Papers

We present a straightforward but effective method for repurposing existing contact-rich dexterous manipulation demonstrations. Starting from inputs of hand and object trajectories, our method outputs high-quality nonlinear trajectory warps that account for intermediate waypoints, environmental barriers, temporal shifts, and varied start/end configurations. Foundational to our method is the utilization of contact distributions, which we show allows us to reliably compute complex and high-dimensional dexterous hand trajectories following a simple object-centric warp specification pipeline. We evaluate our method across 12 variations sourced from 4 demonstrations in a publicly available dataset of human hand motion data, perform baseline comparisons, and demonstrate generalization of our approach to different manipulators. Results and code will be made available on publication.

Recursive self-improvement (RSI) aims to achieve compounding gains by having models improve themselves. While most existing RSI systems optimize external agent harnesses or prompts around a frozen base model, data-centric RSI directly updates the model's own parameters by training on agent-generated data. However, because validating data strategies requires expensive model training, existing methods face a fundamental dilemma: a single agent gets trapped in narrow directions and lacks exploration breadth, while naive parallel search or heavy trace sharing sacrifices long-horizon search depth. To address this challenge, we introduce G"odel Forest, a multi-agent framework that organizes recursive self-improvement as an ensemble of co-evolving search trees. In G"odel Forest, each agent autonomously grows a persistent tree, deepening, branching, or pruning data strategies based on model feedback to secure depth, while parallel trees explore distinct regions of the data space to expand breadth. Crucially, rather than leaving trees isolated or flooding them with heavy execution logs, a dynamically co-evolving memory connects the forest: agents continuously distill their successes and failures into compact procedural lessons anchored to a global leaderboard. Through this forest ecosystem, a dead-end in one tree instantly warns the whole forest against unpromising paths, while an empirical breakthrough quickly seeds new exploration branches in neighboring trees. Evaluated on RSIBench-Data across six diverse domains, G"odel Forest outperforms the single-agent baseline by an average of 10.70% while reducing wall-clock time on five tasks. Ablations confirm that co-evolving shared memory yields a +7.00% gain over independent parallel search, demonstrating that collective distillation is key to scalable self-improvement. The code is available at https://github.com/evolvent-ai/Godel-Forest.

Let $G$ be a finite simple graph and let $A_c(G)$ be the Artinian algebra associated with its cover ideal. We prove that $A_c(G)$ has the WLP when $\tau(G)>|V(G)|/2$, where $\tau(G)$ denotes the size of a minimum vertex cover of $G$. As a consequence, $A_c(G)$ has the WLP with high probability when the Erdős-Rényi random graph model is considered. Moreover, we study the borderline case $\tau(G)=|V(G)|/2$ and as a result, classify the WLP for paths, cycles, Ferrers graphs, and well-covered trees.

Higher-order effects, such as microlensing parallax and lens orbital motion, are essential for characterizing the physical properties of microlensing planetary systems including their masses and distances. In practice, higher-order effects are often included in light-curve modeling only when their signals are strong, because their inclusion in weakly constrained cases tends to drive the inferred parameters toward regions of parameter space that are disfavored by standard Galactic models. This binary choice over physically continuous effects introduces an event-dependent selection function that complicates population-level interpretation and prevents strong and weak detections from being combined within a single uniform framework. In this work, we investigate the origin of instabilities associated with higher-order effects, and show that such instabilities can arise from a mismatch between commonly adopted uninformative priors in the light-curve parameter space and the distributions predicted by Galactic models. Motivated by this, we develop a new Bayesian framework that enables higher-order effects to be incorporated in a stable and uniform manner. As part of this framework, we introduce Galactic Prior Modeling Engine (gapmoe), a dedicated tool that enables efficient evaluation of the Galactic prior density and allows it to be incorporated directly into the light-curve inference. Using simulated events, we demonstrate that our framework robustly recovers the true physical parameters while fully accounting for microlensing parallax and lens orbital motion. This framework eliminates the need for ad-hoc model selection and provides a scalable and statistically consistent pathway for analyzing the large samples expected from next-generation surveys such as the Roman Space Telescope.

