Papers

Multimodal inputs are inherently heterogeneous, not only across modalities but also in the information pathways required for effective prediction. To address this limitation, we propose InstMoE, an adaptive expert routing framework for multimodal learning. InstMoE dynamically routes each input to specialized unimodal and cross-modal experts, allowing the model to adapt its information pathways to the characteristics of the input. However, routing can be misled when modality-specific variations obscure task-relevant semantics. Such irrelevant variations may distort routing decisions, causing inputs to be assigned to inappropriate experts. We therefore introduce a Contrastive Semantic Alignment module, which encourages semantically similar inputs to share task-relevant representations while suppressing irrelevant modality-specific variations. Experiments on multimodal sentiment analysis benchmarks demonstrate that InstMoE achieves state-of-the-art performance on CMU-MOSEI and CH-SIMS v2 while using substantially fewer parameters than competitive baselines. Further analysis shows that different inputs exhibit distinct expert preferences, demonstrating that InstMoE moves beyond fixed fusion toward adaptive multimodal computation.

We develop a family of compact, dispersion-corrected marker-and-cell (MAC)-based finite-difference discretizations for the elastic Helmholtz equation. While dispersion correction is well established for scalar Helmholtz problems, fewer approaches exist for the elastic problems, and most use non-compact stencils. To this end, we introduce a generalized MAC (GMAC) framework that enables compact discretization of the elastic Helmholtz equation. We prove that even though GMAC is non-symmetric, this discretization produces no numerical dissipation and no shear-mode separation. For certain weights, GMAC has lower polarization error than standard MAC. We also introduce an elastic dispersion correction (EDC) process, extending real-shifted-wavenumber dispersion correction methods from acoustic to elastic Helmholtz problems. We prove and demonstrate numerically that GMAC is second-order accurate. Furthermore, we show that EDC substantially reduces dispersion for standard MAC and can provide additional correction in some instances of GMAC. Dispersion analysis shows that the resulting method outperforms existing compact finite-difference discretizations in dispersion accuracy and is competitive with some non-compact schemes. We give numerical evidence for improved phase alignment in mildly heterogeneous media.

Recently, unified high-definition image restoration has attracted considerable attention; however, existing models tend to excessively increase their depth in pursuit of improved generalization, which often yields only limited gains. Meanwhile, loop-based learning paradigms have drawn widespread attention due to their low parameter counts and strong regression capability, as exemplified by GPT-6 and looped Transformers. In this paper, we introduce the loop learning paradigm to address restoration tasks that require cross-domain learning. Specifically, we propose LoopMoEVR, a loop-based mixture-of-experts model capable of handling degraded ultra-high-definition (UHD) inputs. First, a degradation-conditioned low-rank loop embedding is designed to construct input-dependent stage conditions. Second, a spatio-temporal iterative adaptive normalization module, termed IterAda3DN, is developed to fuse local features with global loop context, thereby performing position-wise affine modulation. Finally, the expert branches further integrate the attention-updated local and global video states with the loop conditions to generate dedicated modulation parameters, while an input-conditioned depth predictor adaptively configures the number of loop iterations. With only approximately 0.884M trainable parameters, the proposed model uniformly handles UHD video dehazing, deraining, denoising, and low-light enhancement tasks, achieving state-of-the-art restoration performance on both public benchmarks and real-world scenarios.

Diffusion models achieve strong generative performance but often struggle to align generated samples with human preferences measured by a reward model. A simple yet effective algorithm for test-time alignment is Best-of-$N$ (BoN) sampling, which draws $N$ i.i.d. samples from a pre-trained diffusion model and outputs the single highest-reward sample. Despite its empirical success, BoN makes limited use of reward information, as it is incorporated only at the final selection stage without influencing the reverse diffusion trajectory during sampling. Consequently, BoN sampling does not improve the average alignment of generated samples and is primarily suited to single-output settings. We propose Best-of-$N$ Guidance (BoNG), a novel method that integrates the principle of BoN sampling directly into the reverse diffusion process. BoNG performs online BoN selection over denoising particles and adjusts the reverse diffusion process to steer the particle population toward higher-reward regions during generation. Specifically, by introducing an asymmetric guidance interaction among denoising particles, BoNG uses the current BoN particle as a guidance signal to the rest of the particle population. This particle-level interaction reshapes the sampling process toward higher-reward regions, enabling BoNG to improve not only the final best sample beyond Vanilla BoN sampling, but also the average quality of generated samples. Over 36 empirical comparisons, BoNG achieves the best performance in 29 cases, ranking first in 80.56% of the comparisons against SMC and Vanilla BoN sampling. BoNG also supports multi-output capability, achieving 1.3$\times$ ImageReward score of the latest sample-based guidance method with a 1.6$\times$ speedup. We release the code at https://github.com/aailab-kaist/BoNG.

