Search arXiv⌕ Search

arXiv subjects

Han Wang

Publications and source records attributed to Han Wang.

At least 19 recordsLinked to original sources
arXiv2026

MLE-World: Predicting Machine Learning Experiment Outcomes with a Learned World Model

AI research agents can propose, execute, and improve programs that solve machine learning tasks. However, executing each candidate program requires training and evaluating ML models, limiting how many candidates an agent can explore at each step. We introduce MLE-World, a world model finetuned from Qwen3.5-4B that predicts a program's terminal output (error traces, training logs, and validation scores) from its code and data description without executing it, together with a confidence score estimating the prediction's correctness. Because a prediction costs one inference call rather than a training run, MLE-World makes test time scaling (TTS) affordable for these agents. We build five TTS policies, guided by MLE-World, that compare several candidates per step by their predicted outputs: Sequential revision, Best-of-N, Pairwise judging, Fusion, and a UCB style rule that also adds an exploration bonus from this confidence score. Since predictions can be wrong, we run a candidate for real only when its confidence score is low. On 17 MLE-bench Lite tasks, MLE-World matches the real terminal output on 77.9% of held out programs, versus 41.1% for much larger prompted models. MLE-World guided TTS achieves up to 22.1% higher observed mean performance than the agent without a world model, with less real execution. Project webpage: https://mle-world.github.io/.

cs.LG↗
arXiv2026

Learning Without Clocks: Partition-Invariant Neural Network

Event cameras, spiking cochleas and similar sensors emit asynchronous events, yet most learning systems first bin them into frames or packets, so the output depends on how the stream was sliced. Networks whose state is updated exactly at every event and evolves in closed form between events are invariant to this slicing. We ask what more makes the output (i) independent of the processing order of simultaneous events and (ii) continuous in the time stamps, even when two events cross in time, while (iii) staying sensitive to temporal order. For arbitrary nonlinear event models we prove that (i) and (ii) are each equivalent to pairwise commutation of the event updates, and that once updates commute, all order dependence comes from their non-commutation with the inter-event dynamics and vanishes linearly in the time lag. This identifies a class of time-mediated non-commutative (TM) models and yields a resolution-robustness bound for any model. We construct the minimal TM layer, a learnable, provably Lipschitz generalization of the Hassenstein-Reichardt correlator; prove that no single additive layer can uniformly represent ordered pair interactions, whereas a three-dimensional TM state can; show that each TM layer adds two levels of temporal nesting versus one for additive layers; and give a single-layer higher-order construction (TM-k) and a stacking rule preserving all three properties. We instantiate these results as the Clock-Free Event (CFE) layer, linear in the number of events, give a topology-free variant for temporal point processes with closed-form likelihoods, and specify a preregistered clock-robustness protocol with a physical-time threat model for spike-retiming attacks. In preliminary development-set experiments, the TM state beats an otherwise identical additive state by 3.5-9.1 points on DVS128 Gesture and by 24.5 and 30.6 points on SHD and Speech Commands.

cs.LG↗
arXiv2026

CrystalREPA: Transferring Physical Priors from Universal MLIPs to Crystal Generative Models

Crystal generative models mainly learn what stable crystals look like, with little explicit supervision for what makes them stable. We reveal a substantial representation gap between state-of-the-art crystal generative models and pretrained universal machine learning interatomic potentials (MLIPs) via energy probing, and show this gap can be closed by a simple training-time alignment. We propose Crystal REPresentation Alignment (CrystalREPA), a plug-and-play framework that aligns the atom-wise hidden states of generative encoders with frozen MLIP representations through an element-aware contrastive objective, transferring stability-aware atomistic priors with marginal training overhead and no additional inference cost. Across three generative frameworks, ten MLIP teachers, and two benchmark datasets, CrystalREPA consistently improves the thermodynamic stability, structural validity, and structural fidelity of generated crystals. Equally important, we find that an MLIP's transfer effectiveness is poorly predicted by its accuracy on standard leaderboards (e.g., Matbench Discovery) but strongly predicted by the distinguishability of its atom-wise representation space, yielding a practical, accuracy-independent criterion for selecting MLIP teachers for generative transfer. Code is available at https://github.com/Chengqian-Zhang/CrystalREPA.