Efficient exploration in high-dimensional decision spaces remains a central challenge for decision-making systems. Humans, in contrast, can navigate large decision spaces with remarkable efficiency. Recent behavioral studies suggest that humans reduce dimensionality in large decision spaces by probing candidate feature dimensions, identifying reward-relevant ones, and restricting the effective decision space. Inspired by this mechanism, we propose TDGE (Task-Dimension-Guided Exploration), a human-inspired, model-agnostic algorithm with an automatically constructed task-dimension–feature–item hierarchy. TDGE follows a top-down exploration strategy: it first selects task-relevant feature dimensions, then identifies informative features within those dimensions, and finally recommends concrete items based on the selected features. Experiments on MovieLens-20M, LastFM, and Amazon recommendation datasets show that TDGE substantially improves exploration efficiency and cold-start adaptation over baseline algorithms. Comparisons with other structured algorithms and ablation studies attribute these gains to TDGE's hierarchical structure and semantic feature-space exploration, with robust results across clustering methods and hierarchy depths. Recommendation-trajectory visualizations also show exploration patterns similar to human dimension-guided behavior.

While the "Matthew Effect" and filter bubbles are widely recognized outcome-level biases in recommender systems, we reveal that Diffusion Recommender Models (DRMs) uniquely compound this issue through their generative dynamics. Rather than merely inheriting data imbalances, DRMs trigger a self-reinforcing amplification of popularity bias. We identify that this phenomenon is driven by two compounding mechanisms. First, while optimization loss is universally dominated by high-frequency items across recommenders, DRMs suffer from a unique structural prior mismatch during generation. Because the forward terminal distribution of long-tailed data deviates significantly from the standard Gaussian prior, reverse sampling trajectories inherently collapse toward high-density popular items, fundamentally suppressing niche item generation. To dismantle this self-reinforcing loop, we propose FairDiff, a plug-and-play fairness-aware diffusion framework. To overcome the popularity-dominated loss, we introduce Popularity Condition Guidance (PCG). Rather than altering the training objective, PCG acts as an inference-time distributional reweighting mechanism, mathematically reshaping the score-based gradient field to penalize high-popularity regions and guide trajectories toward niche semantics. Furthermore, we design a Semantic Calibration (SC) Module to bridge the prior mismatch, aligning the forward and reverse distributions via one-step optimal transport. Comprehensive evaluations demonstrate that FairDiff achieves state-of-the-art performance while effectively mitigating the self-reinforcing Matthew Effect, highlighting its value as a general framework for DRMs.

A critical prerequisite of generative recommendation is designing semantic identifiers (SIDs) that are both scalable to large item sets and efficiently learnable. Existing SID learning methods fundamentally rely on Top-1 hard assignment during vector quantization. While heuristic strategies – such as clustering-based initialization or forced post-hoc collision resolution – can artificially inflate codebook coverage, they often disrupt end-to-end semantic alignment and fail to address the underlying optimization bottleneck: sparse gradient propagation. In standard Top-1 assignment, gradients concentrate on a narrow subset of frequently selected codewords, leaving the majority inherently under-trained and causing severe SID collisions. To overcome this limitation natively without relying on complex initialization priors, we propose FineSID, a unified quantization framework that moves beyond Top-1 assignment by enabling fine-grained gradient propagation across the entire codebook. Instead of updating only a single selected codeword, FineSID distributes learning signals to all codewords in a soft, differentiable manner. This design promotes globally balanced codebook optimization while strictly preserving semantic consistency, effectively alleviating SID collisions and stabilizing training in large, high-dimensional codebooks. Extensive experiments on multiple public benchmarks demonstrate that FineSID is robust to initialization configurations and consistently improves both codebook utilization and recommendation accuracy. Our work provides a principled, initialization-agnostic solution for semantic identifier learning, advancing the practicality of generative recommendation.

Porous and perforated leading edges reduce turbulence-interaction noise, but predictions of the reduction have relied on material impedances adjusted to the acoustic data. This paper removes that adjustment. The finite-chord scattering problem is solved in compressible subsonic flow for an arbitrary chordwise admittance, with the shed wake, the Kutta condition and permeability-dependent edge singularities, and closed to the far field with the finite span and the array aperture represented. The admittance of a perforated insert is derived from the Rayleigh conductivity of one aperture with three computed corrections: plate thickness, interaction between neighbouring apertures, and Howe's grazing-flow blockage evaluated at the aperture-averaged velocity of the boundary layer. A bulk porous insert is represented by its measured permeability and pore-fluid inertia. For three perforates at four flow conditions the prediction agrees with measurements to 2.5 dB rms over 84 comparisons, against 4.4 dB with the unscaled impedance used previously. Although the Mach number is 0.05, the chord is acoustically non-compact, and an incompressible solution rises where the measurement falls. Two of three bulk porous inserts are predicted to within 1.7 dB rms where interaction noise dominates. Narrow, equally spaced peaks measured on perforated plates are reproduced by none of the admittances tested; their spacing implies a source convecting at 0.62 of the free-stream speed, absent from a frozen-gust model. A parameter study shows that the perforate geometry acts almost entirely through one low-frequency admittance group, in which the hole radius cancels for holes buried in the laminar boundary layer.