Search agents repeatedly make short decisions about relevance, evidence sufficiency, and search actions. Using generative language models for these decisions introduces latency and unreliable confidence. We present SearchJev, a fast and calibrated System-1 model that separates search decisions from System-2 reasoning and generation. Given a search state and a decision schema, SearchJev directly scores legal options without autoregressive output generation. We propose Soft-Label Learning for Calibrated Decisions (SLCD) to learn decision probabilities from uncertain supervision and calibrate their confidence. In a dual-system search agent, SearchJev handles short decisions and delegates uncertain judgments to System 2, which retains planning, query generation, and answer composition. We also introduce SearchDecision-Bench, a benchmark unifying six types of search decisions for training and evaluation. On SearchDecision-Bench, SEARCHJEV improves decision quality over same-size Qwen3.5 autoregressive models, achieves 5.2-5.3 times faster decisions, and reduces average expected calibration error by 41-74%. On BrowseComp-Plus, the dual-system agents achieve a 3.7-4.7 times speedup in active search time while improving answer accuracy from 45% to up to 54%.

Long-form reasoning makes inference expensive, and speculative decoding mitigates this cost by verifying multiple draft tokens in parallel. Its speedup, however, can fade as context grows and draft acceptance declines. We focus on attention-mass dilution: as softmax normalizes over more visible Keys, the mass concentrated on the highest-scoring Keys can decrease. We introduce SharpDraft, a training-free method that counteracts this effect through cardinality-aware Query scaling, without the computational overhead of online adaptation. Under explicit assumptions, we derive an exact top-$k$ mass correction and deploy a closed-form fixed-slope approximation. Across AIME-26, GPQA-Diamond, and LongGenBench Diary, SharpDraft achieves $2.59$-$3.19\times$ geometric-mean end-to-end speedups over target-only autoregressive decoding when applied to DFlash, PARD, and EAGLE 3.1. With DFlash, it improves decoding speed and outperforms full-parameter and LoRA-based online adaptation in end-to-end speedup, while matching the unmodified drafter's reported peak allocated GPU memory.

Split learning has emerged as a promising paradigm for privacy-preserving LLM fine-tuning, yet its practical deployment is severely hindered by the sequential communication-computation bottleneck. In conventional synchronous pipelines, clients remain idle while waiting for server-side gradients, resulting in substantial training inefficiency. We propose CommuteProp, an asynchronous split-learning algorithm that decouples the training process into two concurrent phases: a cross-block forward-backward pass and an in-block weight update. By overlapping computation with communication, CommuteProp reduces the marginal cycle time from a sum of all stage latencies to the dominant computational bottleneck. We provide a rigorous asynchronous error and convergence analysis. Moreover, we derive an NS preconditioner method based on our analysis to further mitigate staleness-induced noise. Comprehensive experiments indicate that our algorithm achieves substantial throughput gains while maintaining accuracy comparable to synchronous methods; furthermore, it functions as a plug-and-play module that not only accommodates but actively enhances existing privacy enhancement methods.

Reliable learning-based vulnerability detection requires high-quality labels, yet datasets built from vulnerability-fixing commits may label functions as vulnerable simply because they were changed by a security patch. We present VulValidate, a framework that uses LLM agents to coordinate dynamic analysis tools and construct vulnerability-triggering experiments from runtime feedback. Given a labeled function and its fixing patch, VulValidate reconstructs vulnerable and fixed revisions, selects suitable tools and execution paths, refines triggering inputs, and compares runtime behavior to assess function-level attribution. We audit all 35,849 instances originally labeled vulnerable in BigVul, PrimeVul, and DiverseVul. We confirm 20,510 (57.2%), correct 6,819 labels (19.0%), leave 7,981 attacked but undecided (22.3%), and cannot successfully measure 539 (1.5%). After conflict resolution and byte-exact deduplication, the corrected release contains 15,890 distinct confirmed vulnerable function bodies. In a blinded review of 581 sampled decisions, expert consensus supports 90.0%–92.0% of confirmations and 92.6%–99.0% of label corrections. With model parameters fixed, corrected evaluation lowers F1 for all five tested detectors on both BigVul and DiverseVul; retraining with corrected labels improves F1 for four of five detectors on each dataset. We also release a reusable VulValidate skill, corrected datasets, and reproducible evidence for future vulnerability-detection research.

