arXiv:2606.20855v2 Announce Type: replace
Abstract: Sparse tensors represent data that is mostly zero or some other compressible fill pattern. Such datasets can be massive, so optimized tensor algebra libraries and compilers have been developed to exploit these patterns to improve performance. Existing systems, however, frequently lack support for parallelism, especially when outputs are sparse or multiple inputs are sparse. We propose WingSpan, a sparse tensor language enabling unrestricted parallel programming. WingSpan supports arbitrary composition of parallel loops and data structures, matching or exceeding the performance of hand-optimized parallel routines on critical kernels such as SpGEMM. We also introduce a dependence theory for the safety of parallel programs involving sparse tensors and structures beyond sparsity.
Science Journals
arXiv:2606.20959v2 Announce Type: replace
Abstract: Language models may encode both outdated facts and their newer replacements. We introduce Parametric Temporal Conflict (PTC), where the newer fact is present and recoverable, but the default forward pass prefers the outdated one. We release a deterministically verified benchmark of 8,746 Wikidata position-holder transitions and evaluate four open-weight language models across three families. A date-prefix prompt recovers the newer fact in 61-81% of PTC cases. Activation patching flips predictions in 72-85% of cases and localizes the preference to model-specific upper-layer regions. Residual-stream steering outperforms norm-matched random directions, indicating direction-specific representations. These results show that PTC reflects a localized representational preference rather than missing knowledge. Recovery is measured on oracle-identified conflicts because automatic detection remains unreliable. We release the benchmark, code, and statistics.
arXiv:2606.21841v2 Announce Type: replace
Abstract: Multiphysics simulation of semiconductor processes requires simultaneous command of governing equations, numerical methods, software implementation, and process physics, creating a substantial entry barrier for newcomers. We investigate whether large language model (LLM) assistance and curated literature guidance can reduce this barrier during the development of a low-pressure chemical vapor deposition (LPCVD) model for graphene growth using the open-source Multiphysics Object-Oriented Simulation Environment (MOOSE). We compare GPT and Claude across direct prompting, staged development without literature, and staged development with literature guidance. Staged development successfully produced converged simulations but revealed two recurring failure classes. We define a false summit as a simulation that converges and appears physically plausible while implementing incorrect physics, and silent drift as the undetected propagation of unverified assumptions through successive development stages. In our case study, a non-conservative species-transport formulation generated spurious methane depletion that was visually indistinguishable from genuine surface consumption, while an omitted self-limiting surface-reaction term produced physically incorrect growth behavior that the LLM repeatedly rationalized as valid. Literature guidance substantially reduced interaction burden during the most challenging development stage, decreasing prompt counts by 91% for GPT and 64% for Claude. Based on these observations, we propose an eight-category failure taxonomy, model-specific human oversight strategies, and a stage-gated workflow to improve the detectability of hidden modeling errors. The results demonstrate that LLMs can lower the barrier to multiphysics simulation development, but rigorous physical validation remains essential because apparently reasonable solutions may conceal critical defects
arXiv:2606.21857v3 Announce Type: replace
Abstract: Rain-induced attenuation and scattering become significant at terahertz (THz) frequencies, and exploiting this rain sensitivity is necessary both to safeguard link reliability and to enable opportunistic environmental sensing without dedicated instrumentation, a capability that remains largely unvalidated on real outdoor channels above 100 GHz. This article investigates opportunistic rainfall estimation using measured lower-terahertz (THz) channels at 140 and 229 GHz. Outdoor measurements over a 41.5-m rain-exposed path are used to characterize rain-induced attenuation and the rainfall dependence of an effective Rician K-factor. Because the path-representative drop-size distribution (DSD) is unavailable, several propagation-model scenarios based on ITU-R P.838-3 and Mie theory with canonical DSDs are employed to quantify model-form sensitivity. These channel characteristics are then used to generate physics-constrained synthetic received-power sequences for training RainFormer, a compact attention-convolution regression network that combines temporal features with explicit attenuation and fluctuation statistics. Under matched synthetic conditions, RainFormer achieves RMSEs of 0.1782 and 0.2925 mm/h at 140 and 229 GHz, respectively, and outperforms the investigated convolutional and Transformer baselines in most metric-frequency combinations. Direct application to the independent measured dataset produces physically consistent rainfall estimates at 140 GHz and demonstrates that received-power fluctuations provide useful information beyond mean attenuation. The results establish a measurement-informed framework for evaluating lower-THz links as opportunistic rainfall sensors while explicitly accounting for propagation-model uncertainty.