cond-mat.mtrl-sci↗
arXiv2026

Training an MoE LLM in MXFP4: How We Closed the Gap to BF16

Training large language models (LLMs) in sub-8-bit arithmetic promises substantial gains in memory bandwidth and compute throughput, yet FP4 formats introduce quantization error severe enough to destabilize optimization if left unaddressed. We present ALTO, an open-source training recipe for GPT-OSS-20B, a 20-billion-parameter mixture-of-experts (MoE) model, on the MLPerf Small MoE Training benchmark under the MXFP4 microscaling format. The recipe combines established techniques from Abecassis, Felix et al. [arXiv:2509.25149] (hybrid 1D/2D block quantization, randomized Hadamard transforms, and stochastic rounding) with a simplified weight de-oscillation scheme adapted from TetraJet-v2 [arXiv:2510.27527]. On C4 validation at global batch size 16, MXFP4 + RHT + SR + de-oscillation reaches a validation loss of 3.3350 at 16,128 steps, only 0.007 above the BF16 baseline (3.3283), closing roughly 50% of the gap left by MXFP4 without de-oscillation (3.3418). Measured by steps to reach the MLPerf quality target (validation loss 3.34), de-oscillation cuts the FP4 convergence overhead relative to BF16 from +20.0% to +5.0%. We also report negative end-to-end results from differential gradient estimation, outlier clipping, and macro-block scaling, despite each showing operator-level SNR gains. All experiments use simulated MXFP4 kernels on AMD MI300X hardware and wall-clock speed has not been optimized.

cs.AI↗
arXiv2026

Revealing Imaging Characteristics of Thermal Wakes from Underwater Vehicles Without Cooling-Water Discharge

Small uncrewed underwater vehicles are proliferating and threaten maritime security, yet evade sonar and magnetic sensors built for larger submarines. Infrared thermal wakes, generated by hydrodynamic displacement that modulates the cold-skin layer, offer a non-acoustic, non-magnetic alternative even without cooling-water discharge. Existing studies, however, rely on idealized simulations that fail to capture real-world backgrounds, offering limited guidance for target-background discrimination. We establish AVP-IR-Wake, the first large-scale, real-world, physically diverse benchmark of 161,485 frames collected over three years. We show that thermal wakes are extended structures of discrete vortices, violating the compactness assumption of conventional targets. We reveal local indistinguishability: individual wake vortices are statistically identical to wind-induced background vortices in temporal, frequency, and sparse domains, sharing low-frequency information and long-range correlations. This challenges the assumptions of compact, transient, high-frequency targets in conventional detection. Yet we establish global morphological separability: although locally inseparable, the wake as a whole is organized by vehicle trajectory, whereas background clutter is organized by wind-driven thermal fronts. This explains why representative detectors fail and shows that reliable detection must shift from local feature extraction to global trajectory-constrained physical modeling. Because the cold-skin layer varies globally and seasonally, forecasting wake observability will require integrating physical oceanography and atmospheric science, linking underwater surveillance to the changing ocean surface.

cs.CV↗
arXiv2026

SpecAgent: Empowering Program Verification with Agentic Synthesis of Formal Program Specifications

Formal specifications are essential for deductive program verification, providing semantic abstractions for compositional verification of complex software. However, manually constructing specifications is labor-intensive, motivating automated synthesis. Despite recent advances in large language models (LLMs), existing approaches often rely on forward-only workflows and localized repair, limiting their effectiveness on real-world programs with complex dependencies among functions and loops. Verification failures may stem from previously generated specifications, causing cascading failures that local refinement cannot resolve. To address these challenges, we present SpecAgent, an agentic framework for synthesizing high-quality ACSL specifications for real-world C programs. SpecAgent integrates four components: dependency-aware planning to identify specification targets, retrieval-augmented generation (RAG) to provide relevant specification patterns and program context, agentic repair to diagnose defects and revisit dependent specifications, and agentic critique to assess and refine semantic strength beyond proof success. We evaluate SpecAgent on specification synthesis and program verification tasks. On 50 programs from 14 real-world repositories, SpecAgent with DeepSeek-V4 achieves 96.82% precision and 88.95% recall in synthesizing correct and strong specifications, outperforming all baselines. On 811 real-world verification targets, it successfully discharges 604, also surpassing existing baselines. These results demonstrate SpecAgent's effectiveness in synthesizing high-quality specifications and facilitating real-world program verification.