We study differentially private learning problems in the realizable setting, where a hypothesis is specified by $k$ components. A direct iteration of private component learners is obstructed by a simple difficulty: an approximate choice of the next component may destroy exact realizability of the labeled sample, even when the next component is locally accurate. We restore realizability using the LabelBoost procedure of Beimel, Nissim, and Stemmer [SODA '15, Algorithmica '21] and recycle data through two alternating reservoirs. The resulting learner, for a target privacy $\varepsilon$, pays only $\widetilde O(\sqrt{k}/\varepsilon)$ overhead relative to the active sample requirement of a single component learning step at target accuracy $\Theta(\alpha/k)$. For learning $d$-dimensional halfspaces over a finite coordinate grid of size $L$, exact realizability makes the direct component-depth objective quasi-concave. Instantiating the framework with the IPConcave algorithm of Nissim, Tsfadia, and Yan [SODA '26] and with the quasi-concave optimizer of Cohen, Lyu, Nelson, Sarl'os, and Stemmer [STOC '23] yields a realizable sample complexity of $\widetilde{O}\left(\frac{1}{\varepsilon \alpha}\cdot \min\{d^{2.5} \log^*L, \:\: d^{2.5} + d^{1.5} 2^{\log^*L}\}\right),$ which improves on the previously known bound of $\widetilde{O}\left(\frac{1}{\varepsilon \alpha}\cdot\min\{\frac{1}{\alpha}\cdot d^{5.5}\log^*L,\:\: d^{2.5}2^{\log^*L}\}\right).$ We also apply the framework to Boolean compositions: given proper private learners for classes $H_1,\ldots,H_k$, we obtain a proper private learner for $G(H_1,\ldots,H_k)$ for any fixed Boolean function $G:\{0,1\}^k\to\{0,1\}$. Compared with the closure theorem of Alon, Beimel, Moran, and Stemmer [COLT '20], this reduces the overhead on a common component sample bound from $\widetilde O(k/\varepsilon)$ to $\widetilde O(\sqrt{k}/\varepsilon)$.

State-of-the-art neural operators scale to complex meshes via slice-and-process architectures, yet many rely on linear compatibility scores for latent tokenization. Under common feature normalization, such scores are equivalent to isotropic Euclidean clustering, while without normalization they induce unbounded linear decision regions. In both cases, they lack slice-specific anisotropic locality, which can lead to redundant and entangled latent slices. To address this, we propose Metric-Enhanced Token Routing Operator (METRO), a geometry-aware routing mechanism that replaces linear projection with a learnable Mahalanobis metric. By enabling each latent slice to learn a local anisotropic tensor, METRO shapes receptive fields into exponentially localized, oriented ellipsoids that naturally align with flow features like boundary layers and wakes. As a drop-in replacement, METRO yields consistent improvements across both Transformer and Mamba backbones. Empirically, our method achieves substantial performance gains on irregular domains, outperforming baselines on both standard PDE benchmarks and complex industrial design tasks. Finally, METRO exhibits enhanced robustness in out-of-distribution regimes across varying Reynolds numbers and geometric configurations.

Locating relevant files from natural-language requests is a core subtask for agents operating over code repositories. We investigate whether this task can be delegated to compact, specialized models to enable on-device search while reducing the token usage, latency, and inference cost of larger agents. We show that semantic search improves file localization, with gains in accuracy, cross-language transfer, and inference efficiency. To study this setting, we introduce a training framework for file-localization agents built around ColGREP, a local semantic search tool based on late-interaction retrieval models. Our recipe combines weighted supervised fine-tuning on teacher trajectories, assigning turn-level credit based on retrieval outcomes, with reinforcement learning on localization quality. We train three model families with fewer than two billion parameters to formulate search queries, inspect retrieved content, and identify relevant files. On localization tasks derived from SWE-bench Lite and Multi-SWE-bench Flash, ColGREP-equipped agents substantially improve over their base models and outperform corresponding GREP-based agents. In addition to improving localization accuracy, ColGREP reduces mean end-to-end trajectory latency by 44.1% on CPU while using 29.1% fewer tokens, and enables better generalization to programming languages unseen during fine-tuning. These results suggest that compact, tool-specialized localization agents can provide an efficient interface between natural-language requests and large codebases.