arXiv:2606.22394v2 Announce Type: replace
Abstract: Consistency distillation has significantly accelerated diffusion-model inference, but its sampling dynamics remain underexplored. We reveal an asymmetry: although Logit-Normal sampling priors work well for standard iterative generation, consistency distillation exhibits a different difficulty profile (e.g., U-shaped), with optimization bottlenecks concentrated at the boundary stages rather than intermediate steps. To address the limitations of static sampling under evolving learning demands, we propose Curvature-Adaptive Consistency Flow Matching (CACFM). By formulating distillation as a dynamic decision process, CACFM uses a lightweight reinforcement learning agent to probe Probability Flow ODE trajectories and construct an efficiency-oriented curriculum that prioritizes critical regions without manual scheduling. Combined with Flow-adapted DMD and adversarial consistency objectives, our RL-based scheduler achieves state-of-the-art results on large-scale models such as FLUX and SDXL, mitigating structural deformities and preserving high-frequency details in extreme few-step regimes.
arXiv:2606.22683v2 Announce Type: replace
Abstract: Context: Reflection is a fundamental skill in software engineering education, particularly in project-based courses where students learn through extended group work and need to develop their ability to reflect iteratively throughout their work. For students to benefit from reflection, their written reflections need to be assessed so that feedback can guide and improve their reflective practice. However, manually assessing written reflections to guide reflections is time-consuming, and often results in broad, non-specific feedback for a student to improve. Objective: This study builds on reflective writing frameworks to produce an eight-indicator scheme for assessing student reflections in software engineering. Furthermore, this study validates an automated classifier for assessing reflections against the framework, enabling scalable and structured feedback whilst reducing instructor workload. Method: We adapted existing reflection frameworks through iterative refinement to create our eight-indicator framework. Three annotators labelled student reflection texts, establishing moderate to reliable inter-rater agreement. We then trained and evaluated multiple encoder-only transformer models and compared them with decoder-only large language models using zero-shot prompting. Results: The fine-tuned RoBERTa model achieved the strongest performance, substantially outperforming decoder-only models in both accuracy and speed. The classifier demonstrated human-level agreement on most indicators whilst enabling near-instantaneous classification. We provide two model variants optimised for different assessment priorities. Conclusions: Our fine-tuned encoder-only models enable efficient automated assessment of reflective writing. The framework and automated classifier offer a means to provide timely, structured feedback on student reflections in software engineering.
arXiv:2606.22881v2 Announce Type: replace
Abstract: LiDAR (Light Detection and Ranging) sensors capture the surrounding environment as dense 3D point clouds by measuring the time-of-flight of emitted laser pulses, making them foundational across autonomous vehicles, robotics, and large-scale mapping. PCAP (Packet Capture) files from these sensors are the starting point of most 3D perception pipelines, yet internal packet structures, UDP (User Datagram Protocol) port conventions and encoding schemes differ enough across manufacturers that no single tool reads them all. Ouster, Velodyne, Hesai, and Livox each require their own SDK (Software Development Kit), their own environment setup, and their own conversion workflow. Supporting all four means maintaining four disconnected pipelines with no shared infrastructure. The pipeline described here takes a raw PCAP as input and handles vendor identification automatically, scoring six independent file characteristics through a weighted multi-signal approach to determine the source sensor. C++ SDKs handle Ouster and Velodyne, while Hesai and Livox rely on Python-based dpkt parsing where no open source SDK exists. From there, a single command writes output to any of five industry-standard formats. We tested on real outdoor captures. Ouster peaks at 2.08M points per second, Velodyne at 1.47M, both running through native C++ packet decoding. Hesai and Livox land at 110K and 150K respectively, where Python-layer parsing introduces overhead that compounds under sustained load. The 8-10x gap held consistently across runs. Tested on a consumer-grade i3 with 8GB RAM, no vendor configuration required
arXiv:2606.23161v3 Announce Type: replace
Abstract: The minimum distance problem (MDP) for low-density parity-check (LDPC) codes is a central problem in coding theory and is closely related to the analysis of low-weight codewords and error-floor behavior. Although the unrestricted MDP is computationally intractable, its complexity under degree constraints that commonly occur in LDPC code design has remained less clear. In this paper, we study the MDP for left regular and biregular Tanner graphs. For every fixed $J\geq3$, we prove that the standard at-most-weight problem is $\mathrm{NP}$-complete for $J$-left regular Tanner graphs and that its exact-weight variant is $\mathrm{W}[1]$-complete when parameterized by the prescribed weight. For biregular Tanner graphs, we prove $\mathrm{NP}$-completeness for $(3,K)$-regular instances for every fixed $K\geq 3$ by replacing degree-two auxiliary completion blocks with a single-port high-girth gadget. A nonzero relative support inside this gadget induces an essentially cubic graph, so the Moore bound gives an exponential lower bound in the girth and allows a polynomial-size Karp reduction. Combining this right-degree amplification with a replica-and-global-check left-degree amplification yields $\mathrm{NP}$-completeness for $(J,K)$-regular Tanner graphs for every fixed $J,K\geq 3$. The reductions are based on a degree-preserving transformation framework consisting of hyperedge decomposition, check node splitting, and controlled variable replication. These transformations relate different degree distributions while preserving explicit maps among nonzero codewords, even covers, and nonempty $(a,0)$-trapping sets. The results delineate the computational limits of computing minimum distance exactly under natural regularity constraints.