Polyak step size (PS) and Armijo line search (ALS) have received increasing attention in stochastic optimization, with encouraging empirical performance and theoretical guarantees. However, their convergence theory for stochastic heavy ball (SHB) methods remains limited. In this work, we develop a unified convergence analysis for SHB equipped with PS and ALS. To this end, we introduce a modified Armijo rule that closely parallels the Polyak step size, together with a decoupling analysis that isolates the historical dependence induced by momentum. For SHB with standard PS and ALS, we establish expected convergence for strongly convex, convex, and non-convex objectives without interpolation or restrictive conditions on the momentum parameter. Under interpolation or strong growth, we further strengthen the results to almost sure rates and last-iterate convergence. Moreover, for general settings beyond interpolation, we prove almost sure convergence to the exact optimum or to stationarity for SHB with diminishing variants of PS and ALS. These results provide a more comprehensive theoretical view of Polyak step size and Armijo line search for stochastic heavy ball methods.

A coalition that deviates once can profit many times when what it sells keeps working. In a threshold-encrypted mempool, a leading defense against maximal extractable value (MEV), a quorum of the decryption committee that sells its decryption capability to a front-runner exposes every later block that the capability still decrypts. We ask how large a penalty, such as slashable stake, deters this kind of collusion. In our repeated game, a single leak by any coalition in a monotone family of authorized coalitions (for example, any $k$ of the $n$ committee members) unlocks a set of future rounds, costs a one-time penalty, and ends the coalition's participation. We show that every dynamic deviation reduces to choosing a leak time, so deterrence holds if and only if each coalition's penalty covers the largest discounted value that a single leak reaches. Without discounting, over $T$ rounds of unit value full reuse needs a penalty of $T$ while binding each leak to its own round needs $1$, so no penalty that is constant in the horizon deters unbounded reuse; a reuse window of $w$ rounds costs at most $w$ times the largest per-round value. The cheapest profile of per-party stakes that deters every coalition solves a covering linear program. For blockchain design, per-epoch keys cut the required stake from the value of a key's lifetime to the value of one epoch; we calibrate the gap on Ethereum front-running data and place Ferveo and Shutter in the model. The analysis extends to sealed-bid auctions, multi-authority voting, and federated learning under a shared key.

Frequency-modulated continuous-wave (FMCW) techniques underpin both coherent optical ranging and microwave radar, creating a common demand for highly coherent and highly linear frequency-chirped sources across the optical and microwave domains. Conventional semiconductor lasers, however, are fundamentally constrained by trade-offs among linewidth, linear tunability, and operational robustness. Here, we present a robust self-linearized quantum-dot (QD) coherent swept-source architecture through the co-design of source physics and system control. The chaos-free characteristics of QD lasers enable stable low-quality-factor external-cavity locking, yielding a Lorentzian linewidth of 12.6 Hz. Unlike self-injection-locked lasers, broadband external optical feedback allows the laser to maintain high optical coherence and stable operation over a wide current-tuning range, thereby enabling robust turnkey operation together with a chirp bandwidth of 23.2 GHz. Furthermore, a statistically gated real-time iterative learning control (ILC) strategy reduces the chirp nonlinearity $(1-R^2)$ to as low as $8.96\times10^{-8}$, while maintaining excellent environmental stability under laser-temperature variations. To demonstrate practicality, we demonstrate isolator-free coherent LiDAR and photonic generation of frequency-agile, linearly chirped microwave waveforms, establishing a common FMCW source platform for optical ranging and radar waveform synthesis. To the best of our knowledge, we demonstrate for the first time a QD-based coherent swept source that simultaneously combines hertz-level linewidth and ultrahigh chirp linearity, which we envision as a unified source platform for FMCW signal generation and coherent sensing across the optical and microwave domains.