Many latent neural operators represent input and output fields in a stationary latent chart. In particular, common latent routing mechanisms use fixed or shared assignment weights for feature projection and reconstruction, limiting their ability to model transport-dominated systems where coherent structures move relative to fixed coordinate frames. We propose Advectra, a transport-aware latent operator that introduces a regularized kinematic coordinate map to decouple source and target coordinate systems. This yields an approximately co-moving latent reference frame and enables asymmetric feature aggregation and reconstruction. Combined with a geometry-aware ordering mechanism for state-space models, Advectra captures advective dynamics while maintaining stable global interactions. Advectra achieves the best performance among evaluated geometry-constrained and form-free baselines on advection-dominated benchmarks, including passive scalar transport in Navier–Stokes flows and Rayleigh–Taylor instability, while demonstrating strong generalization on real-world engineering tasks. These results highlight the benefit of explicit moving-frame structure in neural operators for non-stationary physics.

As multimodal large language models (MLLMs) reason for longer, attention to the initial visual input diminishes, weakening visual grounding. Visual memory reintroduces visual evidence during reasoning. We conduct a controlled analysis of visual memory along three axes: curation, organization, and access. We find that local evidence benefits from global context, compact latent representations balance accuracy and visual-context cost, and the utility of memory access depends on the reasoning state. Guided by these findings, we propose ReMAP (Reasoning-Time Memory-Augmented Perception), which couples two complementary latent memories: a static, question-conditioned Global memory that preserves scene and cross-image context, and a dynamic Local memory that uses this context as an anchor while selecting and re-encoding region-level evidence according to the current reasoning state. Both memories return compact latent tokens inserted into the reasoning sequence, and a reinforcement-learning access policy trained with branched rollouts decides when to continue reasoning or invoke Global or Local memory. On ten benchmark families, ReMAP outperforms prior visual-memory methods on all four multi-image benchmarks, exceeding the strongest prior results on MuirBench and MIMIC by 8.38 and 14.84 percentage points. Across four backbone families, enabling memory access improves over the same trained model with memory disabled, and on shared V*Bench, CV-Bench-2D, and MuirBench questions ReMAP reduces the visual tokens entering the reasoning sequence by 51.0-76.8% relative to the native-resolution backbone. Further analyses show that Global and Local memory form distinct yet complementary latent representations. Together, these components restore the perceptual cycle by letting the reasoning state trigger targeted visual retrieval, with the retrieved evidence guiding subsequent reasoning.

Latent Dirichlet Allocation (LDA) and models derived from it remain widely used topic models. LDA observes each document as a bag-of-words and models each topic by a categorical distribution over the vocabulary, so that it uses neither the internal structure of the document nor the similarity in meaning between words. Earlier work has responded to this limitation in two ways: many models have introduced embeddings, and some have assigned topics to sentences rather than to words. Their combination, a topic model that observes sentence embeddings, remains little explored. We propose vMF Sentence LDA (vSLDA), which keeps the admixture structure of LDA, observes each sentence as its L2-normalized embedding and models each topic by a von Mises-Fisher (vMF) distribution, which matches the cosine geometry of sentence embeddings. Its per-topic parameter count and per-iteration cost are linear in the embedding dimension, versus quadratic for the full-covariance Gaussian distribution in the existing model over sentence embeddings. We evaluate vSLDA where the limitation is expected to matter most, among topics that share much of their vocabulary: the topics that subdivide the one subject of a collection, and the narrow topics that result when a corpus is divided into a large number of topics. On two corpora, vSLDA attains the best mean rank against eight baselines when the fine categories within each coarse category are classified from the document-topic distributions. On the whole corpus, its advantage appears or widens as the number of topics grows. Weighting the word frequencies of each sentence by its topic posterior yields expected topic-word counts of the same form as those of LDA, so that the standard topic coherence and diversity measures apply to models that assign topics to sentences. On their product, topic quality, vSLDA leads in most within-category conditions of both corpora.

