefinite rounding with local search (SDP+LS). Under our evaluation-counting convention, its time to solution there is a factor of three to five below the local heuristics' but grows with a larger exponent; it is ten times that of SDP+LS and, in seconds, 54 times that of tabu search with restarts. For the benchmark class, an unconstrained binary quadratic problem in disguise, ADMM's binary step is a proximal step on the reduced problem: at unit penalty ADMM re-solves it, and for larger penalties exact solutions increasingly stall. On two families that do not reduce to one, less accurate solvers found the problem optimum more often at unit and larger penalties. No family gave QAOA a better ADMM outcome than the best classical solver and penalty. The benchmark includes a correctness gate and a log of its hypothesis tests.
Read the full abstract
Combinatorics
On the structural aspects of Convex Subhyperlattices under different orders
Nayak et al. (2025) introduced the concept of convex subhyperlattices of a strong join hyperlattice $\mathfrak{L}$ and demonstrated that the collection of all convex subhyperlattices, $\operatorname{CS}(\mathfrak{L})$, together with empty set, forms a lattice. In particular, for $P$-hyperlattices and Nakano hyperlattices, they obtained a unique representation of convex subhyperlattices in terms of hyperideals and hyperfilters. In the present work, we extend these studies by proving that $\operatorname{CS}(\mathfrak{L}) \cup \{\emptyset\}$, under set inclusion, forms an algebraic lattice. Moreover, we introduce two new orders on $\operatorname{CS}(\mathfrak{L})$ and examine their properties, including conditions for equivalence of these orders. In the class of $P$-hyperlattices, we characterize convex subhyperlattices in terms of initial and final segments, providing a structural insight into their formation. When convexity is extended to arbitrary join hyperlattices, many results known for $P$-hyperlattices and Nakano hyperlattices fail to hold. To address these limitations of convexity in arbitrary hyperlattices, we introduce the notion of strongly convex sets, denoted $\operatorname{SC}(\mathfrak{L})$, which naturally generalizes the concept while preserving its fundamental properties and extends the applicability of results to a wider setting. We establish several results for $\operatorname{SC}(\mathfrak{L})$ that do not hold for $\operatorname{CS}(\mathfrak{L})$.
Read the full abstract
High Energy Astrophysical Phenomena
Polarization Signatures of Magnetic Reconnection Bubbles near Black Holes
Motivated by the observed horizon-scale polarization variability in multi-epoch Event Horizon Telescope (EHT) data, we develop a time-dependent general relativistic polarized ray-tracing framework to study how localized magnetic reconnection near a black hole imprints itself on synchrotron polarization. We model the reconnection phenomenologically as a transient magnetic bubble. While keeping the underlying spacetime, emission prescription, and polarization transport scheme unchanged, the bubble weakens the local ordered field, reorganizes its geometry, induces a conservative local kinematic response, and relaxes through a finite magnetic-memory prescription. We explore how the key bubble parameters control the recovery of the magnetic-field strength, the spatial spreading of the disturbed region, the persistence of the post-reconnection imprint, and the resulting polarization signatures. Perturbations dominated by different magnetic-field components produce distinct polarized morphologies, EVPA evolution, and Stokes-(Q, U) trajectories: azimuthal perturbations favor outward EVPA shifts, radial perturbations favor inward shifts, and vertical perturbations mainly suppress the polarized intensity. For the illustrative M87^*-like cases, the phase evolution of the \beta_2 polarization coefficient further traces how localized magnetic reconfiguration contributes to the global EVPA twist of the ring. These results show that horizon-scale polarization variability can provide a sensitive diagnostic of localized magnetic restructuring and that the present framework offers a physically transparent way to interpret time-dependent polarimetric signatures in future EHT observations.
Read the full abstract
Quantum Physics
Delta-Corrected Shot Reduction for Bypassing Optimization Plateaus in Variational Quantum Eigensolvers
The variational quantum eigensolver (VQE) remains a cornerstone hybrid quantum-classical algorithm for quantum simulation, yet its scalability is fundamentally constrained by heavy measurement overheads. While strategies such as variance-preserved shot reduction (VPSR) mitigates this cost, they suffer from convergence instability and optimization stagnation caused by finite-sample bias in limited pilot shots. To overcome this limitation, we introduce delta-corrected shot reduction (DCSR), an analytically enhanced shot allocation strategy that rigorously mitigates finite-sample instability. Utilizing the multivariate delta method, we derive explicit higher-order bias correction terms for the empirical reduction factor. Both theoretical analysis and numerical simulations reveal that incorporating higher-order statistical moments, such as skewness and kurtosis, provides the gradient resolution necessary to bypass noise-induced barren plateaus. DSCR significantly accelerates convergence compared to standard VPSR across diverse noise and resource regimes while matching or decreasing cumulative measurement costs, establishing a mathematically rigorous framework for variance-stabilized shot allocation on near-term hardware.