arXiv:2606.25045v2 Announce Type: replace
Abstract: Metallic Magnetic Calorimeters (MMCs) are a promising new tool for high precision X-ray spectroscopy. However, the complexity of the detector response and the need for scalable processing pipelines pose significant challenges for their widespread adoption. In this work, we explore the application of Machine Learning (ML) methods to address these challenges and enhance the performance of MMCs. We demonstrate how ML can be used for pulse classification and artifact rejection, as well as for pulse shape analysis and feature extraction. By leveraging unsupervised learning techniques for label auto-discovery and supervised learning for classification and regression tasks, we show that ML can provide robust and scalable solutions for MMC signal processing. Our results indicate that ML-based approaches can achieve comparable performance to traditional methods while offering greater adaptability and efficiency, paving the way for the next generation of high-precision X-ray spectroscopy with MMCs.
arXiv:2606.25325v2 Announce Type: replace
Abstract: We find that current emotion-oriented Omni-MLLMs still lack reliable omni-modal perception: they (i) underutilize multimodal cues in their reasoning trajectories and (ii) exhibit unfaithful behavior, often hallucinating modality-specific statements from other modalities. Building on these insights, we propose OPPO (Omni-Perception Policy Optimization), a reinforcement learning framework that explicitly optimizes multimodal perception. First, an Omni-Perception Reward decomposes ground-truth reasoning into fine-grained visual, acoustic, and emotion cues and rewards trajectories that semantically recover these cues. Second, an Omni-Perception Loss compares the policy under full and unimodally masked inputs, applying a KL penalty only to modality-specific evidence tokens to suppress cross-modal hallucination. We further introduce MEP-Bench, a diagnostic benchmark that quantifies utilization and faithfulness. Experiments show that OPPO achieves state-of-the-art performance on MER-UniBench and MME-Emotion, while substantially improving utilization and faithfulness scores on MEP-Bench, highlighting the importance of sufficient and faithful omni perception for multimodal emotion reasoning.
arXiv:2606.25578v2 Announce Type: replace
Abstract: Hairstyle transfer has practical applications such as virtual try-on, yet remains challenging when the source and reference exhibit large head-pose discrepancies. We propose H-Adapter, which improves pose robustness by training with a region-specific loss that disentangles hair and non-hair objectives and thereby induces spatially disentangled cross-attention, from which a source-aligned hair edit mask is derived to guide diffusion-based inpainting. Experiments on pose-agnostic and pose-different subsets demonstrate strong quantitative results, including the best FID, $\mathrm{FID}_{\mathrm{CLIP}}$, and CLIP-I under pose differences, while maintaining competitive non-hair preservation and improving qualitative fidelity to fine-grained reference hairstyle details. Beyond source-conditioned transfer, H-Adapter supports practical extensions including text-to-image generation, auxiliary prompt-based hair color control, and compatibility with an identity-preserving IP-Adapter variant. We also introduce a VLM-as-a-judge protocol and observe consistent gains in hairstyle faithfulness, non-hair preservation, and artifact quality.