Researchers increasingly use Large Language Models as judges (LLM-as-a-judge) to evaluate model outputs. Yet there are no standards for how to design these judges. Typically, researchers choose the prompt, rating scale, and model intuitively. If these choices change the judge's verdicts, two studies can reach different conclusions about the same facts. To address this risk and to provide an empirical basis for judge designs, we evaluate 10 reasoning models across multiple designs on two tasks: a scalar rating of sentence sentiment and toxicity (over 500 items per category), as well as a binary accuracy classification of question-answer pairs (n=600). For the rating tasks, despite judges showing significant disagreements with the human ground truth, the practical size of differences is small enough to consider most judges reliable (mean absolute deviation of 0.11 points on a 1 - 7 scale); toxicity judges even outperform standard classifiers. Judges are also highly accurate on average (96.5%) for the accuracy classification task. However, design choices can produce shifts: changing the rating scale alone can shift measured bias by up to 0.93 points (rating task), and while accuracy levels are rarely impacted, design choices consistently impact judge leniency (classification task; leniency drop of 28.9 percentage points when using detailed prompts, and up to 56.1 percentage points when switching models). Counterintuitively, lower reasoning effort affects neither accuracy nor leniency. Across both tasks, model identity is the dominant source of variance. These findings suggest that while LLM judges are broadly trustworthy in aggregate, design choices can be meaningful sources of variance. Given the growing reliance on automated evaluation in LLM research, we intend this study as a methodological reference for designing more robust and replicable LLM-as-a-judge pipelines.

LLM gateways enforce separate tenant quotas while sharing model deployments and cooldown records that temporarily exclude failing backends. However, a tenant's request failure can update these shared records and restrict other tenants' access to serviceable deployments. We identify two attacks that exploit this gap in LiteLLM. The first uses requests rejected at the key's requests-per-minute (RPM) limit: caller-supplied identifiers for known registered deployments reach failure handling, allowing two rejected requests to redirect another tenant to fallback with zero upstream calls from those requests. The second uses admitted traffic to create cooldown records that persist after backend capacity recovers. To address these failures, we design an origin check that blocks deployment updates from key RPM rejections and tenant-scoped cooldown that preserves the triggering tenant's back-off while retaining shared records for backend faults. Experiments with authenticated proxies and self-hosted vLLM demonstrate that the attacks can force fallback or denial while deployments remain serviceable. Across five paired four-worker runs, tenant scoping reduces victim fallback after recovery from 54/60 to 0/60, while increasing attempts against exhausted shared quotas. These findings show that tenant isolation must cover both request admission and the failure handling that governs shared deployment availability.

Contrastive Reinforcement Learning (CRL) learns representations that estimate goal reachability, yet its policy remains conditioned on raw goals and therefore does not directly exploit the geometry encoded by its critic. We introduce Direction-Conditioned Policies (DCP), a method built around a small modification to CRL: DCP selects previously visited states as waypoints during online training and conditions the policy on their direction and distance in representation space. At deployment, DCP applies the same interface directly to the final goal, requiring neither waypoint selection nor planning. Across nine navigation and manipulation tasks, DCP attains higher final success rates than CRL on seven tasks and spends more time near the goal on seven. Controlled maze experiments further show that DCP captures shortest-path geometry more accurately and that the supplied direction causally influences the actor's behavior. We identify waypoint coverage and ranking as limits to exploration, and show that learned candidate generation improves goal reaching in two controlled mazes.

Music is universally present in human societies. Ethnomusicologists have long been documenting the diverse expressions of human musicality, and comparative musicology has recently brought several studies of folksong to a more global scale. Such cross-cultural research has not been conducted on lyrics: the language barrier has so far prevented work with multi-lingual data. However, Natural Language Processing (NLP) technologies have reached a stage where this language barrier may no longer be prohibitive. Combining folksong lyrics corpora across five languages, we machine-translate them to a pivot language with a pre-trained neural topic model, and we examine the relationship between content and social function within each language, and across languages for wedding songs. As expected, human evaluation of translation results shows that non-Indo-European languages suffer from overall worse translation quality. Experiments with topic models then indicate that the content of lyrics is at best partially related to the social function of folksongs across all languages. These experiments are just first steps into cross-cultural folk musics lyrics analysis; however, they do indicate that a previously unobserved web of cross-cultural relationships beyond ethnomusicological typologies may be uncovered through the study of what people sing across the world's diverse folk musics.

