Papers

Evidence that pricing algorithms collude comes from markets in which every seller runs the same algorithm. We ask what happens when they do not. Four language models from four providers, each at its cheapest tier, price in a four-firm logit Bertrand market without communication, in every homogeneous, two-by-two and fully mixed composi- tion (11 cells, 20 runs, 200 periods). Collusion is a property of the model: Claude and Gemini markets reach 72 to 79 percent of the monopoly rent, DeepSeek markets 24 percent, and GPT markets none, though GPT prices drift above the monopoly level rather than toward competition. Mixing does not reduce collusion by itself. Markets containing Gemini, which opens at the highest price and settles highest, are more col- lusive than the homogeneous markets they are built from; markets containing Claude, which opens lower and follows its rivals down, are less so; and only the fully mixed market is significantly less collusive than the average homogeneous one. Stability de- pends on the least stable participant: two GPT firms suffice to keep any market from converging. Inside mixed markets the rent is shared in a transitive order, DeepSeek over Claude over Gemini over GPT, that inverts the anchor ranking. The model that raises the price collects the least of the rent, as the price-leadership model predicts.

Traditional digital pen interfaces use menu buttons to change the pen mode, which results in time and cognitive load spent on round-trip interactions and mode errors from tapping small mode selection buttons. This work presents the Magic Pen, a technique which uses machine learning to automatically switch between digital pen modes without requiring explicit mode changes. Magic Pen is driven by an LSTM model trained on pen data collected from 27 participants across two studies and uses transfer learning to iteratively tune the model towards how a specific user annotates. Error mitigation techniques using a flick gesture or on-screen tap are incorporated to correct mode errors or remove a stroke quickly. We evaluated Magic Pen in a comparative study with 18 participants, followed by iterative improvements and a deployment study with 8 participants. Magic Pen was preferred compared to a conventional menu-based approach, and transfer learning allowed for greater model predictability and stability.

We give provable classical subexponential algorithms for secret recovery in growing parameter families associated with NIST third-round lattice candidates. For the Kyber/ML-KEM, FrodoKEM, SABER, NTRU LPRime, and Dilithium/ML-DSA families studied here, polynomial moduli and polylogarithmic coefficient scales yield recovery of the short secret component in expected time and space $2^{(1/2+o(1))n/\ln\ln n}$. For noisy or rounded linear relations, we exploit an exact gap in the squared Euclidean norm of a comparison vector defined by each coordinate guess. One Gaussian list suffices to identify every secret coordinate by binary search, without enumerating the others. We establish the required sampling guarantees through new geometric bounds for structured public operators over prime and power of two moduli. The construction builds on the Wagner-style Gaussian sampling framework of Ducas, Engelberts, and Loyer (CRYPTO 2025) and the low-error decision-LWE algorithm of Han, Gao, and Hu (2026). For quotient relations of NTRU type, we develop an affine slice search: fixing coordinates restricts candidate pairs to slices of the public lattice, and their estimated Gaussian masses guide the choice of each next coordinate. In the stated modulus window, Falcon's key generation quality condition supplies the required mass bound. The algorithm then recovers an equivalent signing key with high probability in time and space $2^{O(n/\ln\ln n)}$. The same search recovers the short key core for cyclic NTRU-HPS/HRSS. Together, these results give subexponential algorithms for problem families associated with all seven NIST third-round lattice candidates. Despite the subexponential complexity, our results do not establish a reduction in the concrete security of the currently specified parameter sets.

Large language model agents are being increasingly deployed as autonomous scientists, designing experiments and inferring mechanistic world models with minimal human oversight. Yet identifiability is often overlooked: when a plateau is reached, the agent needs to know whether it is not yet capable enough or the model simply is not identifiable from the data, in which case no amount of further experimentation of the same kind can help. We propose the Identifiability-Driven Experimental Agent (LLM-IDEA) for closed-loop discovery with an identifiability engine that returns a three-way plateau verdict: capability limit, resolvable within the design class, or certified exhausted. On ODEBench, 60 of the 62 systems with free constants are identifiable at round 0; the RC circuit is certified exhausted for every experiment that protocol can run, and a harvesting model is resolvable by one added initial condition. The identifiability engine reproduces known verdicts on Lotka-Volterra, Van der Pol, Lorenz, and a pharmacokinetic model, where it recommends the intravenous arm pharmacologists use, and it ranks the depth scorer of our own benchmark last among four observation designs. On the DiscoverPhysics benchmark, it finds two public worlds whose explanation rubric rewards a distinction no legal experiment can make, and every model there with accurate trajectories failed the explanation grade (15 of 15, against 5 of 9 in identifiable worlds, p = 0.012). On the Alien Universe, a two-body testbed we propose in which a force law switches between a provably non-identifiable and an identifiable protocol, LLM-IDEA on the identifiable protocol reaches discovery depth at least three on 8/8 seeds versus 1/8 without it. An autonomous discovery agent can thus compute, rather than guess, whether a plateau calls for more search, a better experiment of the same kind, or a different kind of experiment.