Let $A=(a_{ij})$ be an $n\times n$ real matrix and let $\lambda$ be a partition of $n$. Let $\phi^\lambda$ be the class function dual to the Young permutation character, and let $$ \phi^\lambda[A] = \sum_{\sigma\in\mathfrak{S}_n}\phi^\lambda(\sigma) \prod_{i=1}^{n}a_{i\sigma(i)} $$ be the corresponding monomial immanant. Stembridge [Canad. J. Math. 44 (1992), pp. 1079-1099] posed the following open problem: If all minors of $A$ of order at most $r$ are nonnegative, and the partition $\lambda$ has length at most $r$, is it true that $\phi^\lambda[A]\ge 0$? This paper solves the problem for all hook partitions.

Topological nodal-line semimetals are three-dimensional quantum materials characterized by band crossings that form closed loops in momentum space. In $\mathcal{PT}$-symmetric realizations, these nodal rings are stabilized in the absence of spin-orbit coupling, giving rise to drumhead surface states and unconventional transport responses. In this work, we study charge transport across a nodal-line semimetal containing a finite electrostatic barrier, with both leads described by the same equilibrium material. By solving the corresponding scattering problem, we show that the transmission across the barrier exhibits {Klein-tunneling behavior protected at normal incidence by the nodal topology}, despite the extended nodal-line dispersion, which can be traced back to Berry-curvature-induced momentum locking. Using the Landauer-Büttiker formalism, we derive general expressions for the linear and nonlinear conductances, including both longitudinal and Hall components, and evaluate them at zero and finite temperature. Our analytical and numerical results elucidate the dependence of the conductance on barrier height and width, as well as on a $\mathcal{PT}$-breaking mass term. We identify distinct transport regimes in which nonlinear contributions are strongly enhanced and transverse Hall currents emerge, providing clear transport signatures of nodal-line topology and suggesting potential routes toward device applications.

We present GenLimitLib, a source-aligned Lean 4 library for language generation in the limit. Introduced by Kleinberg and Mullainathan at NeurIPS 2024, language generation in the limit studies a theoretical question motivated by LLMs: how to generate valid new strings from observed examples. This young and rapidly evolving field offers a natural testbed for studying large-scale formalization. GenLimitLib contains formal developments for 30 papers. It extracts shared definitions and reusable proof components while preserving paper-specific assumptions and statements, and records relationships across papers. In this way, GenLimitLib provides a concrete and structured view of the literature. We show through mathematical case studies and LLM experiments how our library can support both human mathematical research and AI-assisted research. Our Library: https://github.com/pengzhang91/generation-in-the-limit-lib.

The pre-training of Large Language Models (LLMs) is increasingly conducted across multiple data centers. As training scales to a larger number of accelerators, the fraction of time spent on computation decreases, while the fraction spent on communication increases. Therefore, frequent synchronization becomes a growing bottleneck. Local update methods reduce this cost by allowing workers to perform several optimizer steps between synchronizations. Most local update methods set the number of local optimizer steps between synchronizations before training and keep this interval fixed throughout the run. However, the best interval can change during the entire train process. If the interval and optimizer are adapted to the current training state, the communication frequency is reduced while maintaining the training performance. In this work, we introduce AutoLoCo, an adaptive training framework to reduce communication in LLM training. It adapts the local interval using scalar training statistics and corrects each outer update. Our method is motivated by two observations: 1) the appropriate local interval varies across training stages, and 2) changing the number of inner steps per interval creates a mismatch with an unchanged outer optimizer, requiring a correction to the outer update. We optimize this mismatch by correction of the outer optimizer for the momentum and the learning rate using the accumulated inner learning rate. Our experiments under communication constraints demonstrate that AutoLoCo reduces communication frequency by 27% relative to DiLoCo while maintaining training performance.