arXiv:2606.26454v2 Announce Type: replace
Abstract: By promoting vectors to spheres and enabling explicit model construction, neural networks can perform symbolic-level syllogistic reasoning without training data. We identify two fundamental limitations that prevent conventional data-driven machine learning systems from achieving this capability: training data generated by the combination table cannot distinguish all 24 valid syllogism types, and end-to-end premise-to-conclusion mapping creates contradictory targets within neural components. Experiments with two representative conventional systems, GPT-5 using linguistic inputs and Euler Net using visual inputs, support this analysis. ChatGPT GPT-5 may reach 100% accuracy in syllogistic reasoning, but with hallucinations. Because the learning process terminates upon reaching 100% accuracy, the system cannot progress beyond empirical accuracy to symbolic level reasoning. Random test data reduced Euler Net's accuracy to 56%. Repeatedly expanding the training set increased its accuracy to 97%, with perfect performance on 8 syllogism types. However, because unintended inputs cannot be exhaustively covered, even 100% test accuracy does not imply symbolic-level reasoning. Since syllogistic reasoning underpins logical reasoning and human rationality, these results suggest that increasing data and training time alone cannot ensure symbolic level logical reasoning.
arXiv:2606.26985v2 Announce Type: replace
Abstract: 3D Gaussian Splatting (3DGS) supports fast training and real-time rendering, but its optimization process remains difficult to interpret. Existing viewers mainly expose the final reconstructed scene and offer limited support for explaining how Gaussian properties contribute to visible artifacts or evolve during training. We present Vis4GS, a multi-view visual analytics tool for primitive-level diagnosis of 3DGS reconstruction artifacts. Built on the original 3DGS viewer and training framework, Vis4GS links rendered artifacts to Gaussian properties, View Coverage, training progress, and Gaussian genealogy through four linked views: an interactive Gaussian analysis view, a property timeline view, a Gaussian densification tree view, and a log and control panel. The system supports Gaussian selection, blur and needle-like artifact scoring, View Coverage analysis, and multiscale genealogy exploration of clone, split, prune, and clone-split events. By connecting scene-level artifacts with primitive-level evidence and optimization history, Vis4GS enables a structured workflow for diagnosing reconstruction failures beyond final-image inspection and global metrics. A user study also shows that Vis4GS provides stronger support for usability and artifact understanding than the original 3DGS viewer.
arXiv:2606.27088v2 Announce Type: replace
Abstract: Mesh subdivision is a fundamental operation for converting coarse, editable meshes into high-resolution surfaces, with broad applications in digital asset creation. Classical rule-based schemes rely on fixed local refinement rules and often produce over-smoothed surfaces. Recent neural subdivision methods improve detail synthesis, but remain constrained by local modeling and exhibit limited generalizability. We present SubdivAR, a neural mesh subdivision framework based on our proposed Mesh Autoregressive Representation (MAR). MAR arranges meshes at different subdivision levels into an ordered scale sequence, reformulating subdivision as autoregressive next-scale prediction. To support this formulation, we introduce a Hybrid Topology-Aware Transformer that combines global semantic attention with topology-constrained local feature aggregation. SubdivAR adopts a next-scale coordinate prediction paradigm, regressing vertex offsets at each refinement stage to preserve subdivision topology while recovering fine-grained geometric details. To enable reliable learning, we construct FII-40K, a curated dataset of nearly 40,000 high-quality meshes with multi-level subdivision supervision. Experiments show that SubdivAR outperforms state-of-the-art baselines, reducing Hausdorff Distance and Chamfer Distance by 18.8% and 14.2%, respectively, and demonstrates strong robustness on complex open-surface geometries.
arXiv:2606.27163v2 Announce Type: replace
Abstract: I describe my solution to the LeHome Challenge 2026, an ICRA 2026 competition on bimanual garment folding. The system placed 1st of 62 teams in the online (simulation) round and 2nd in the real-world final. It improves a vision-language-action (VLA) policy with a reinforcement-learning loop. The policy is its own value function: the same network that predicts actions also predicts success, progress, and a few task-relevant future quantities, and those predictions drive advantage estimation, live failure detection, and candidate selection. The work mostly recombines existing RL ideas with engineering and optimization contributions that can be used together as one recipe or individually: AWR + RECAP combined for flow-matching VLA; an asynchronous distributed training / rollout pipeline through HuggingFace Hub; inference-time hyperparameters optimization via Thompson sampling; a sim-to-real recipe with camera-alignment tooling, heavy augmentation and DAgger-like HIL data collection.