This study presents the Neuro-Memory Fuzzy Inference System (NeMeFIS), a hierarchical machine learning architecture that asymmetrically models acceleration and deceleration in car following behavior by integrating five human memory types procedural, working, episodic, semantic, and declarative. By linking external variables to memory functions via metaheuristics and validating them through factor and p-value analyses, NeMeFIS uncovers latent cognitive influences across Arterial, Collector, and Rural Highway corridors for different types of vehicles. Results from 54 different trained models emphasize cognitive thresholds shaped by driver perception limits and cognitive load. The trained NeMeFIS models outperform traditional statistical and conventional machine learning models in replicating realistic driving behavior, including comparisons with Linear Regression, ANFIS, and LSTM architectures. Fuzzy rule analysis reveals that declarative memory demands the highest rule, especially during deceleration, indicating complex braking decisions. Procedural memory drives acceleration, while semantic and declarative memory guide deceleration. Risk perception also emerges as a key factor, particularly on urban roads. Validated on both heterogeneous and homogeneous datasets, NeMeFIS offers a robust framework for modeling driver cognition. The findings support psychotherapeutic applications and the development of adaptive, human-like decision systems in Connected and Autonomous Vehicles (CAVs) to enhance traffic safety.

Vision-language models (VLMs) have demonstrated impressive capabilities but suffer from substantial computational overhead, as vision tokens dominate the input sequence. This motivates vision token compression as a key direction to alleviate the burden. However, with the emergence of hybrid architectures incorporating linear attention (\eg, Qwen3.5), prior methods designed for softmax attention struggle to generalize. Our analysis reveals that both attention- and similarity-based approaches suffer notable performance degradation, underscoring the urgent need for compression methods tailored to this regime. To this end, we propose V-CoLA, an efficient training-free token compression framework specifically designed for linear attention. V-CoLA introduces a novel uniqueness-aware importance criterion for identifying critical vision tokens, coupled with an adaptive token merging strategy that performs compression. All components are optimized at the implementation level to remain compatible with the chunk-wise parallelism of linear attention, ensuring strong practical value. Extensive experiments across multiple benchmarks demonstrate the superiority of V-CoLA: it achieves 99.5% of the original performance with only 50.0% of vision tokens, and over 88.0% with as few as 12.5%, while delivering a 1.86$\times$ to 6.15$\times$ prefill speedup. Our code is available at https://github.com/alibaba/EfficientRL.

Continuous structural connectivity describes the connection between any two points of the cortical surface by an intensity function. Based on the endpoints of tractography streamlines, existing methods estimate this function mostly by kernel smoothing, which only gives a point estimate. We propose a Bayesian framework that estimates continuous connectivity with uncertainty quantification. The endpoints of tractography streamlines, mapped to two spheres, are modeled by a log-Gaussian Cox process on the product of the spheres with a Matérn Gaussian process prior. With the finite element representation of the prior, the Laplace approximation of the posterior enables the computation on a fine mesh with over 400,000 latent variables per hemisphere pair. Numerical experiments show the accuracy of the point estimate of this approach and the coverage of its credible intervals. In the Adolescent Brain Cognitive Development Study, the proposed method predicts held-out streamlines better than kernel smoothing.