cs.SE↗
arXiv2026

Where Prefill Expert Reduction Hurts MoE Models: A Few Prompt Tokens Carry Most of the Short-Answer F1 Loss

Running prompt tokens through fewer routed experts during prefill makes mixture-of-experts (MoE) serving cheaper, but what it damages has not been measured. We locate the damage with token-, layer- and attention-path interventions in four MoE families on short-answer question answering. At half the experts the F1 loss is mostly answer form: the model still finds the answer but no longer answers briefly. The loss is made at a few tokens: the two chat-template tokens before the answer in Qwen3-30B-A3B, and the first 8-32 tokens, where the instruction sits, in DeepSeek-V2-Lite. Keeping them at full experts removes most of the F1 loss and about half and a fifth of the increase in answer NLL. These tokens are not the familiar suspects: reduction leaves the first-token sink within 4%, and restoring super experts or massive activations recovers at most a quarter of the loss. On real prompts the localisation becomes a placement rule: with about 4% of tokens at full experts, putting them on the task interface keeps F1 within one point of the unmodified model, while the same budget scattered in the documents is about 3 points worse in both prespecified primary tests of this comparison. Time to first token falls about 14% on 11K-token prompts.

cs.CL↗
arXiv2026

HybridResearcher: Bridging the Structured-Unstructured Boundary for Deep Research Agents

While LLM agents are increasingly capable of gathering and synthesizing evidence from a single homogeneous source, such as structured tables or unstructured data, real-world analytical questions frequently span heterogeneous sources. For example, answering one enterprise question may require both a result computed over an internal database and evidence found on the web. Recent evaluations on benchmarks like HybridDeepResearch show that even frontier agents struggle when evidence must cross this boundary. While these benchmarks have exposed the limitation, methodologies to explicitly teach cross-source reasoning remain absent. To address this, we introduce HybridResearcher, a 4B agent trained to bridge evidence across the structured and unstructured boundary. To build it, we synthesize cross-source bridge trajectories by starting with validated single-source tasks-such as existing text-to-SQL or web search queries-and fabricating the missing source around a bridge entity drawn from the seed's own output. We retain a sample only when strict construction gates confirm that neither side alone suffices, ensuring the data requires true cross-source interdependence. Finally, we train on this curated dataset using supervised fine-tuning followed by reinforcement learning. Evaluated on the existing HybridDeepResearch benchmark, HybridResearcher improves over our base model, Qwen3.5-4B, by 30.0 points pass@8 and surpasses the far larger Qwen3.5-397B-A17B by 8.3 points, while also improving over its base model on established, held-out single-source benchmarks (BrowseComp and BrowseComp-Plus), demonstrating that this complex cross-source reasoning can be taught directly to small models without triggering catastrophic forgetting.

cs.AI↗
arXiv2026

Turning-Point Impact Compromises the Biomechanical Integrity of Cortical Bone during Oscillating Sawing

Purpose Oscillating bone sawing is widely used in orthopedic procedures such as total knee arthroplasty, where preservation of the mechanical integrity of the prepared bone surface is essential for implant stability. However, the reciprocating motion of oscillating saws can generate transient impact events near trajectory turning points, potentially inducing surface damage and reducing local load-bearing capacity. This study aimed to investigate the biomechanical consequences of turning-point impact loading during oscillating bone sawing. Methods Oscillating sawing experiments were performed on bovine cortical bone using a conventional trajectory and an optimized trajectory designed to suppress turning-point impacts. Dynamic sawing forces were measured to quantify impact events. Surface damage on the groove wall surface was characterized using scanning electron microscopy (SEM) and optical imaging. Local and global mechanical responses were evaluated using compression tests. Results The conventional trajectory produced pronounced periodic force spikes and generated extensive surface cracks and breakout regions. These impact-dominated events were associated with reduced local compressive strength of the prepared bone surface. In contrast, the optimized trajectory significantly suppressed peak impact forces, resulting in smoother groove wall surfaces with fewer defects. Consequently, the optimized trajectory increased the local compressive strength (p < 0.001) and improved the global compressive strength of the specimens (p = 0.001), while the elastic modulus remained largely unchanged. Conclusion Turning-point impacts during oscillating bone sawing compromise the biomechanical integrity of cortical bone by promoting surface damage and reducing local load-bearing capacity. Suppressing these transient impacts helps preserve the mechanical integrity of surgically prepared bone surfaces.