Teleoperation is becoming increasingly important for collecting high-quality demonstrations to teach robots dexterous manipulation skills. For dexterous manipulation, bare-hand tracking provides a practical way to control robotic hands and demonstrate coordinated finger movements without gloves or exoskeletons. However, this type of teleoperation faces two key feedback limitations: a lack of force feedback and visual feedback of occluded region. The absence of force feedback hinders precise and safe manipulation, as operators must infer contact force visually rather than feel them directly. Occlusion by objects or other robot parts impairs the assessment of hand positions, approach distances, and pre-grasp configurations suited to the object's shape. To address these limitations, we propose a digital-twin-based augmented reality (AR) teleoperation interface that integrates bare-hand tracking, bimanual robotic hands, and virtual representations of the robot and task objects. The interface renders tactile measurements on robot hand meshes as visuo-force feedback. Furthermore, it offers a user-controlled virtual view panel or occlusion-aware transparency rendering to provide occlusion-mitigating visual feedback. We evaluated the interface through user studies involving Task 1 and Task 2, examining how force visualization and visual assistance support demonstration collection in tasks requiring careful force regulation and manipulation under occlusion.

LLM-based text-to-speech (TTS) models can generate emotionally expressive speech, but how reference emotion is routed through the model and realized in decoded speech remains unclear. We introduce two emotion-sensitive metrics for matched neutral and emotional syntheses—a codec trajectory score and a late residual direction score—and use them to score activation-patching interventions. Under controlled matched-reference conditions, this analysis identifies a sparse source-to-readout component-level circuit: 23–27 attention heads and MLPs per emotion, roughly 5% of the components considered, recover or suppress 74–88% of the late emotion-readout shift on held-out cases. The circuit combines a shared component backbone with emotion-specific components; cross-emotion activation swaps reduce the target readout in 47 of 48 cases. In decoded speech, the same intervention produces consistent changes in pitch, energy, and spectral brightness over 24 matched pairs per emotion. A readout-matched residual-direction baseline produces only 17–27% of the intervention's pitch effect, showing that internal readout movement alone does not explain the decoded acoustic changes. These results trace a compact causal route from reference-derived prefix information to emotion-relevant properties of generated speech.

Accurate retinal vessel segmentation requires features that capture vascular geometry while preserving information for fine-scale reconstruction. We propose CGDD-Net, a context-guided dynamic detail modeling network that connects adaptive feature extraction to a shared decoder pathway. Context-Guided Scale-Adaptive Deformable Encoding (CSDE) combines fixed-grid convolution with deformable local attention to capture vascular patterns at multiple spatial extents. Spatially Adaptive Multi-Kernel Gating (SAMG) selects receptive-field responses at each location. Dynamic Cross-Scale Detail Fusion (DCDF) aligns the gated intermediate features and compresses them into an eight-channel representation, which is reused at three decoder resolutions together with selected encoder skips. This design consolidates intermediate information before decoding instead of transferring each middle-stage feature through a separate direct skip. On DRIVE, CHASE\_DB1, STARE, and HRF, CGDD-Net achieves AUC values of 0.9824, 0.9938, 0.9895, and 0.9874, with F1 scores of 0.8323, 0.8102, 0.8510, and 0.8157, respectively. The complete model contains 1.96 million trainable parameters. In cumulative ablations, the full model improves F1 over the internal baseline by 2.48, 0.61, and 3.49 percentage points on DRIVE, CHASE\_DB1, and STARE. Twelve directed cross-dataset experiments further characterize transfer without target-domain adaptation. The results support shared intermediate detail delivery as an effective, parameter-compact architecture for retinal vessel segmentation. Code is available at \url{https://github.com/lixincheng-xcl/CGDD-Net}.

Backtests of LLM trading agents often assume that every order fills at the closing price. We ask whether this choice changes only reported returns or also the order of the agents. Five prompted LLM signal policies and seven classical baselines trade the same synthetic price paths under six execution settings, from near-ideal fills to latency, spread, participation, and impact stresses. The main experiment contains $2{,}462$ runs with matched decision frequencies and paired market paths. On the compressed two-asset board, agreement between the near-ideal and default-stress rankings falls to Kendall $\tau_b=0.21$ in the high-volatility regime, compared with $0.82$ in the calm regime. The seed-bootstrap intervals, $[0.00,0.52]$ and $[0.48,0.94]$, are wide and overlap. On a fixed 11-policy board, agreement rises from 0.24 with two assets to 0.85 with ten; the two-asset point estimate differs substantially from the wider settings we tested. Rank changes are related to turnover, and comparisons with buy-and-hold also depend on how that anchor is initialized. The experiment does not compare LLM trading skill. It shows that, on a short horizon, an execution convention can become part of the benchmark's headline. Execution assumptions and rank stability should be reported alongside returns.