Testing equality of two high-dimensional covariance matrices is challenging when many entries differ only slightly. Dependence among sample covariance entries can also affect the finite-sample size and power of tests that aggregate their differences. We propose a studentized $\ell_2$-type statistic that averages squared, marginally standardized differences between corresponding entries of the two sample covariance matrices. Before constructing the statistic, we apply the same nonsingular linear transformation to both samples. In the transformed coordinates, the covariance-equality hypothesis is unchanged, whereas the standardized differences and correlations among their estimators generally change. The transformation can therefore incorporate structural or scientific information without reducing dimension. We approximate the null distribution using a Gaussian multiplier bootstrap that uses coordinatewise studentization and avoids forming or inverting the full covariance matrix of the vectorized sample covariance entries. For deterministic transformations, we derive nonasymptotic Gaussian and bootstrap approximation bounds and establish asymptotic size validity and power consistency. We also show that replacing a population transformation by a same-sample estimator leaves the statistic and bootstrap critical value asymptotically unchanged in relative terms under an operator-norm convergence condition. Simulations show that suitable transformations improve size accuracy and power under weak dependence. An analysis of breast cancer gene-expression data illustrates the method.

Large-scale, diverse datasets have driven the success of LLMs and VLMs. But VLAs for robotics remain limited by the cost and complexity of real-world data collection. While simulation offers a scalable alternative, its potential for sim-to-real VLA learning in mobile manipulation remains largely underexplored. We introduce SimVLA, an end-to-end framework that trains VLAs entirely on synthetic simulation data without teleoperation for mobile manipulation. SimVLA is first pre-trained on two complementary simulation-derived datasets: SimAction, a large-scale robot action dataset spanning 35 diverse mobile manipulation tasks, generated by composing atomic skills, and SimVQA, which leverages privileged simulator state to provide spatial, geometric, and subtask-level visual-language supervision. We further post-train SimVLA on a mixture of SimAction and SimDeploy, a dataset collected from policy rollouts across diverse simulated environments. We evaluate SimVLA on tasks including restocking, pouring, and cleaning, and show zero-shot transfer to real-world mobile manipulation, including real home environments. SimVLA outperforms policies trained on 50 in-domain real-world demonstrations, suggesting that simulation can enable scalable sim-to-real mobile manipulation. We further demonstrate the value of multiple complementary forms of supervision for effectively leveraging simulation in VLA training.

On-policy distillation (OPD) enables effective capability transfer between language models, yet the mechanisms underlying its failures are not fully understood. Across code generation and mathematical reasoning, OPD with larger-scale teachers exhibits early loss plateaus, with an average final loss reduction of 25.1% after 200 updates, compared with 96.2% for self-RL teachers, obtained by further reinforcement learning (RL) training of the initial student. To understand this difference, we analyze OPD as an idealized continuous-time dynamical system in the small-learning-rate limit. Our training-log diagnostics associate these plateaus with an early decline in a gradient-based learning-signal proxy while substantial loss remains; these measurements do not establish why the underlying gradient weakens. We further prove a local recovery guarantee for teachers sufficiently close to the initial student in a shared parameterization under regularity conditions, offering a conditional explanation for the success of self-RL teachers in our experiments. Across runs with and without loss plateaus, we observe small relative parameter changes (0.025-0.098%) and high similarity between the student's representations before and after OPD (linear CKA $>0.98$ across layers). These observations suggest that limited representation adaptation may contribute to learning-signal collapse, a hypothesis that remains to be tested. Code is available at https://github.com/leizhao7/opd-learning-signals.

The Crystallography Open Database (COD) contains over half a million experimentally determined crystal structures. Validation of crystallographic data proceeds at three hierarchical levels, the third of which assesses the physical plausibility of structures. This level has typically relied on empirical heuristics or expert judgment, which are slow and error-prone given the size of crystallographic databases and the rate at which new structures are deposited. Here we assess whether universal machine-learning interatomic potentials (MLIPs) can automate this level of validation. Of the 522 086 COD entries, 317 457 (60.8%) passed a data-quality filter removing disordered structures, formula mismatches and unmodelled solvent. These structures were relaxed with M3GNet, and 314 153 (99.0%) converged without a large change in cell volume. For most entries the relaxed and deposited volumes are nearly identical. Convergence and volume change thus provide simple criteria for identifying relaxations that do not confirm the deposited structure. A manual inspection of 327 structures showed that no single descriptor threshold (energy, largest force, or atomic displacement) separates valid from invalid structures. A comparison of M3GNet, CHGNet, PET-MAD and PET-OAM on 524 selected structures revealed model-specific failures: M3GNet distorts cyclopentadienyl and other $\pi$ ligands, and both M3GNet and CHGNet distort thiophene and thiazole rings, whereas the PET models preserve these motifs. All models detected missing hydrogen atoms, and at least some detected spurious hydrogens, incorrectly assigned atom types, and two previously unreported coordinate errors. No single model detected all error types. MLIP relaxation is therefore a useful but model-dependent screening tool, and combining several models is preferable to relying on any single one.