Visual reprogramming is a parameter-efficient method for adapting pretrained models, yet its training can remain computationally expensive: even with a frozen backbone, visual prompts are often optimized through the full model for hundreds of epochs. Before changing what the pretrained model sees, we ask whether we are fully using what it already tells us. We find that modeling the full source response can already yield strong downstream predictions without prompt optimization. Motivated by this observation, we introduce You Only Reprogram Once (YORO), which constructs a downstream predictor from the frozen response space in a single forward-only traversal. Its Bayesian Discriminant Mapping (BDM) derives a covariance-aware affine mapping from streaming class statistics, requiring no backpropagation, optimizer updates, or repeated visits to the training set. When further input adaptation helps, YORO-FP optionally refines the visual prompt for 20 epochs. BDM also extends naturally to CLIP by treating attribute-prompt similarities as source responses. Across three full-data settings, YORO improves average accuracy over the strongest prior gradient-free mapping by 18.4–24.4%. On 16-shot CLIP, it raises the four-backbone average from 71.4% to 77.2%. YORO-FP provides further gains on selected tasks, while validation often retains the one-pass predictor. These results suggest a different default for visual reprogramming: read out the frozen response first, and optimize the input only when needed.

We study federated learning (FL) with adversarial clients, where the goal is to minimize the average loss of the honest (non-adversarial) clients without knowing their identity. Under heterogeneity, a single shared model parameter is statistically inappropriate: it cannot capture the distinct data-generating processes across clients, incurring an irreducible model-heterogeneity bias and severely limiting robustness to adversarial clients (a.k.a. Byzantine-robustness). We address this problem through representation learning, where each client learns a personalized linear head, while collaboratively estimating a shared nonlinear representation through Byzantine-robust aggregation. We demonstrate that the heterogeneity among honest representation gradients is controlled by the representation error and statistical errors that decay either with the number of data samples per client ($\tau$) or the number of iterations ($T$). In particular, our non-asymptotic parameter recovery error bound reveals three terms: (i) an initialization-dependent error that goes away with $T$, (ii) finite-sample noise terms that decreases with $\tau$ and the number of honest clients, and (iii) a stochastic gradient variance term that also reduces with $T$. Importantly, with no irreducible model-heterogeneity bias in our bounds. We extend the regression analysis to multiclass classification, and empirically validate it on CIFAR-10, FEMNIST, and School Exam Score datasets.

The strong generalization performance of on-policy post-training paradigms has motivated studies of their parameter update behaviors. However, these studies treat the observed behaviors only as byproducts in on-policy training, overlooking their potential to serve as optimization principles for improving the generalization of other paradigms such as supervised fine-tuning (SFT). To address this limitation, we investigate whether there exists a specific on-policy update behavior that can achieve such improvements. First, our analyses reveal that SFT updates parameters along consistent directions, while the on-policy paradigm continuously adjusts the direction during training. This difference inspires us to focus on the cumulative update direction of each parameter as a promising behavior. Then, we evaluate its effectiveness for improving generalization by proposing On-Policy direction-constrained Supervised Fine-Tuning (OPSFT), which constrains SFT updates to the direction identified by on-policy paradigms. The strong performance of OPSFT indicates that the generalization advantage of on-policy paradigms can be transferred to SFT through the parameter update direction. Once such a direction is identified, even SFT can generalize with its updates constrained to this direction. This finding offers two practical benefits by combining the strong generalization of on-policy paradigms with the advantages of SFT, including the high training efficiency and ability to leverage high-quality trajectories. For efficiency, we identify update directions that support strong generalization using a few on-policy training steps, and subsequently apply OPSFT to achieve high training efficiency. For leveraging high-quality trajectories, OPSFT can utilize these trajectories to continue improving a post-trained model along its update direction without disrupting the ability learned from on-policy training.

An efficient single-photon nonlinearity is a key resource for photonic quantum technologies. Single-photon nonlinearities have been demonstrated with quantum dots (QDs) across a range of cavity and waveguide geometries. However, none of these realizations simultaneously provide high-efficiency direct fiber coupling and compatibility with scalable on-chip integration. The nanobeam cavity addresses these limitations through its in-plane geometry, which enables highly efficient, direct fiber coupling alongside scalable homogeneous and heterogeneous on-chip integration. Here, we report a single-photon nonlinearity from a nanobeam cavity coupled to a single quantum dot. The nanobeam interfaces directly with a single-mode fiber with an efficiency of 60%. Driven by resonant picosecond pulses, the platform achieves a single-photon nonlinearity with a low threshold of 0.24 incident photons per pulse. The system operates in the strong-coupling regime, exhibiting a coupling rate of g/2{\pi} = 29.1 GHz and a cooperativity of C = 3.98. These results establish the nanobeam cavity-QD system as a compact, chip-integrable, and fiber-compatible building block for scalable photonic quantum technologies.