arXiv:2606.27412v2 Announce Type: replace
Abstract: 3D Scene Graph Generation (3DSGG) represents 3D scenes as structured object--relation--object graphs for spatial understanding. In observer-centric spatial perception, the same scene may be expressed under different local observer frames while its structure remains unchanged. However, existing models typically assume a fixed scene-aligned reference frame and may produce semantically inconsistent predictions when the scene is re-expressed in another observer frame. We attribute this failure to the heterogeneous frame dependency of relational predicates. Directional predicates such as $\textit{left}$, $\textit{front}$, $\textit{right}$, and $\textit{behind}$ are $\textbf{Observer-Dependent Relations}$, whereas most contact, support, and semantic predicates, such as $\textit{standing on}$ and $\textit{attached to}$, are approximately $\textbf{Observer-Independent Relations}$. Conventional models do not distinguish these frame responses, leading to degraded relation prediction under observer-frame reorientation. We introduce $\textbf{Observer-Aware Relations (OAR)}$, which combines observer-aware geometric encoding and relation specialization, supported by frame-stable object encoding, for unified multi-label predicate prediction. Experiments on 3DSSG show that OAR consistently outperforms baselines across controlled observer-frame reorientations without training-time frame-reorientation augmentation, while remaining competitive on the standard benchmark. The project page is available at https://oar-predicate.github.io/.
arXiv:2606.28079v3 Announce Type: replace
Abstract: The rapid proliferation of automated, multi-vector malware threats poses a significant risk to heterogeneous, resource constrained cyber-physical networks. Conventional epidemiological models often treat security defenses as static parameters, failing to capture the strategic, asymmetric maneuvers between an attacker and a defender. To address the gap, this paper proposes a Game-Theory-Integrated Modified Multi- Wireless Sensor Epidemic Malware Propagation (GTI-mSEMP) framework. This paper analyzed and compared the operational trajectories of Susceptible (S) and Recovered (R) node populations across three different operational regimes: Balanced Matchup, Exploit Surge and Hardened Defense. Numerical simulation results capture the real-time transient dynamics of the network state variables, demonstrating how the epidemic curve shifts when either the defensive or offensive scaling vectors hold an efficiency advantage. The proposed mathematical and numerical framework provides a rigorous foundation that can be deployed in highly adversarial network environments to evaluate dynamic malware propagation and predict localized node population states.
arXiv:2606.28196v2 Announce Type: replace
Abstract: Traditionally, dexterous manipulation controllers are designed using analytic models constrained by strong assumptions about the hand and the objects being manipulated. Reinforcement learning (RL) has become another common approach in which skills are explored openly in an end-to-end manner but is inefficient because of unnoticeable instability and conflicts in learning objectives. This paper attempts to efficiently explore stable and accurate manipulation skills by decomposing dexterous skills into multiple simpler/analyzable components. Each skill component is subsequently learned with constraints and guidance from classical physics and control theory. Our work shows that for stable grasp, in-grasp reposition/reorientation with different objects, sensor/motor noise, latency, and frictional conditions, skill learning becomes efficient and stable with prior knowledge from theory.
arXiv:2606.28534v2 Announce Type: replace
Abstract: The increasing demand for high-performance computing in plasma physics has driven scalable and resilient simulation methods capable of efficiently exploiting modern multi-GPU architectures. This work extends a portable hybrid MPI+OpenMP implementation of BIT1, focusing on high-performance resilience for accelerated Particle-in-Cell (PIC) Monte Carlo (MC) simulations under both uniform and non-uniform load conditions. Scalable particle load balancing and robust checkpoint/restart mechanisms across Nvidia and AMD accelerators are integrated with standardized I/O using openPMD and ADIOS2. This leverages BP4 for high-performance file-based checkpointing and SST for in-memory data streaming, enabling efficient data movement, resilient large-scale execution, seamless continuation from existing checkpoints, and effective handling of computational and I/O workloads. Advanced HPC profiling and tracing tools, including Nvidia Nsight Systems and AMD ROC-Profiler with Perfetto, provide detailed insights into computation, communication, and system-level behavior for optimization. Performance results on Frontier (OLCF-5), MN5, and LUMI-G demonstrate strong and weak scaling up to 800 GPUs, validating the framework for large-scale PIC MC simulations, while in-situ analysis and visualization using scalable I/O further enhance scientific insight without interrupting multi-GPU execution on current and future exascale systems.