KV-cache entries are stored before their future queries are known, but each decoding query needs precision in different places. We study this mismatch using separate budgets for retained bits and bits fetched per query. ReadKV stores each key and value in a progressive code whose prefixes support different reconstruction precisions. For each query, it allocates key-channel prefixes using the query, computes attention from the reconstructed keys, and then allocates value-token prefixes using that attention. Stored entries remain unchanged. Each stage optimizes a calibrated distortion objective under a fixed budget; we prove exact allocation under diminishing refinement gains and relate these objectives to attention-output error. We also exhibit a finite-dimensional attention family where query-dependent access strictly outperforms every query-independent reader at the same read budget, even with unrestricted competing encoders and decoders. Across six base models, reading four bits on average from an eight-bit cache increases C4 perplexity by at most 0.66%, using about one quarter of the logical reads and half the retained capacity of a 16-bit cache. It is consistently more accurate than storing and fully reading four bits at the same payload-read budget. Retaining more bits than each query fetches is aimed at long-context decoding, where the cache bytes moved per step, rather than the weights, dominate cost. Long-context question answering and retrieval on two instruction-tuned models provide additional quality evidence. On the tested 8K-token, batch-one, single-layer workload on an NVIDIA A10G, a restricted eight-bit ReadKV reader with a two-bit mean payload-read budget has 39% lower latency than the tested TurboQuant codec.

The topological Hall effect of skyrmions is a promising electrical probe of magnetic textures, but its small magnitude limits applications. In this work, we theoretically investigate the topological Hall effect induced by sparsely distributed skyrmions in the weak-coupling limit, focusing on how the size and shape of the skyrmion affect the Hall effect. To this end, we used a formalism based on angular-momentum eigenstates, which enables systematic analysis within reasonable computation time. We found that, in clean materials where the electron mean-free path is comparable to or larger than the skyrmion size, the Hall effect depends considerably on the size and shape of the skyrmion. In particular, the Hall effect is largest when $k_F\lambda\sim 1$, and its magnitude can be tuned by the domain-wall width of skyrmions. These findings provide a basis for investigating how the shape and size of skyrmions affect transport properties and demonstrate that the topological Hall effect is controllable by engineering these properties.

We develop a theory of bi-contact plugs on oriented $3$-manifolds to construct Anosov flows or, more generally, structurally stable nonsingular flows. For a hyperbolic plug with filling Morse-Smale boundary laminations, we prove that every strongly transverse gluing map can be isotoped through strongly transverse maps to identify a strongly adapted unoriented contact form. The resulting flow is hyperbolic and is Anosov when the quotient is closed. This answers a question of Béguin-Bonatti-Yu. The result also extends to gluing nonsingular partially hyperbolic flows. Anosov completion by reflection embeds each such hyperbolic plug in an Anosov flow on its oriented double. Further applications include a classification of structurally generic projectively Anosov flows without saddle periodic orbits and Morse-Smale examples with prescribed saddle count. Generalizing the construction of Bonatti-Bowden-Potrie, finite capping embeds any bi-contact plug in a closed projectively Anosov flow with the same topological entropy, preserving partial hyperbolicity for attracting plugs. We also construct flows with controlled invariant torus dynamics on torus bundles and on doubles of plugs with coorientable boundary bi-foliations. Finally, every cooriented partially hyperbolic bi-contact plug admits a Liouville structure on its thickening, with Liouville trajectories projecting to positively reparametrized flow lines. Repelling plugs yield four-dimensional Liouville domains, often with chaotic skeletons.

