arXiv subjects

Yang Zhang

Publications and source records attributed to Yang Zhang.

At least 19 recordsLinked to original sources

arXiv2026

Select-TTT: Query-Conditioned Selective Fast Weight Updates for In-Place Test-Time Training

Test-time training (TTT) enables a deployed language model to internalize incoming context on-the-fly into a small set of fast-weight parameters, rather than merely treating it as transient context for attention. In-Place TTT recently made this idea practical for large language models by repurposing the down-projection of standard MLP blocks as fast weights. However, we find that its benefit largely disappears on realistic web long-context question answering. We attribute this to In-Place TTT's uniform write schedule: every chunk is written with equal strength, so irrelevant content dilutes query-relevant information in fast-weight memory. Building on this insight, we propose Select-TTT, a model-agnostic, query-conditioned rule for In-Place TTT that scales each chunk's write strength by its relevance to the user query. This relevance signal requires no supervision and no new parameters: it is read directly from the model's frozen output head, as the probability mass that a chunk's hidden states assign to the query tokens. A temperature-scaled softmax turns these scores into learning-rate multipliers, leaving the chunk-wise, context-parallel computation of In-Place TTT intact. Select-TTT consistently improves long-context question answering on two benchmarks and two backbones. These findings highlight the value of query-conditioned fast weight writes for practical test-time training on long-context question answering, which is also potentially promising for long-horizon agent tasks involving multiple user queries over evolving contexts.

cs.LG↗

arXiv2026

Probing the singlet-triplet scotogenic model with lepton flavor violation, gravitational waves and dark matter

We perform a comprehensive phenomenological analysis of neutrino masses, lepton-flavor violation, gravitational-wave signature induced by the first-order electroweak phase transition, and scalar dark matter within the framework of the singlet--triplet scotogenic model. We derive general analytical expressions for the one-loop Wilson coefficients of the penguin and box contributions to lepton-flavor-violating processes, considering generic scalar and fermion mediators. These general results are then applied to the singlet--triplet scotogenic model to calculate the branching ratios of $μ\to eγ$ and $μ\to e\overline{e}e$, as well as the coherent $μ$--$e$ conversion rate. We systematically investigate the lepton-flavor-violation phenomenology and the constraints from current and future experiments. It is found that the coupling $λ_5$ plays a key role in connecting neutrino masses, lepton flavor violation, and dark matter direct detection. We also study the finite-temperature effective potential and the gravitational-wave signatures of the resulting first-order phase transition. By analyzing the thermal history of the model, we identify the viable parameter space allowing for a first-order electroweak phase transition in the early Universe and find that the portal parameters $λ_1^Ω$ and $μ_1$ play a crucial role in determining the transition strength. The corresponding stochastic gravitational-wave spectra can be probed by future gravitational-wave observatories. As a representative example, we identify a benchmark point that simultaneously satisfies the current constraints from lepton-flavor-violation and dark-matter experiments and generates a stochastic gravitational-wave spectrum that may be probed by the Laser Interferometer Space Antenna, the Big Bang Observer, and the Ultimate Deci-hertz Interferometer Gravitational-wave Observatory.

hep-ph↗

arXiv2026

Rollback: Experience Backtracking in World Models for Policy Improvement

Action-conditioned world models offer a scalable alternative to costly real-world interaction for improving robot policies. However, existing world-model-based methods largely treat world models as forward-only surrogates of the physical environment, leaving untapped a capability no physical environment affords: rewinding to past decision points to explore counterfactual futures. We introduce Rollback, a policy improvement paradigm that turns world models from passive simulators into counterfactual engines: it rolls back to critical decision points, branches into alternative actions, and contrasts their outcomes to attribute success to individual decisions. This concentrates imagination where it matters most and distills counterfactual outcomes into fine-grained, action-level supervision without a separately learned value model. We build a scalable training system that unifies distributed world-model search with asynchronous real-robot execution, forming a self-reinforcing cycle: real-world experience continually sharpens the world model, and a sharper world model in turn drives further policy improvement. We term the world-model side of this cycle test-time physics grounding, a capability that hand-built physics simulators lack: while the world model serves as the training environment, real transitions from the policy it trains correct its dynamics through a lightweight LoRA pipeline. Experiments on three challenging real-world tasks show that Rollback improves success rates over the strongest baseline by an absolute 10.67%, 11.11%, and 11.85% on block extraction, ball tossing, and tabletop curling, respectively, while achieving more stable improvement throughout training. Although initialized from teleoperated demonstrations far below human level, Rollback surpasses the average success rate of human players on both ball tossing and tabletop curling within 8 hours of training.

cs.RO↗

arXiv2026

WorldOracle: Benchmarking Action-Conditioned World Models against Real-World Executions