Test-time training (TTT) lets a model store information in its weights during inference. When the model learns from its own output, however, each update also changes the model that generates the next training example. Across 128K-token streams, retaining generated-text updates worsens prediction on independent human-written text with three TTT-E2E model configurations (labeled 125M, 760M, and 3B). The same failure occurs when Adam updates Qwen3-4B's existing weights. The same update mechanisms can improve on real text, so writing itself is not the failure. Three matched comparisons trace the causal pathway. Fixed Generation removes over 98% of the damage at 125M and 760M by using a frozen model to generate training chunks. Recorded Replay separates the loss caused by reading degraded text from the additional loss stored by updating on it. A paired one-update comparison then shows the local conflict: an update predicts its source better but new real text worse. This cost grows after Closed Loop adaptation, with a few trajectories accounting for most large failures. Finally, Settlement evaluates the candidate state on independent real text before commitment. It leaves mean endpoint gaps of 0.07 and -0.02 nats at 125M and 760M while retaining real-text adaptation. These results motivate checking prediction on independent evidence before retaining an update.

Low-light UAV-based RGB-infrared oriented small-vehicle detection is important for nighttime traffic monitoring, emergency response, and urban inspection. Illumination variations, headlight glare, local shadows, and thermal-response degradation cause spatially varying modality reliability, while the small visual extent of vehicles further weakens boundaries, orientation cues, and thermal responses. Accordingly, selecting trustworthy observations based on local modality reliability while further exploiting complementary discriminative information in regions with ambiguous modality preference is key to constructing effective multimodal representations. Based on this insight, we propose ReDiffNet, a reliability-conditioned differential representation network in which modality reliability guides both evidence selection and complementary recovery. Specifically, degradation-aware reliability learning estimates relative spatial reliability, uncertainty-guided differential recovery exploits cross-modal differences to recover complementary cues in ambiguous regions, and reliability-conditioned reconstruction integrates retained and recovered evidence into a unified representation. ReDiffNet achieves 85.3% and 73.9% mAP50 on DroneVehicle and VEDAI, respectively, supporting its effectiveness.

Learning population dynamics from unpaired temporal marginals is an ill-posed inverse problem that requires structural assumptions on the underlying dynamics. We introduce $\textit{Discrete Action Matching}$ (DAM), a finite-state counterpart of Action Matching based on discrete Wasserstein geometry. For a prescribed marginal path and transport geometry, we derive an action-minimization objective for its canonical minimum-kinetic-energy current. Our key observation is that the density dependence of the discrete action reduces to neighboring density ratios. Along an empirical interpolation of the snapshots, DAM first estimates these ratios and then learns an action potential. The learned fields also define a graph-supported Markov sampler. Experiments on controlled synthetic dynamics and real mouse gastrulation data evaluate marginal reconstruction and interpolation. Additional experiments approximate numerical surface-transport paths from paired samples.

On-policy self-distillation (OPSD) provides denser token-level supervision and better computational efficiency than Reinforcement Learning with Verifiable Rewards (RLVR). However, this denser supervision may introduce substantial noise and training instability. Existing improvements often rely on high-variance per-token statistics and introduce extra hyperparameters and trade-offs. Based on the advantage formulation in RLVR, we analyze the OPSD objective from the same perspective, incorporating outcome correctness signals. We find that vanilla OPSD imposes insufficient penalties and excessive rewards on incorrect trajectories because it applies a fixed divergence objective regardless of outcome correctness. Furthermore, the reliability of teacher supervision is associated with both trajectory outcome and the cumulative average teacher entropy along the rollout. Based on these observations, we propose Outcome-Guided On-Policy Self-Distillation (OG-OPSD), which dynamically adapts both the divergence objective and distillation position according to binary outcome rewards and the cumulative average teacher entropy. Extensive experiments show that OG-OPSD consistently improves the performance of vanilla OPSD and multiple strong baselines in mathematical reasoning, multimodal reasoning, and out-of-distribution tasks across Qwen3 models at 1.7B, 4B, and 8B scales, as well as Qwen3-VL-2B.