Maneuver detection of non-cooperative satellites is an essential task of space situational awareness. The interacting multiple model (IMM) filter has been applied to detect unannounced maneuvers, but the standard IMM assumes a fixed Markov transition probability matrix (TPM) that must be tuned without knowledge of the maneuver rate of the target. To avoid this tuning, adaptive-TPM approaches update the TPM online from statistics computed inside the IMM. However, these statistics already depend on the TPM through the mixing step, and this dependence forms a feedback loop. Under the dynamics mismatch of orbit tracking, the loop can keep the maneuver probability high in the absence of a maneuver or suppress its rise when a maneuver occurs. In this paper, the TPM is driven by the normalized innovation squared (NIS) of a reference coast filter that is not mixed with the IMM, and the NIS is mapped continuously to the coast-to-maneuver transition probability. The reference-driven IMM (RD-IMM) is evaluated with four space-based optical sensors tracking a geosynchronous target. RD-IMM detects small burns that the fixed-TPM IMM fails to detect, with few false declarations in maneuver-free runs, and avoids both failures of the closed-loop adaptation. Moreover, RD-IMM reduces the position error after a small in-track burn by more than 50% compared with the single-filter approaches.

Human body point clouds are a versatile representation for AI-enabled human sensing. However, existing methods using LiDAR, radar, and depth cameras suffer from inherent drawbacks in high cost, sparse reconstruction, and privacy concerns, etc. In this paper, we exploit low-cost thermal arrays and present TAP3D, the first system to reconstruct 3D human point clouds from body heat signatures, offering significant advantages in cost, density, human sensitivity, and privacy. To overcome major challenges in depth estimation, thermal interference, and multi-person separation, we propose a novel physics-informed design, which integrates a forward thermal physics model with two distinct modules: multi-primitive estimation for self-supervised joint recovery of depth and other thermal properties, and geometric perspective fusion for suppressing interference and disentangling multiple people. We implement TAP3D using a single commodity thermal array sensor and build a large-scale dataset (160K samples, 8 environments, 11 users) for evaluation. TAP3D achieves remarkable accuracy for dense point cloud generation, enabling downstream tasks like fall detection (91.46%), indoor tracking (21.86 cm MAE), and human mesh recovery (4.87 cm error). By transforming body heat into point clouds for the first time, TAP3D pioneers a new paradigm for privacy-first, fully passive human sensing for many applications. TAP3D is open-sourced at https://github.com/aiot-lab/TAP3D.

Extending recent works on the restricted divisor function over integers, we consider an analogue of this function over the function field. We provide some exact and asymptotic formulas related to this function.

Hermite and Laguerre $\beta$-ensembles are important and well studied models in random matrix theory, with the special cases $\beta=1,2,4$ corresponding to classical matrix ensembles. Although sharp moderate-deviation estimates have been established for the largest eigenvalues, substantially less is known for other edge eigenvalues. Even for the limiting distributions $TW_\beta^{(k)}$, the leading constants in the tail asymptotics are not known for general $\beta$ and $k\ge2$. In this paper, for each fixed $k\ge2$, we prove matching upper and lower moderate deviation bounds for the right tail of the $k$-th largest eigenvalue, with the optimal exponential constant $2\beta k/3$, for general $\beta$. We also establish sharp lower tail estimates for the second largest eigenvalue of the Gaussian and Laguerre orthogonal ensembles, with exponential constant $1/24$, same as for the largest eigenvalue. For $TW_\beta^{(k)}$, we obtain optimal upper and lower tail asymptotics for general $\beta$ and $k\geq 1$. We also obtain new stochastic domination results for the edge eigenvalues. Our proofs combine variational arguments from tridiagonal matrix models, last passage percolation, stochastic comparisons of sums of edge eigenvalues, and superposition-decimation identities.

Diffusion models have achieved remarkable success in image generation, yet tracing their outputs to individual training examples remains challenging. Existing attribution methods often compress factor-specific effects into scalar responses, making distinct internal changes indistinguishable. This is particularly limiting for diffusion models, where semantic factors emerge through evolving representation dynamics during denoising. We therefore reformulate diffusion data attribution as attributing factor-induced internal response trajectories. In this paper, we propose a novel Concept Attribution method through Dynamic Trajectories(CADT). We argue that attribution should therefore ask not only which examples matter, but also how their influence unfolds during generation. Specifically, we construct matched counterfactual pairs at identical noisy states to isolate factor-specific representation displacements, and model their directional and magnitude evolution across denoising as dynamic attribution signatures. For each training example and generated query, CADT extracts stage-wise feature vectors and integrates them along the denoising process to form a trajectory descriptor. Applying the same construction across the training set yields a bank of factor-specific trajectory descriptors. The covariance statistics of this bank are then used to construct . CADT uses this covariance-aware positive-semidefinite kernel to calibrate the query and training representations, and compares the calibrated query trajectory with each training trajectory to produce the final training-sample attribution scores. Experiments on multiple public datasets show consistent improvements over existing diffusion attribution baselines across hierarchical, compositional, and style attribution.