arXiv:2606.30868v2 Announce Type: replace
Abstract: Two observers of the same physical system may differ in gauge by a group element acting on their common vector representations, and recovering that element from finite, noisy paired observations is useful in both theory and experiment. The Kabsch and Horn algorithms solve this problem for rotated frames in $\mathbb R^3$ (i.e. $SO(3)$); our earlier Lie algebra method extends it to the Lorentz group $SO(3,1)_+$. Here we report explicit formulae for the Lie algebra method on the classical matrix Lie groups ($GL(n)$, $SL(n)$, $SO(n)$, $U(n)$, $SO(p,q)$, $Sp(n)$, $Spin(n)$, and $SE(n)$) over the real and complex fields. The four steps (pseudoinverse, matrix logarithm, projection onto the Lie algebra, matrix exponential) are exact for noiseless data. Only the projection is group-dependent, and we show it yields the unique least squares-optimal element of the Lie algebra whenever its image lies in $\mathfrak g$ and its residual is orthogonal to $\mathfrak g$. For noisy data the method is optimal only to leading order, so we add a quasi-Newton correction whose accuracy interpolates between the uncorrected method and direct least squares optimization. The projections, their optimality, and the identity underlying the correction are formally proven in Lean 4.31.0 with Mathlib, and numerical experiments are benchmarked in Julia.
arXiv:2606.32017v3 Announce Type: replace
Abstract: Agentic reinforcement learning requires assigning credit to environment-facing actions such as searches, clicks, edits, navigation commands, and object interactions. Standard GRPO uses the final verifier outcome as a uniform advantage over all action tokens. This outcome signal is useful but structurally incomplete: it punishes useful exploration in failed rollouts and reinforces redundant or regressive actions in successful rollouts. We propose TRIAGE, a role-typed credit assignment framework that adds a semantic role axis to outcome credit. A structured judge classifies each segment as decisive progress, useful exploration, no-progress infrastructure, or regression, and a fixed role-conditioned rule maps these labels to bounded segment-level process rewards. This keeps verifier outcomes as the source of optimization direction while correcting the two main blind spots of outcome-only credit. We further show that the Bayes-optimal role-measurable correction is the L2 projection of the per-segment advantage residual onto the role variable, and that TRIAGE's fixed role constants approximate this projection, reducing advantage estimation error whenever the judge is reliable; we connect this to lower-variance policy gradients. Across ALFWorld, Search-QA, and WebShop, TRIAGE improves success rates over GRPO for two policy models and outperforms both a scalar judge-derived process reward and an outcome-supervised shared-backbone value baseline. Ablations show that the gain comes from role typing rather than merely adding dense rewards: reliable detection of regression inside successful trajectories is the dominant contributor, while exploration credit provides a consistent secondary gain; on completed ALFWorld and WebShop rollouts, TRIAGE also reduces environment-facing turns by an additional $10.4\%$ and $14.8\%$ relative to GRPO.
arXiv:2607.00573v2 Announce Type: replace
Abstract: Diffusion MRI probes brain microstructure with particular sensitivity to early cerebrovascular and neurodegenerative changes. Neurite Orientation Dispersion and Density Imaging (NODDI) decomposes the diffusion signal into three biophysically interpretable maps: neurite density index (NDI), orientation dispersion index (ODI), and free water fraction (FWF), capturing neurite packing, fiber coherence, and extracellular fluid. These 3D maps offer a rich substrate for transferable microstructural representations, yet integrating them is challenging: standard representation learning struggles to disentangle the unique information in each map from their shared and synergistic interactions. We present BrainFIBRE, the first foundation model for brain microstructure, pretrained on NODDI-derived maps from 55,592 UK Biobank participants. We propose Self-supervised Partial Information Decomposition (SPID), which extends PID-guided multimodal learning to the self-supervised regime for the first time. A novel Counterfactual Candidate Construction (CCC) paradigm perturbs inter-modality alignment through modality dropping and swapping, providing the contrastive signal for a Mixture-of-Experts architecture to disentangle unique, synergistic, and redundant information without any downstream label. On both Caucasian and Asian cohorts, BrainFIBRE achieves state-of-the-art performance across diverse tasks predicting age, sex, cerebrovascular and neurodegenerative markers, and cognition, while yielding neurobiologically interpretable representations that reveal task- and cohort-specific interaction patterns. BrainFIBRE establishes a versatile foundation for neuroimaging analysis at the microstructural level.