Prior work suggests that sentiment can often be captured by approximately linear directions in LLM activation spaces, but a single direction may not fully capture sentiment representations. In natural communication, sentiment is shaped not only by polarity but also by usage factors, such as tone and audience adaptation. We test whether these factors systematically modulate sentiment representations beyond a shared sentiment direction. We construct a controlled paired dataset that holds event content fixed while varying sentiment polarity and usage factors, and analyze Llama, Mistral, and Gemma. We identify a shared sentiment direction, remove it, and test the residual structure through erasure and generation-time tone steering. Across models, the shared direction is robust (median cosine 0.953-0.975), yet removing it leaves 0.833-0.909 of the original positive-negative representation-difference norm. The residuals contain compact, reproducible usage-conditioned structure. Targeted erasure weakens held-out usage metrics more than random and label-shuffled controls. On Llama, outputs steered along residualized tone components are preferred in 92.8% of blind target-tone comparisons while preserving the requested sentiment polarity in 98.7% of evaluated outputs.

Adapting step-distilled text-to-image (T2I) models through post-training incurs additional computational costs and affects native few-step generation behavior. This motivates a complementary route beyond style-specific adaptation: drawing on the visual knowledge already encoded in step-distilled T2I models to elicit stylistic capabilities through language. Pursuing this direction requires textual guidance that captures how visual attributes jointly define a style and remain applicable as the depicted content changes. To explore this approach, we introduce StyleForge, a fully automatic, training-free framework that expresses reference styles as reusable rendering instructions. By integrating overall rendering characteristics with local color and lighting behavior, StyleForge organizes visual evidence from reference images into a coherent specification of how the target style should be expressed. The specification is then compiled into textual guidance that can be reused across content prompts, enabling frozen step-distilled T2I models to render different subjects and scenes in the reference style while retaining native few-step generation. Extensive experiments show relative gains of up to 29.47% in generation quality scores over the strongest baseline, while Pareto analysis indicates that improved stylization is accompanied by strong adherence to the requested content. Code is available at https://github.com/sunshy-1/StyleForge.

Secure device-to-device (D2D) caching requires users to generate delivery signals from their own caches and to recover their requested files without learning the other files. We construct schemes that meet both requirements when communication is limited to nearby users. The network consists of $K$ users on a cycle, a library of $N$ independent files, and a reception radius $r$. Delivery uses one round of noiseless orthogonal local broadcasts, with no server transmission. Each user's cache has capacity $M$ file units. An outer secret-sharing code assigns file shares to senders, and a secure local caching scheme at each sender delivers the requested shares. The cache budget includes the sender's library representation, random state, and delivery keys. We prove that each receiver's cache and all audible broadcasts jointly reveal no information about nonrequested files, despite the dependence between the outer shares. The converse counts transmissions crossing a contiguous user block and uses secrecy to remove one user's observation from the entropy bound. Let $h=2r$ and $K\ge h+1$. The private coded caching (PCC) instantiation attains the minimum peak sum load $K/h$ for every $M\ge Nh$ when $N>1$. For $N\ge K$, its memory-sharing envelope is within a factor $15$ of the optimum throughout $M\in[N/h+2,Nh]$. We also apply lifting without secrecy constraints, achieving the optimal load on $M\in[hN/(h+1),N]$ and a factor-$9/2$ guarantee on $M\in[N/(h+1),N]$ when $N\ge K$.

This paper develops a mesh-free numerical method for the one-phase Mullins-Sekerka flow, in which the interface bounding a set moves as the Wasserstein gradient flow of its perimeter. One difficulty in computing this flow is that two pieces of the interface with opposite normals form a hidden boundary when components merge. Our approach represents the interface by an oriented point cloud, which we treat as an oriented varifold. We introduce a quantity that measures how much nearby normals cancel each other and use it to detect the hidden boundary. Based on this quantity, we construct a perimeter estimator, prove that its expected error decays at an algebraic rate in the number of points, also in the presence of a hidden boundary, and use it in a minimizing movement scheme for the flow in two dimensions. Numerical experiments show that the scheme computes mergers of components through the hidden boundary, while the perimeter term in the scheme decreases at every step and the area is nearly conserved. As hidden boundaries also appear in other perimeter-driven interface problems, we expect our oriented representation to be applicable to them as well.