Group dance generation aims to synthesize coordinated multi-dancer choreography from music, with broad applications in animation and interactive content creation. This task requires modeling dense inter-person dependencies to ensure spatial coordination, while naturally preserving individual dancer identities. Existing approaches model all dancers jointly with end-to-end transformers, which tie the architecture to a fixed group size and entangle per-dancer identities across frames. We propose ChainDance, a scalable framework that reformulates group dance generation as a Chain-of-Dancers: a sequential decomposition over per-dancer conditional distributions, allowing a single model to scale across variable group sizes without retraining and naturally preserving per-dancer identity. Built on a frozen single-dancer diffusion backbone, ChainDance introduces two lightweight modules: a Role-Aware Text Encoder (RATE) for per-dancer semantic conditioning, and a Group-Aware Motion Encoder (GAME) that aggregates previously generated dancers via a distance-weighted graph convolutional network, and incorporates a training-free noise optimization procedure at inference time to enforce global spatial coherence. Experiments on AIOZ-GDance demonstrate that ChainDance achieves state-of-the-art motion quality and group coordination while structurally preserving per-dancer identity, with $3$-$4\times$ fewer parameters and requiring $3$-$6\times$ less training time compared to prior approaches.

Low-cost sensors enable scalable indoor air quality monitoring but require calibration because of nonlinear distortions, noise, and temporal drift. The conventional strict pairwise calibration setting requires a co-located reference sensor at each deployment location and does not account for spatial and temporal heterogeneity. To address these limitations, we introduce a six-month dataset comprising multivariate indoor air-quality measurements from low-cost and reference sensors with contextual metadata collected at five locations. Using this dataset, we define four evaluation scenarios. The reference-efficient and location-transfer scenarios evaluate spatial generalization, whereas the long-term drift and event-conditioned scenarios assess robustness to gradual and abrupt distribution shifts. Based on these scenarios, we derive design requirements and propose a lightweight temporal model that combines input-window compression with residual temporal and feature fusion. Experiments show strong calibration performance across all four scenarios with low edge-inference cost.

Conditional mutual information (CMI) probes spatial non-Markovianity and plays a central role in a recently developed framework for defining mixed-state phases based on local reversibility. This framework sharply distinguishes exponentially decaying from algebraically decaying CMI, since the latter obstructs equivalence to a state with finite Markov length. Here we show that states that flow to the same renormalization-group (RG) fixed point can nevertheless exhibit qualitatively different CMI. Starting from a critical Gibbs state of a finite-range local classical Hamiltonian, whose CMI vanishes beyond the interaction range, we show that an RG-irrelevant perturbation that breaks detailed balance generically generates algebraically decaying CMI. Crucially, these power-law tails are cutoff-suppressed: although they lead to an infinite Markov length at every fixed lattice cutoff, they vanish as a positive power of the cutoff in the continuum limit. We construct solvable models that exhibit this phenomenon and predict that the Ising-symmetric critical point of Toom's cellular automata generically exhibits such behavior. We also find analogous cutoff-suppressed CMI in the high-temperature paramagnetic phase of the long-range Ising paramagnet, in contrast with the genuine power-law CMI at the finite-temperature critical point in this same model that survives the continuum limit. We also derive a general result that in one spatial dimension, CMI decays faster than the inverse square of the buffer length if and only if the long-distance physics is Markovian.

We study a hybrid system of coupled cavities, with or without two-level atoms doped inside. By employing a building-block construction facilitated by nonlinear Stark coupling, we identify a manifold of exact many-body eigenstates that remain analytically tractable across arbitrary lattice geometries and atom distributions. We demonstrate that these exact states exhibit a dual physical nature depending on their spectral positioning: when embedded within the chaotic many-body spectrum, they manifest as quantum scars that violate the eigenstate thermalization hypothesis and undergo weak ergodicity breaking; conversely, as ground states within fixed-total-excitation sectors, they exhibit off-diagonal long-range order and form polariton condensates. Furthermore, numerical calculations of local excitation-number fluctuations suggest that negative Stark coupling shifts the Mott-insulator to superfluid transition toward weaker inter-cavity hopping. Our findings provide a versatile framework for engineering both coherent quantum phases and non-thermalizing states in diverse light-matter architectures.