Training models to act in accordance with an explicitly defined set of principles, or "constitution," has shown promise as a robust and transparent mechanism for AI alignment. However, the generality and flexibility of such methods remain unclear. Here, we show that constitution-consistent behavior can be distilled from synthetic corpora into lightweight objects (low-rank adapters and steering vectors). Despite never seeing a harmful request or jailbreak during training, such objects increase jailbreak defense success and measured alignment – particularly at long context lengths and against multi-turn attacks, where they outperform both prompted and steered baselines. Subtracting control-trained from constitution-trained objects further accentuates these effects, yielding defenses we call "constitutional adapters" (CAs). CAs can be trained on a base model, transferred zero-shot to its post-trained checkpoint, and scaled at inference time to predictably trade off defense for benign compliance. Taken together, these results recommend CAs as a lightweight, portable, and tunable lever for mitigating misalignment and misuse in API deployments.

Reionization is the last major transition in the Universe that transforms the intergalactic medium (IGM) from neutral to highly ionized, yet its spatial structure remains poorly constrained. Using an astrophysics foundation model trained for James Webb Space Telescope datasets, we identify a luminous galaxy at redshift $z=7.567$ showing significant Ly$\alpha$ transmission at $z=6.80\pm0.07$. This transmission, named “the First Ly$\alpha$ transmission in the Reionization Epoch at $z\sim7$" (FL$\alpha$RE-z7), indicates a cosmological-scale ionized structure extending over $210.5^{+55.7}_{-53.5}$ comoving Megaparsecs at cosmic dawn. The hydrogen neutral fraction of the ionized structure is $x_{\rm HI}=(2.07^{+0.29}_{-0.30})\times10^{-6}$, about $10^5$ times lower than the cosmic mean at similar redshifts $(x_{\rm HI}\approx0.4)$. Cosmological simulations suggest a probability of $\approx10^{-5}$ for a random sightline to reproduce the transmission. We find no significant galaxy overdensity associated with the transmission region, in tension with canonical inside-out reionization models. FL$\alpha$RE-z7 severely challenges existing reionization models in the standard $\Lambda$CDM Universe, indicating extreme spatial fluctuation in ionizing radiation or IGM temperature. Our result marks the first example of a foundation model discovering a previously unpredicted astrophysical phenomenon, demonstrating the power of deep learning methods for mining large astronomical datasets and finding unexpected new astrophysics.

Continual test-time adaptation (CTTA) adapts a source model to an unlabeled test stream whose distribution may change over time. Existing TTA methods often assess prediction reliability using confidence or entropy, which primarily reflect the model's self-certainty for the current sample. In CTTA, accumulated target observations can provide complementary evidence for correcting the source prediction, but this history may become misaligned as the target distribution changes. The key question is therefore not how much the correction differs from the source prediction, but whether and how strongly it should be applied. This paper proposes Gain-Aware INtervention (GAIN), a backpropagation-free CTTA framework guided by a simple principle: history proposes, gain decides. GAIN maintains compact target statistics to form a correction proposal and a posterior-predictive evaluator that accounts for estimation uncertainty. The resulting source-relative gain estimates the proposal's benefit and determines a sample-specific intervention strength along a continuous path through efficient one-dimensional optimization. Gain-controlled predictions then update the target statistics online, limiting the propagation of unreliable corrections, all without backpropagation, sample storage, or replay. Across five benchmarks, our method achieves strong predictive performance, with favorable accuracy–calibration–efficiency trade-offs in continual adaptation. On ImageNet-C, for example, GAIN achieves 61.9% accuracy with near-source calibration. It remains stable under diverse and challenging continual shifts while running 15.9x faster than a representative optimization-based CTTA baseline.

Large language models make weight storage and memory traffic major inference costs, motivating low-precision formats that represent each weight with only a few bits. Such formats use a scale to map floating-point values into a small codebook; NVFP4 improves local range utilization by letting every 16 E2M1 weights share an E4M3 block scale. Choosing that scale is difficult in GPTQ because quantizing one column updates those that follow, so evaluating a block independently can misestimate its final reconstruction error. Large models pose a second challenge: full-precision weights, calibration activations, and second-order state cannot all remain on one accelerator, while assigning complete layers to devices leaves each time-consuming layer solve serial. We introduce Schur Replay, a scale-selection algorithm that reproduces the GPTQ updates caused by each block scale and scores the resulting block error after accounting for compensation from unquantized columns. Separately, our execution infrastructure keeps only the active layer resident, tiers activations across device, host, and disk, retires full-precision layers after export, and distributes independent output rows across tensor-parallel ranks. Together, the algorithm and infrastructure attain $99.35\%$ and $100.84\%$ question-weighted recovery from BF16 across seven benchmarks on Qwen3.5-397B-A17B and Llama-3.3-70B-Instruct. On the 397B model, the infrastructure reduces measured per-layer time by $15.17\times$ over ModelOpt and $23.14\times$ over LLM Compressor, with lower memory used per GPU.