We introduce WorldOracle, a benchmark specifically designed to evaluate action-conditioned world models (ACWMs) against paired real-world executions. Given the same initial observation and low-level action sequence, WorldOracle asks whether a predicted rollout reproduces what happens in the real world, including physical consequences and task-state transitions. We evaluate models along four complementary dimensions: physical state correctness, interaction spatial alignment, key interaction area fidelity, and multi-view action coherence. To support this evaluation, we curate real-world training and evaluation suites that cover different interaction dynamics and embodiments. The evaluation suite contains long, multi-segment episodes with both successful and failed executions, enabling evaluation across varied action outcomes and behavior distributions. Experiments on seven representative ACWMs reveal distinct model strengths and failure modes. WorldOracle provides a systematic foundation for measuring and advancing action-conditioned world modeling for manipulation. We release the datasets, evaluation code, and leaderboard: https://worldoracle-benchmark.github.io/.

cs.RO↗

arXiv2026

Exact Magnetic Confinement of a Doped Hole by Kinetic Frustration

We establish an exact mechanism of magnetic confinement driven purely by kinetic frustration. In a finite open sawtooth $U=\infty$ Hubbard chain of $L$ unit cells doped with one hole, the ground state at magnetization $M$ contains $M$ polarized triplet dimers surrounding a connected singlet region of length $\ell_{\rm s}=L-M$. A magnetic field selects $M$ and hence the confinement length. The hole forms a standing wave in the singlet region with exponential tails in the triplet background. Confinement requires neither disorder nor an exchange-generated string tension: destructive interference between direct and apex-mediated hopping makes the triplet bonds reflecting. At $t_2/t_1=[2+4\cos(π/(\ell_{\rm s}+2))]^{-1/2}$, they become perfectly reflecting, producing a position-degenerate compact localized state. At every fixed field $0<h<h_{\rm sat}$, the singlet region remains finite as $L\to\infty$ and shrinks stepwise with increasing field, whereas at $h=0$ it spans the system.

cond-mat.str-el↗

arXiv2026

Distributions or Quantiles? A Convex Variational Theory of Worst-Case Risk

Worst-case risk evaluation is central to finance and operations research, yet existing analytical results remain fragmented and largely confined to ambiguity sets with specific structures. Using spectral risk measures as the underlying class, we develop a convex variational theory that unifies these analyses and extends them to richer ambiguity models. The theory rests on a representation principle: use the quantile or distribution formulation that preserves convexity of the full ambiguity set. In both representations, the resulting optimality conditions provide a systematic route to identify the worst-case law and determine its risk value. This approach recovers classical central-moment bounds and yields closed-form or semi-closed-form solutions for practically motivated models involving Wasserstein ambiguity with tail stress, two-sided expected shortfall (ES) bands, asymmetric partial moments, geometric-moment bounds, and exponential- or Gaussian-reference Kullback--Leibler (KL) neighborhoods. Beyond these analytical results, the same theory yields projection-based algorithms with fixed-grid convergence under the stated assumptions. In a high-dimensional portfolio application, we establish an exact projection identity for mean-anchored joint KL ambiguity under a Gaussian reference. The resulting standardized worst-case ES coefficient applies to every portfolio, reducing robust portfolio selection to an exact second-order cone program after a single calibration.

q-fin.RM↗

arXiv2026

Prediction of the Geoneutrino Signal at the Jinping Neutrino Experiment

We present refined predictions of the geoneutrino flux for the 500-tonne Jinping Neutrino Experiment (JNE). We construct a local crustal model with a resolution of $0.2^\circ\times0.2^\circ$ based on the SWChinaCVM-2.0 seismic velocity model and integrate it with CRUST1.0 to establish our reference model. For comparison, we also calculate the geoneutrino flux using ECM1 and CRUST1.0 as standalone global crustal models. Assuming an energy scale of 500~p.e./MeV and an exposure of 500~t$\times$1500~days, we predict the inverse beta decay event spectrum, including both the geoneutrino signal and the reactor antineutrino background. The distinct spectral shapes of the $^{238}$U, $^{232}$Th, and reactor antineutrino components allow their respective contributions to be obtained through spectral fitting. These predictions provide inputs for the JNE data analysis, particularly for extracting the mantle geoneutrino signal and testing competing bulk silicate Earth models.

hep-ex↗

arXiv2026

CARTE: A Benchmark for Mapping Language Model Knowledge Across France

We introduce CARTE (Culturally Anchored Regional-Territorial Evaluation), a multiple-choice benchmark for evaluating the ability of large language models (LLMs) to discriminate fine grained, geographically grounded and regionally differentiated knowledge within France. While prior benchmarks focus on national-level cultural understanding, they largely overlook intra-country variation and the need to distinguish between closely related regional contexts. CARTE addresses this gap by introducing 3,132 questions spanning the 13 metropolitan regions of France and covering 12 thematic domains, including culture, language, demographics, economy, environment, and mobility. Questions underwent four-stage validation: human review, LLM-as-a-judge, model-based difficulty calibration, and a second human review. We further introduce CARTE-LV, a subset targeting Linguistic Variation across French regions, enabling focused evaluation of language-related differences. We evaluate 32 open-weight LLMs of 1B to 30B parameters and three frontier API models across zero-, one-, and three-shot settings. Our experiments reveal performance disparities across regions and model scales, suggesting systematic gaps in pretraining coverage and limited robustness to intra-national variation.