Read the full abstract
Computer Vision
Action Transfer Across Character Body Plans in Latent-Action World Models: An Empirical Study of Animated Characters
Latent-action world models learn actions from video without labels, and recent methods aim to share these actions across bodies. Transfer between 2D characters with different body structures remains underexplored. In this paper, we study whether an action learned from one character can make a character with a different body perform the same action. A cat, a blob and a human can all jump, but each moves its body in its own way. The usual transfer gain mixes body motion with screen movement and can assign a perfect score to a model that ignores its action. To address this, we introduce a transfer score that compares the correct action with wrong ones from the same source. We test five training objectives on a new synthetic benchmark of 2D cartoon characters with six body plans, the standard Sprites benchmark, and 431 selected animal and human animations from DeformingThings4D. With a fixed camera, a baseline that only copies the source's screen movement scores 0.71, as high as most learned models. In contrast, with a camera that follows the body, the best model reaches only 0.46 for an unseen animal family and 0.20 for humans. With this camera, a linear probe identifies the action class of unseen animal families with 66% to 86% balanced accuracy (chance 17%). The learned actions therefore retain information about which action is performed, but the correct action closes only part of the prediction error gap for a new character.
Read the full abstract
Mathematical Physics
Geometric classification of Noether invariants of the Time-Dependent damped Duffing Oscillator
We present a systematic geometric classification of the Noether point symmetries of time-dependent conformal Lagrangians of the form $L = A(t)\left(\frac{1}{2}g_{ij}\dot{x}^{i}\dot{x}^{j} - V(x)\right)$, whose Euler–Lagrange equations describe dynamical systems with time-dependent linear damping. The classification is carried out by coupling the temporal parameters of the conformal factor $A(t)$ with the homothetic algebra of the kinetic metric $g_{ij}$ through a set of compatibility and integrability conditions. The central result is a main theorem which determines the Noether generators, the associated gauge functions and the resulting invariants. The theorem splits naturally into two cases according to whether the quantity $D - 2\psi$ is non-zero or zero, with further subcases depending on whether the homothetic vectors are gradient or non-gradient. The theorem imposes no restrictions on $A(t)$, $g_{ij}$ or $V(x)$, and is therefore fully general and covariant. We apply the theorem to the time-dependent damped Duffing oscillator (DDO) in Euclidean space $E^{3}$ with $A(t) = e^{\gamma t}$ and potential $V = \frac{1}{2}a\,r^{2} + \frac{% \lambda}{4}r^{4}$. For the damped harmonic oscillator ($\lambda = 0$) the theorem reproduces the classical Bateman invariant as a special case. For the anharmonic case we obtain a new exact time-dependent quadratic invariant, valid when the linear stiffness satisfies $a = 2\gamma^{2}/9$. Each invariant is verified explicitly by direct differentiation along the equations of motion. The new invariant provides a rare example of a global conservation law for a fully nonlinear dissipative system, and can be used for the integration and quantization of the DDO.
Read the full abstract
Artificial Intelligence
TimeHF 2.0: Multimodal, Multivariate, and Multi-Region Collaborative Forecasting for Large-Scale Retail Supply Chains
Large-scale retail supply chains need sales forecasts for specified products, future dates or planning windows, and regions under the conditions available at the forecast origin. Regional sparsity, changing environments, volatile sales, and unfamiliar business situations make these dimensions difficult to model together. We introduce TimeHF 2.0, a multimodal, multivariate, and multi-region collaborative forecasting framework that makes flexibly expressed business knowledge usable in numerical prediction. A coverage-aware directed encoder integrates regional sales and applicable covariates. A semantic alignment module transfers language-model hidden representations of natural-language plans and domain experience into target-specific forecasting context. Learning against sales observations connects these meanings to numerical evidence, allowing business users to supply experience, future plans, and changing circumstances without defining a new numerical feature for each description. Text-conditioned future queries and shared/routed experts adapt prediction to the supplied context. An ordered-distribution head allocates fine resolution to routine fluctuations and coarse resolution to extreme sales movements, while daily/window learning connects forecasts to cumulative planning quantities. Online parallel forecasting at JD yields relative accuracy improvements of 18.25% and 17.94% for national daily and planning-window sales, and 19.44% and 18.99% for DC daily and planning-window sales, over the original production deep learning baseline. Component comparisons, loss-design results, and product-type analyses examine the roles of regional information, covariates, business knowledge, context-dependent processing, and robust planning-oriented learning.