arXiv:2607.02104v2 Announce Type: replace
Abstract: Large language models (LLMs) are increasingly used as cheap, scalable judges that compare candidate outputs pairwise. Because such judges prefer verbose or well-formatted answers, the natural fix is to add bias covariates to a Bradley--Terry model and estimate the bias away. We show this cannot work as advertised: the quality/bias split is \emph{not identified} by pairwise comparisons, and the failure is exact -- across $48$ real judge-pools the profile likelihood over the coefficient is flat to $\mathbf{0.0000}$ \textbf{nats}, and scaling the comparisons $26\times$ buys none. A ``debiased'' score is selected by the prior, not recovered from data. Our contribution is accordingly not a better estimator but a characterization of \emph{when prior-based correction is justified}, plus designs that supply the missing information when it is not. The assumption the prior encodes -- quality is a priori uncorrelated with the covariate -- pays only while $\mathrm{corr}(\theta,x)$ stays below a crossing point (configuration-dependent, $0.22$--$0.60$), which is what makes the same model help on LLMBar and hurt on SummEval and Nectar. We give two escapes: a \textbf{trusted-anchor gate} that decides per (judge, covariate, task) (no false enables in $6{,}000$ decisions at $K\ge10$ anchors, a rate our sample bounds at $\le6\%$), and a \textbf{paired rendering design}. Across fifteen real LLM judges bias is heterogeneous and capability-dependent: correction improves \topk{} recall by $0.20$--$0.32$ on five biased-but-competent cheap judges and is a no-op on frontier ones (Spearman $\rho{=}{-}0.84$ between competence and gain over the $14$ competent judges, $p{<}10^{-3}$), concentrating the benefit where at-scale evaluation happens.
arXiv:2606.06547v2 Announce Type: replace
Abstract: Diffusion Large Language Models (dLLMs) refine tokens iteratively but commit them irreversibly, leading to a "stability lag" where early decisions remain fragile even after being written. We reveal that Post-Training Quantization (PTQ) error easily flips these borderline decisions at the write frontier, which are then permanently locked in and amplified. To address this, we propose Frontier-Aware Instability-Reweighted Calibration (FAIR-Calib), a two-stage PTQ framework for dLLMs. Stage I probes a full-precision teacher to estimate a position prior that combines frontier hits and masked-stage reliability. Stage II performs off-policy, layer-wise calibration by minimizing a reweighted hidden-state MSE, effectively prioritizing the protection of fragile frontier states without requiring expensive end-to-end diffusion rollouts. We further theoretically justify our weighted objective as a surrogate for output KL divergence. Empirically, FAIR-Calib consistently outperforms state-of-the-art baselines on LLaDA and Dream (W4A4), significantly reducing frontier decision flips and suppressing post-commit mismatches across diverse benchmarks.
arXiv:2604.09497v2 Announce Type: replace
Abstract: Accurate evaluation is central to the large language model (LLM) ecosystem, guiding model selection and downstream adoption across diverse use cases. In practice, however, evaluating generative outputs typically relies on rigid lexical methods to extract and assess answers, which can conflate a model's true problem-solving ability with its compliance with predefined formatting guidelines. While recent LLM-as-a-Judge approaches mitigate this issue by assessing semantic correctness rather than strict structural conformity, they also introduce substantial computational overhead, making evaluation costly. In this work, we first systematically investigate the limitations of lexical evaluation through a large-scale empirical study spanning 36 models and 15 downstream tasks, demonstrating that such methods correlate poorly with human judgments. To address this limitation, we introduce BERT-as-a-Judge, an encoder-driven approach for assessing answer correctness in reference-based generative settings, robust to variations in output phrasing, and requiring only lightweight training on synthetically annotated question-candidate-reference triplets. We show that it consistently outperforms the lexical baseline while matching the performance of much larger LLM judges, providing a compelling trade-off between the two and enabling reliable, scalable evaluation. Finally, through extensive experimentation, we provide detailed insights into BERT-as-a-Judge's performance to offer practical guidance for practitioners, and release all project artifacts to foster downstream adoption.