q-bio.TO↗
arXiv2026

Atomistic prediction of magnetic behaviour across chemistry, structure and temperature

Controlling magnetic phases and properties through composition and external conditions is central to the development of magnetic materials. Atomistic foundation models offer a route to predicting such behaviour across diverse materials, but magnetic materials remain challenging because their atomic structure and local moments are strongly coupled, while temperature and disorder drive magnetic moments into non-collinear states with evolving orientations and magnitudes. Broadly pretrained atomistic models still sample only limited magnetic diversity, while finite-temperature descriptions of coupled lattice and magnetic fluctuations remain largely material-specific. Here we introduce MagPT, pretrained across 48 elements on 912,492 spin-lattice configurations spanning structural, orientational and longitudinal magnetic fluctuations over 0 $\sim$ 1,000 K and 0 $\sim$ 10 GPa. By treating species, atomic positions and vector moments together, MagPT distinguishes magnetic states that share the same atomic structure but differ in spin configuration and energy. It captures magnetic behaviour across composition, temperature and length scales, from Slater-Pauling-type composition-magnetism relations in transition-metal alloys and finite-temperature phase evolution in elemental ferromagnets and magnetic shape-memory Heusler alloys, to emergent long-wavelength non-collinear spin textures in chiral magnets and local chemistry-magnetism coupling in high-entropy alloys. MagPT moves atomistic studies of magnetic materials beyond material-specific approaches, establishing a general framework for exploring and designing magnetic matter across diverse chemistries and conditions.

cond-mat.mtrl-sci↗
arXiv2026

Cracking Gravitational Wave Multiple Ringdown Modes in Space

Ringdown signals from perturbed black holes (BHs) offer a clean window into BH spacetime, strong-field gravity, and fundamental physics. Presently the quasi-normal modes of stellar-mass BH ringdowns have been successfully extracted in the ground-based gravitational wave (GW) observations. Looking ahead, the future space-borne observatories will listen to the ringdowns from massive BH binary coalescences more loudly and resolve multiple modes to unprecedented precision, which calls for efficient approaches to mitigate the sharply increasing computational burden. We develop a practical ringdown analysis pipeline for space-borne detectors by implementing FIREFLY, a novel acceleration algorithm validated in ground-based detectors, and for the first time demonstrate its compatibility and effectiveness with the time-delay interferometry (TDI) observables. With high fidelity, we achieve an over two-order-of-magnitude speedup for a simulated ringdown signal including six modes, providing a viable and scalable route for multi-mode ringdown analysis in the space context. This new approach has sound statistical interpretation and is extensible to other GW sources in band.

gr-qc↗
arXiv2026

What Matters in Preference Post-Training of Continuous Autoregressive Speech Language Models

Continuous autoregressive speech language models combine an autoregressive (AR) backbone with a diffusion or flow-matching head. Preference post-training requires choosing how to construct pairs and which modules to update. Controlled interventions show that a perturbed acoustic patch changes subsequent backbone states, yet larger representation changes do not necessarily reduce speech quality. We construct preference pairs through noise or inference-configuration perturbations and formulate a preference update based on reference-relative flow-matching errors. Updating the backbone with a fixed head yields stronger reward improvements than head-only adaptation in our evaluated setting. Across flow-time and module ablations, the lowest training loss does not consistently produce the best speech. These findings support preference post-training that considers both acoustic sampling and autoregressive conditioning, with training choices validated on generated utterances.