Recurrent reasoning models have attracted growing attention for scaling test-time computation, typically by iteratively refining latent states with shared parameters. However, these models apply each learned update with a fixed unit scale, which can be conservative when updates make persistent progress and overly aggressive when they fluctuate, limiting the benefit of additional loops. To understand how the scale should vary along the trajectory, we first analyze the sensitivity of terminal loss to recurrent update scale. We show that its temporal average admits an exact decomposition into persistent-progress and centered-fluctuation contributions. Based on this, we introduce the Trajectory Adaptive Progress-Fluctuation Scheduler (TAPS), which tracks their balance across recurrent updates and adapts the step size online. Theoretically, we establish sufficient conditions under which TAPS reduces expected terminal loss and reaches a target quality in fewer recurrent loops. Empirically, we show that TAPS improves terminal accuracy across structured reasoning tasks without retraining. By further incorporating the progress-fluctuation principle into training, TAPS yields additional accuracy gains with up to 1.56 times wall-clock speedup at matched baseline accuracy. The broad applicability of TAPS is supported by its effectiveness across diverse recurrent architectures and inference strategies. Together, these results establish update scale as complementary control axis of recurrent inference alongside architecture and depth.

Open-ended generation lacks canonical answers, making pointwise rewards difficult to calibrate for group-based reinforcement learning. Directly ranking same-query rollouts provides a more suitable relative reward signal, but existing ranking-based reward methods can incur substantial judging cost. We introduce RankBuffer, which maintains an ordered, query-specific buffer of previously judged responses as a reusable quality scale. Each rollout is first inserted into an anchor interval through an independent coarse judgment, after which only rollouts assigned to the same interval undergo local fine ranking. The resulting complete order is converted into bounded rank rewards, while boundary expansion, local refinement, and inactive-anchor pruning adapt the buffer as the policy evolves. Across four open-ended benchmarks, RankBuffer consistently outperforms all pointwise baselines. It also achieves nearly on-par performance with the strongest ranking-based reward baseline while substantially reducing judging cost. Ablations demonstrate the importance of both local fine ranking and anchor response content, while buffer analyses show that rollout-derived anchors progressively extend and refine the covered quality scale. These results establish response reuse as an effective approach to efficient relative reward construction.

Long-context reasoning in large language models incurs substantial computation and memory costs. Visual text compression (VTC) reduces input length by rendering text as images, but fixed-resolution rendering creates a compression-performance trade-off: low DPI saves tokens at the expense of legibility, whereas high DPI spends tokens on irrelevant content. We introduce FocusVTC, which breaks this trade-off through adaptive resolution while preserving general multimodal capabilities. It combines compressed low-DPI global views with selective region enhancement, integrating enhanced views into ongoing reasoning. We construct 29.4K high-quality Reasoning-Evidence Localization (REL) chain-of-thought examples (REL-CoT) that link reasoning traces to page indices and bounding boxes. Multi-resolution REL supervised fine-tuning (REL-SFT) teaches the model to localize relevant regions, and Group Relative Policy Optimization learns when to enhance resolution and how to use the resulting observations, without a separate continual-pretraining stage. At 72 DPI on RULER v1, FocusVTC scores 87.4 at $2.9\times$ input compression, including tool observations, versus 57.5 for Glyph at $3.0\times$ input compression. It surpasses its text-input backbone on LongBench (56.40 versus 55.86), improves the MRCR macro-average by 13.91 points, and achieves a 51.19 macro-average on VTCBench. The MRCR latency evaluation also shows a $2.79\times$ online end-to-end speedup over Text. General multimodal capabilities are preserved, with MMMU increasing from 65.12 to 66.73 and MME from 2424.02 to 2457.62.