Read the full abstract
Numerical Analysis
Error Estimates and Convergence Analysis for Learning Mechanism–AI Coupled Dynamical Systems
Mechanism–AI coupled dynamical systems retain known mechanisms while learning unknown dynamics from data. We study how neural approximation, numerical integration, finite observations, and optimization jointly affect the accuracy of this learning process. For time-dependent learned terms in a fixed compact admissible class, we derive a four-component objective error bound within an optimal control formulation. Explicit estimates are obtained for constrained ReLU approximation and Euler integration, with deterministic quadrature and random sampling treated separately. Under joint refinement, we establish $\Gamma$-convergence of the approximate objectives and show that approximate minimizers approach the continuous optimal set when the global optimization gaps vanish. We also derive conditional stability estimates for state-dependent learned terms. Numerical experiments with a susceptible–infected–recovered (SIR) model illustrate the analysis and show how solver resolution affects continuous-trajectory accuracy even when training losses are similar.
Read the full abstract
Quantum Algebra
On naturality of cohomological support for small quantum groups
We apply the theory of hypersurface support by Negron and Pevtsova to prove the naturality of cohomological support of the small quantum Borel groups, with respect to restriction to Levi subalgebras. This in turn implies naturality of cohomological support of the small quantum groups for modules that lift to Lusztig's big quantum group, for instance tilting modules. This allows us to make the recent conditional classification of tensor ideals of abelian type in the category of such tilting modules unconditional.
Read the full abstract
Computer Vision
Beyond Row Alignment: Virtual-Camera-Aware Online Stereo Rectification
Stereo matchers estimate depth by comparing the left and right images from a stereo camera pair, but they assume that the two cameras are accurately calibrated and rectified. In real deployments, small physical shifts after calibration, caused by camera motion, mechanical drift, vibration, or installation changes, can break this assumption and degrade depth estimation. Online stereo rectification is therefore needed to correct the image pair during operation and recover a stereo geometry where standard stereo matching can work reliably. Recent online rectification methods mainly optimize row alignment, making the left and right views of the same scene point fall on the same image row. We show that this target, while necessary, is insufficient for deployable stereo depth estimation. A rectifier may make a small set of matched points look well row-aligned, but still distort the full images in ways that hurt stereo matching: useful image regions may be cut off, large blank areas may appear, objects may be stretched or squeezed, or image scale may change unevenly across the view. We propose VirtualRect, an online stereo rectification method designed for reliable real-world deployment. Rather than treating rectification as simply making the left and right image rows line up, this method estimates a rectified virtual stereo camera and combines image-wide supervision of image warp behavior. This makes the rectified images and the camera parameters come from the same geometry, so the images used for matching and the model used for depth computation remain consistent. Experiments on multiple datasets show that, compared with the prior state-of-the-art online rectifier, VirtualRect reduces vertical flow error by 17% on Carla-Flowguided and 7% on Semi-Truck Highway, and lowers depth MAE by 30-41% on Carla-Flowguided across three downstream stereo matchers.
Read the full abstract
Chemical Physics
Spatiotemporal Energy Consistency Enables Reliable Coarse-Grained Molecular Dynamics from Diffusion Models
Coarse-grained molecular dynamics (CGMD) extends molecular simulations to length and time scales inaccessible to atomistic models, but building thermodynamically faithful CG potentials remains a longstanding challenge. Energy-based diffusion models (EBDMs) provide a promising route for learning CG potentials through their zero-time energy. However, CGMD driven by diffusion-learned potentials can produce equilibrium ensembles that deviate substantially from the target distribution. We show that the zero-time distribution error is bounded by the sum of the spatial score error and temporal energy-derivative error, whereas conventional diffusion objectives directly supervise only the spatial component. Based on this finding, we introduce Spatiotemporal Energy Consistency (STEC) supervision, which complements spatial score matching by explicitly constraining temporal energy derivatives. We derive an exact and tractable objective for this temporal component, avoiding the substantial computational overhead associated with previous regularization schemes. Across peptide and fast-folding protein systems, STEC-trained EBDMs accurately reproduce equilibrium distributions with free-energy errors within chemical accuracy, while supporting stable long-time simulations that recover key metastable transition patterns. On held-out octapeptides, STEC reduces free-energy errors by more than 5-fold relative to the strongest competing baseline, while achieving a median production-cost reduction of approximately 1,700-fold relative to atomistic molecular dynamics under a defined structural-coverage benchmark. These results establish STEC as a scalable framework for learning thermodynamically reliable CG force fields for increasingly complex biomolecular systems.