cs.SD↗
arXiv2026

SpecSyn: LLM-based Synthesis and Refinement of Formal Specifications for Real-world Program Verification

Program verification is a formal technique to rigorously ensure the correctness and fault-freeness of software systems. However, constructing comprehensive interprocedural specifications for full verification obligations is time-consuming and labor-intensive, giving rise to automated specification generation approaches. Despite the significant advancements in these approaches brought by Large Language Models (LLMs), existing LLM-empowered approaches still suffer from significant limitations: they lack effective strategies for handling sizable input programs, and are typically equipped with no mechanisms to evaluate and guarantee the strength of the generated specifications. The limitations impair their ability to extract precise specifications from real-world complicated programs to support the verification of target properties, thereby hindering the applicability of existing approaches in verification tasks on real-world programs. To remedy this gap, we propose SpecSyn, a novel LLM-based specification generation method. SpecSyn first decomposes the input program into individual segments, which are handled respectively by the subsequent iterative specification generation process. Innovatively, we incorporate into the process a specification refinement mechanism based on semantic-non-equivalent program mutations and variant discrimination, assessing and enhancing the semantic strength of the generated specifications. Extensive experiments show that SpecSyn maintains high precision over 90% and outstanding recall over 75%, significantly outperforming existing LLM-based approaches. In further evaluations, SpecSyn successfully handles 1022 out of 1421 target properties for open-source programs, proving its applicability on real-world program verification tasks.

cs.SE↗
arXiv2026

Genesis: Visual Augmentation Primitives for Multimodal Chain-of-Thought Reasoning

Recent multimodal large language models achieve promising results on visual reasoning tasks, approaching human-level performance on existing STEM benchmarks. However, these capabilities degrade under complex real-world scenes. We identify that chain-of-thought reasoning remains largely confined to the linguistic space, rarely anchoring each step in visual evidence. The main cause is the absence of such data during training. In this work, we introduce a training data augmentation pipeline, \textbf{Visual Augmentation Primitives} (VAP), which leverages state-of-the-art vision generative models to synthesize diverse photorealistic training samples and corresponding reasoning text. VAP introduces some basic operators and applies them to a seed image and gets a variant. It also constructs coherent multimodal reasoning chains throughout the generation process. Such training data force models to perceive visual content and perform logical reasoning according to them at every step. We further introduce multimodal chain-of-thought rubric-based reinforcement learning to enhance this. It rewards not only text reasoning but also visual reasoning by scoring intermediate reasoning steps for logical validity, completeness, and consistency. Moreover, we construct the UNREAL benchmark to systematically evaluate visual reasoning on hyperrealistic scenes across controllable difficulty levels. Experiments show that a 4B model built on this framework surpasses 8B baselines, with performance improving consistently as the model scales to 32B, validating the scaling law for visual reasoning.

cs.CV↗
arXiv2026

Learning Electron Density with Virtual Atoms for Transferable Atomistic Representations

Electron density is a fundamental quantity of electronic structure, yet most machine learning density models treat it solely as a reconstruction target and are developed separately from machine learning interatomic potentials. Here, we introduce a virtual-atom formulation within the DPA3 framework, in which density query points are represented as virtual nodes and processed by the same parameterized encoder used for real atoms. An asymmetric neighbor construction allows virtual atoms to receive information from nearby real atoms without altering the real-atom interaction graph. The resulting framework supports density, energy, and force prediction through task-specific readout heads. The proposed model achieves state-of-the-art density reconstruction accuracy on the Materials Project benchmark, matches the best reported performance on nickel-manganese-cobalt battery cathode materials, and is competitive with dedicated density models on QM9. When density labels are used jointly with energy and force supervision, the model improves in-domain accuracy, makes local environment information more accessible under linear probing, and reduces zero-shot force errors on larger QM40 molecules. Density-only checkpoints also provide useful initializations for downstream energy-force learning under the fixed fine-tuning protocol. Together, these results support electron density as a source of electronic-structure supervision for interatomic-potential representations.

physics.chem-ph↗
arXiv2026

NucleoNet: A Nuclear-Aware Determinant Ansatz for Multi-Geometry Neural Variational Monte Carlo