A core challenge in short-video fact-checking is identifying which evidence is sufficient to support a verification conclusion. Existing approaches either give the verifier all available evidence, introducing noise, or select evidence by topical relevance, which conflates relatedness with sufficiency. We identify evidential sufficiency as the selection criterion: whether a subset of evidence is adequate to support a confident verdict without redundancy. We introduce MiniVer-V, a benchmark of 195 short videos with three-way verdict annotations (supported, refuted, insufficient) and 5,510 multimodal evidence units spanning visual keyframes, speech transcripts, and web-retrieved external sources. We propose a two-layer verification framework that separates claim-video consistency, assessed from internal evidence, from factual verdict determination, which additionally requires external corroboration. On top of it, a sufficiency-driven greedy search assembles evidence until a sufficiency threshold is met and outputs insufficient when the candidate pool is exhausted, rather than forcing a verdict. With Claude Sonnet 4, the method reaches a Macro-F1 of 0.510 using 4.5 evidence units on average (16% of the full evidence set), statistically indistinguishable from the full-evidence baseline (0.518 with 27.7 units), while significantly improving recognition of insufficient cases over the same search without abstention. The efficiency result replicates with GPT-5.5 and holds only partially with an open-weight Qwen2.5-72B verifier. Ablations show that external evidence is indispensable for factual determination, while internal video evidence grounds the verdict in claim-video consistency. These findings suggest that evidence-efficient verification is achievable, and that explicit abstention is needed when evidence is genuinely inadequate.

Kikuchi lines and bands in reflection high-energy electron diffraction (RHEED) carry information on crystal geometry and electron scattering, but are often obscured by intense diffraction streaks. We combine multiscale convolution with spatial detail from skip connections and independent supervision that allows overlapping regions to localize streaks and candidate Kikuchi regions separately. Using manual polygon annotations made before model-assisted editing, the Kikuchi intersection-over-union (IoU) on 29 laboratory test images from separate growth batches was $0.6290 \pm 0.0137$. We then performed supervised adaptation to public chalcogenide images and stratified images by Kikuchi visibility using the median ratings of three observers who rated the images separately, two of whom rated with predictions hidden. With all 244 training images, IoU for the clear-feature group of 10 test images was $0.5014 \pm 0.0336$, whereas ambiguous images gave lower values. With 100 training images, clear-group IoU was $0.5171 \pm 0.0073$; expanding the training set also reduced responses on images without target features. Reported uncertainties are sample standard deviations across four laboratory runs or three adaptation runs. Public-image adaptation used reference masks of mixed provenance, including prediction-derived drafts. The method converts visual cues into inspectable spatial regions, providing a basis for analysis of line positions, intersections, and local intensity.

Small vision-language models may be able to read external evidence yet struggle to obtain it. We introduce Harness Compilation (HC), an offline procedure that adapts the division of work between a frozen small VLM and its external harness. A large teacher uses student execution traces to revise reusable content and control, while a separate validation set selects the deployed harness. Deployment requires neither weight updates nor teacher calls. Across seven visual question-answering settings with students of at most 9B parameters, HC improves scores over bare students by 9.9-23.9 points, averaged over three independent builds per setting. Interventions on five runtime decision types (invocation, selection, argument generation, evidence integration and abstention) show why this allocation matters: requesting evidence and generating open queries can be costly, whereas bounded choices and reading supplied text can remain useful student work. Fact cards benefit all ten evaluated students, but decision policies transfer unevenly. Recompilation for a new student model helps when the transferred interface no longer fits the student. With 100 practice items, HC exceeds answer-only LoRA on three tasks. Larger training budgets can match or surpass a fixed harness, while combining the two improves SlideVQA beyond either alone. These findings support allocating work from measured student behavior rather than uniformly removing decisions.