We study how first-order differential observables affect the zero-frequency behavior of black-hole wave equations. Quasinormal modes and late-time tails do not always change in the same way. We give a simple condition for when an observable removes the leading threshold term of a Green function. For Schwarzschild Regge-Wheeler modes, the relevant operator is determined by the regular static solution. The transformation is nondegenerate at nonzero frequency but becomes globally degenerate at zero frequency. As a result, the nonzero quasinormal-mode problem is unchanged, while the leading fixed-radius late-time tail gains one extra inverse power of time.

We investigate the role of initial orbital angular momentum in generating elliptic flow in off-central heavy-ion collisions. The conventional expanding fireball picture is extended for the first time to a simultaneously expanding and rotating fireball characterized by an angular velocity. Spherical and spheroidal emission geometries with two different rotating flow profiles, reminiscent of global and differential rotation, are considered for our investigation. While the transverse-momentum spectra at midrapidity are largely insensitive to the geometry and flow profile for realistic values of angular velocity, the elliptic flow is strongly affected by anisotropic expansion and rotation. A spherical fireball generates elliptic flow solely through the momentum anisotropy induced by rotation, but its magnitude remains below the observed values for realistic angular velocities. In contrast, an anisotropically expanding and rotating spheroidal fireball exhibits a $\sim 10\%$ enhancement of elliptic flow for $\sqrt{s_{\rm NN}}=7.7$ GeV, which reduces to a $2\%$ for $\sqrt{s_{\rm NN}}=39$ GeV relative to its non-rotating counterpart. Within our fireball framework, this indicates that vorticity can contribute at the $2-10\%$ level to the observed elliptic flow.

Vision-language pre-training has reshaped image clustering, giving rise to language-assisted image clustering (LaIC), which leverages textual semantics to complement visual representations. Despite the rapid proliferation of LaIC methods, it remains unclear how much LaIC has actually advanced image clustering, as existing studies generally suffer from major limitations, including inconsistent experimental settings, inadequate dataset selection, and limited evaluation dimensions. To address this gap, we introduce VLM4Cluster, a comprehensive benchmark for image clustering in the era of pre-trained vision-language models (VLMs). VLM4Cluster implements 17 representative methods spanning classical, deep, and language-assisted image clustering, and evaluates them on 20 datasets covering classical, challenging, fine-grained, large-scale, and out-of-distribution settings. Beyond effectiveness, VLM4Cluster systematically investigates image clustering along three complementary dimensions: robustness to adversarial perturbations, generalization under distribution shifts, and computational efficiency. Our study shows that LaIC substantially advances the clustering performance frontier on many semantically demanding benchmarks, generally exhibits stronger generalization under distribution shifts, and achieves a more favorable effectiveness-efficiency trade-off. However, its gains become less consistent on large-scale and fine-grained datasets, while language assistance does not systematically reduce sensitivity to adversarial perturbations. VLM4Cluster is released at https://github.com/YuanweiHuu/VLM4Cluster.

Coding agents change running software: they patch a service's code or overwrite its stored state, and then act on their own expectation of how the service will respond afterwards. A wrong expectation may surface only several calls later. Function-level code-execution benchmarks omit persistent service state, and agent benchmarks score the actions an agent takes or the final state it reaches. We introduce CTE-Bench, which measures whether a model can predict how an intervention changes a stateful service's future behavior, without asking it to choose actions. Each scenario gives the model Python service code, the calls and responses observed before the intervention, the intervention itself (a source edit or a state overwrite), and 40 fixed future calls; the model predicts every future response, and predictions are checked by executing the service. Three memory protocols control whether the model sees the correct earlier responses, none of them, or its own earlier predictions. CTE-Bench-Core-v1 contains 255 scenarios over six deterministic Python services, giving 10,200 predictions per model. The main score is effect-step value match (VM): exact response equality on the 2,476 future calls whose response the intervention changes. With correct earlier responses revealed, four API-hosted models (DeepSeek V4-Flash, Kimi K2.5, Qwen3.6-35B-A3B, and Claude Sonnet 4.6) reach 54.3%-61.7% effect-step VM. Hiding those responses lowers effect-step VM to 23.2%-29.0%; conditioning on self-generated predictions gives 24.8%-33.2%, and at most 1.2% of scenarios are predicted exactly end to end. Current models thus track intervention effects mainly when correct feedback is supplied, and their errors compound over a rollout. We release CTE-Bench-Core-v1 with its executable oracle, evaluation scripts, and an evaluation card mapping each claim to its protocol.