Neural-network variational Monte Carlo provides an accurate route to solving the electronic Schrödinger equation, but many neural wavefunctions are optimized for a single nuclear geometry, requiring separate calculations to map out a potential energy surface. We introduce NucleoNet, a nuclear-aware determinant ansatz in which electrons and nuclei form a unified particle set propagated through a shared electron--nucleus backbone that jointly evolves one-particle and two-particle representations. Nuclei carry learned representations that evolve alongside the electronic representations and condition orbital construction, while antisymmetry is imposed only by electronic determinants. On equilibrium small-molecule benchmarks, NucleoNet achieves accuracy comparable to established neural-wavefunction baselines with substantially fewer parameters. For each molecular system, a single model optimized jointly across geometries describes the N$_2$ dissociation curve, Li displacement in LiCN, and the ethane and formamide minimum-energy paths for bond dissociation. Sparse-geometry training generalizes to unseen geometries for N$_2$ and LiCN without further optimization, and N$_2$ ablations show that nuclear streams substantially improve this interpolation. These results show that a compact wavefunction with jointly evolved nuclear representations can describe the geometries of a molecular system with a single set of parameters.

physics.chem-ph↗
arXiv2026

FORTE: A Production Framework for Feature Evaluation and Deprecation in Industrial Recommender Systems

Production recommender systems have feature-lifecycle asymmetry: adding new features must produce gains on some metric, while feature removal must not regress any metric. In practice, feature deprecation is hard to prioritize as a feature can fail to be removed if it interacts with other features, its importance changes daily, it cannot be evaluated before production models change or data drift occurs, it is in contention for A/B test space with other launches, it causes regressions in an online A/B test, or it is consumed by multiple models that need to coordinate its deprecation. We present FORTE, a model-agnostic production framework for discovering, validating, and removing features. Its efficient perturbation implementation allows daily feature evaluation. To account for interactions, it evaluates candidate feature sets constructed from all the low-rank features through grouped ablation and its offline results are sufficiently predictive of online A/B tests to allow deprecations to proceed to online testing with confidence. Lineage tracing identifies all downstream consumer models to realize the full savings of removing a feature from the training set. We validate FORTE over 2 large-scale recommender systems and find that it identifies deprecable feature sets which do not cause online regressions, while the impact on efficiency depends on model characteristics. On a search-relevance model, FORTE resulted in 3 nested candidate sets of 32, 48, and 59 features. Online guardrails rejected the largest treatment. The 48-feature set was selected for deprecation, removing 43.3% of the model's features. On a large ranking model, FORTE identified 8 candidates for removal. Retraining and A/B testing confirmed no regression. Its launch reduced the trainable parameters by 29.9%, improved serving throughput by 19.8-25.5%, and reduced annual training costs by hundreds of thousands of dollars.

cs.LG↗
arXiv2026

On the Steering Dimensionality of Refusal in Language Models

Existing activation steering methods often assume that a high-level concept can be mediated by a single steering direction. To support this, two complementary interventions should be achieved: additive steering should induce the target behavior, while directional ablation should suppress it. Yet behaviors may occupy richer activation geometries beyond a single direction, and semantically similar behaviors may be represented by distinct directions. In this work, we study how many directions can reliably control two different types of refusal behaviors: refusal triggered by the safety alignment and refusal in general contexts. Given the limited expressive capability of a single steering vector, we study the general setting of steering subspaces and introduce the notion of steering dimensionality as the minimum subspace dimensionality required to reliably control a behavior. We characterize sufficient steering subspaces that cover the full extent of the target behavior through both the (monotonic) improvement before the sufficient dimensionality, and the saturation beyond it. Empirically, we find that refusal triggered by safety alignment is 1-dim steerable, while multiple distinct steering directions can achieve comparable control. In contrast, refusal in general contexts exhibits substantially richer activation geometry where even 5-dim steering subspaces fail to reliably capture its full steerable variation. Our results reveal that the activation geometry underlying refusal is highly context-dependent and can be substantially more complex than a single linear steering direction.

cs.AI↗