Cross-modal vector retrieval is widely used in multimodal systems, such as search engines and vector databases. It typically operates in out-of-distribution (OOD) settings, where query vectors follow a distribution that differs from that of the vectors stored in the database. In such cases, conventional indexes suffer significant performance degradation, and even methods specially designed for OOD remain limited by inefficient use of query modal characteristics, restricted GPU parallelism, and inadequate support for dynamic updates. We present CORAL, a novel GPU-accelerated graph-based vector index for scalable cross-modal retrieval, featuring hierarchical memory management that spans GPU, CPU, and disk. Specifically, CORAL incrementally incorporates the characteristics of query modality and terminates index construction timely. Crucially, it introduces coverage-aware adaptive pruning to address the imbalanced coverage of the query vector's neighbors. Moreover, CORAL presents a fully neighborhood-aware projection approach to efficiently utilize GPUs for highly parallel index construction, and a targeted connectivity enhancement method to refine the index structure. Besides, CORAL also supports modal-semantics-based vector insertion and topology-repairing deletion that restore node connectivity. Experimental results demonstrate that CORAL outperforms existing methods with up to 1.6 times the throughput at matched recall while reducing construction time by up to 56%. Furthermore, it exhibits remarkable resilience under dynamic updates and remains effective at the billion scale.

Time series forecasting aims to predict future values from historical observations and auxiliary features. We propose SteerCast, a retrieval-based latent steering method that improves decoder-only forecaster at inference time, without updating its parameters. SteerCast constructs a database from the training set by storing a representation of each history window together with a steering vector computed in the forecaster's latent space, defined as the difference between representations induced by the ground-truth continuation and by the model's own prediction. At test time, SteerCast retrieves nearest neighbors for a query history, aggregates their steering vectors, and injects the resulting signal into the forecaster's hidden states at every step of autoregressive generation, guiding predictions toward trajectories consistent with similar training cases. Experiments across diverse multivariate benchmarks and multiple horizons show that SteerCast consistently improves forecasting accuracy over the fine-tuned backbone and retrieval-based baselines, while requiring no additional training beyond the original fine-tuning and using only the training set as a retrieval corpus.

Multimodal Large Language Models (MLLMs) have demonstrated strong potential for sequential recommendation through their ability to reason over complex multimodal data. However, existing approaches either rely solely on the target user's own interaction history, neglecting collaborative signals from neighboring users, or incur substantial computational overhead through repeated MLLM inference over long interaction histories. To address these challenges, we propose MGRASRec, a multimodal graph retrieval-augmented framework for sequential recommendation. MGRASRec injects collaborative filtering signals conditioned on the candidate item directly into the MLLM prompt by retrieving structured paths from a user-item interaction graph, extended via multimodal similarity to increase coverage beyond exact co-interaction overlap. This retrieval also surfaces the history items most relevant to the candidate at no additional cost, removing the need for recurrent summarization and keeping inference to a single forward pass per candidate. All components are unified into an augmented prompt for parameter-efficient fine-tuning of an MLLM. Extensive evaluations across three publicly available datasets validate the effectiveness of MGRASRec, achieving the best performance on all metrics with particularly strong gains in ranking quality.

We establish an exact Fourier restriction criterion for a class of weighted Moran measures. Assuming uniformly bounded one-level frame bounds, we prove that, for every \(s>2\), the measure itself already detects the full extension estimate: \(L^2(\mu)\to L^s(\mathbb R)\) extension holds if and only if \(\widehat\mu\in L^s(\mathbb R)\), equivalently, if and only if the one-level Fourier moments \(G_{s,n}\) are summable, where $G_{s,n}$ is the $s$th moment of the nonzero Fourier coefficients of the $n$th digit law. The sufficient direction is obtained from a finite-frame stability estimate and its multiscale iteration, while necessity follows from disjoint resonant frequency grids. This gives, in particular, an exact formula for the critical restriction exponent and a precise criterion for endpoint attainment. The criterion reveals that the optimal restriction range is not determined by standard Fourier or dimensional data. We construct singular continuous spectral Moran measures with the same scales, digit cardinalities, exponential spectrum, and maximal nonzero Fourier coefficients. They also satisfy the same global Fourier decay bound, which is sharp along the same sequence of resonant frequencies, yet their critical restriction exponents are different. More generally, after fixing the summability rate of the maximal Fourier coefficients, we determine the optimal interval of possible restriction exponents and show that, at every interior critical exponent, the same fixed data allow both attainment and failure of the critical endpoint. These measures may simultaneously have $ \dim_{\mathrm F}^{\theta}\mu=\theta$ and $ D_q(\mu)=1,$for all admissible $\theta$ and $q$, and pointwise local dimension one everywhere on their supports.