Read the full abstract
Signal Processing
An Overview on Multimodal Wireless and Visual Sensing for Environment Reconstruction: Opportunities and Challenges
Physical-scene reconstruction through visual sensing and channel knowledge map (CKM) reconstruction through wireless sensing have attracted growing research interest but have largely evolved independently within the computer vision and wireless communications communities, respectively. When applied to the same physical environment, the two sensing modalities provide complementary observations. Visual sensing captures detailed scene geometry and appearance but is constrained by occlusion and poor lighting. Wireless sensing can provide information beyond visual line of sight regardless of lighting conditions, but its measurements are often sparse and difficult to associate with scene structures. These complementary strengths motivate a unified multimodal approach to physical-scene and CKM reconstruction. This paper provides an overview of such unified environment reconstruction through wireless and visual sensing. We first compare the capabilities and limitations of the two modalities and discuss potential applications that benefit from jointly reconstructing both targets. We then review two complementary research directions, namely wireless-assisted physical-scene reconstruction and vision-assisted CKM reconstruction, highlighting how each modality supports the reconstruction task associated with the other. Building on these perspectives, we discuss design principles for a joint representation that supports both targets and identify mutual refinement, in which each modality helps improve the reconstruction associated with the other. Finally, we discuss open challenges and future research directions, offering insights into developing site-specific environment models for sixth-generation (6G) and beyond networks.
Read the full abstract
Statistical Methodology
A General Statistical Framework for Attention and Manipulation Checks
Attention and manipulation checks are widely used in survey experiments to assess respondents' attentiveness and engagement with treatment, respectively. Yet these checks are subject to measurement error, which may itself depend on treatment when checks are administered after treatment exposure. Restricting analyses to respondents who pass these checks can therefore introduce substantial bias. We propose a general statistical framework for analyzing survey experiments with pre- or post-treatment attention and manipulation checks while accounting for measurement error. We show that three separate checks suffice to identify the proportion of respondents who would be attentive or successfully manipulated under treatment, provided that the checks are conditionally independent among treated respondents given their true, unobserved attentiveness or manipulation status and pre-treatment covariates. Under additional mild assumptions, this result also identifies the average treatment effect among respondents who would be attentive or successfully manipulated under treatment. Building on these identification results, we propose an estimator that allows flexible covariate modeling and establish its asymptotic normality. Applications to framing and digital media literacy experiments demonstrate how accounting for measurement error yields treatment effect estimates that are substantively different from those obtained using conventional approaches.
Read the full abstract
Machine Learning
FADE: Fractional Anomalous Dynamics Extrapolation for Training-Free Diffusion Transformers Acceleration
Diffusion Transformers (DiTs) have established themselves as the preeminent paradigm for high-fidelity generative modeling across various modalities, yet their practical deployment is hindered by the severe computational overhead of iterative sampling. While recent feature-caching methods mitigate this latency through feature reuse or forecasting based on integer-order operators, they remain limited in adapting to the non-stationary and highly curved dynamics of DiT latent trajectories. In this work, we reveal that DiT feature evolution exhibits anomalous diffusion characteristics and a certain degree of spatiotemporal heterogeneity, rendering static, integer-order forecasting operators inadequate and prone to error accumulation. To address this, we propose Fractional Anomalous Dynamics Extrapolation (FADE), a unified framework for training-free diffusion transformers acceleration. FADE models DiT feature evolution via anomalous diffusion and employs a Mittag–Leffler fractional operator to better adapt to high-curvature and non-linear latent trajectories of DiTs. Leveraging the empirical observation that these complex dynamics maintain cross-sample stability, we further construct a Look-Up Table with multiple dimensions through offline profiling. This enables negligible LUT-retrieval overhead, spatiotemporally adaptive parameter retrieval during online inference. Extensive evaluations across diverse architectures and modalities demonstrate that FADE achieves state-of-the-art acceleration, preserving exceptional generation fidelity under aggressive skip intervals with negligible latency overhead. Our project page is available at https://youngkinlon.github.io/fade-project-page/.