cs.CL↗

arXiv2026

GenXR-Bench: Benchmarking Coding Agents for Interactive XR Development

Coding agents are rapidly advancing from single-file generation to long-horizon repository-scale engineering, yet the evaluation of their capabilities to author interactive extended reality (XR) applications remains an open challenge. Unlike traditional software, XR applications operate within dynamic 3D physical environments where runtime robustness depends jointly on continuous user motion, spatial input, and environmental sensing. We introduce GenXR-Bench, the first benchmark for the end-to-end evaluation of coding agents in XR software development. GenXR-Bench contains 112 expert-authored tasks spanning 24 feature implementations, 21 bug repairs, and 67 application generation tasks. Held-out tests combine scripted and adaptive interactions with programmatic and visual assertions to evaluate core functionality and runtime behavior. GenXR-Bench's testing suite achieves 96.5% accuracy against blinded expert judgments of 124 candidate applications executed on consumer XR hardware. Evaluating 10 models on GenXR-Bench, we find that the strongest configuration achieves only 50.9% overall Pass@1. Failure analysis identifies recurring misunderstandings in asynchronous sensor lifecycles, physical grounding, and embodied interaction semantics. GenXR-Bench establishes a reproducible testbed for evaluating coding agents in the interactive and physically situated conditions of spatial computing.

cs.AI↗

arXiv2026

RiboBridge: protein-conditioned de novo RNA binder structure design

Designing RNA molecules that bind to target protein structures requires generating both an RNA conformation and its bound complex geometry, without assuming a predefined binding site. Existing approaches generally model these coupled processes within a unified representation, entangling RNA conformation generation with protein-relative placement. In this work, we introduce RiboBridge, a structure-first framework that separates RNA conformational generation from protein--RNA assembly through cross-frame interaction geometry. RiboBridge first proposes multiple candidate RNA-binding interfaces on the protein surface, generates canonical RNA conformations within a pretrained RNA structural latent space (RiboSphere), and then predicts protein--RNA interaction constraints to bridge the independently defined RNA and protein coordinate frames. The resulting interaction geometry is converted into a geometrically consistent protein-relative pose through rigid placement. We evaluate RiboBridge on a temporally split, sequence-filtered, and redundancy-controlled benchmark of 124 protein--RNA complexes. Under post-hoc interface-correspondence evaluation, where native information is used only after generation to select among already generated predicted-pocket candidates, RiboBridge achieves the highest fNAT and DockQ and the lowest interface and ligand RMSDs among compared methods. Component ablations and stage-wise diagnostics further characterize the contributions of RNA conformational generation, interaction prediction, and rigid placement, identifying the main sources of complex-recovery error.

cs.LG↗

arXiv2026

WildFakeToolBench: Benchmarking AI-Generated Image Detectors with Web-Sourced Image Processing

As AI generated images are getting popular these days, discussions on such generations are widely existed in current public forums, software repositories, and photography tutorials. Moreover, with the help of LLM-assisted code translation, malicious actors can adopt such information to evade detectors without specialized knowledge or local GPU execution of the resulting tools. To evaluate such risks, we introduce \textsc{WildFakeToolBench}, a benchmark suite of $156$ executable image-processing tools derived from $220$ public sources across $142$ domains. Each tool retains its source evidence and passes syntax, execution, and output-distortion checks without using feedback from the evaluated detectors. On $105$ generated images from five generators, we evaluate six state-of-the-art detectors with images processed by the tools in our collected bench to assess their effectiveness against such in-the-wild risks. The strongest individual tools achieve attack success rates of $25.25\%$--$100\%$ across detectors, exceeding the strongest of six selected image-processing baselines for each detector. Evaluating subsets of an eight-tool pool yields worst-case coverage above $80\%$ for every detector. Effectiveness is sparse and detector-specific: high coverage across candidates does not imply reliable one-shot evasion. These findings not only highlight the potential of such web-sourced risks, but also show the potential on establishing AI-image detector robustness evaluation framework based on those in-the-wild information.

cs.CV↗

arXiv2026

SWE-Clar: Can Coding Agents Clarify Incomplete Client Requirements and Deliver Greenfield Software?

Greenfield software development in freelance settings requires coding agents to turn incomplete client requests into working software. This makes requirement clarification essential, yet clients may be unable to provide continuous feedback over long development cycles, and receiving their answers does not guarantee correct understanding of complex requirements. These challenges motivate two benchmark design choices: clarification precedes code implementation, and its effectiveness is assessed through explicit tests of requirement understanding. Guided by these choices, we introduce SWE-Clar, a benchmark grounded in real, high-value client commissions to evaluate requirement clarification and code implementation in this setting. It comprises 100 tasks based on real software development jobs on Upwork, each with at least US\$4,000 in total payments. Each task begins with multi-round interaction between a coding agent and a client agent, followed by separate runs for code implementation with acceptance testing and clarification assessment. For the latter, we construct 512 questions paired with reference answers from real client-freelancer interactions, covering both explicitly stated requirements and those inferred from the conversational context. Both runs begin from the same clarification context but remain isolated, preventing assessment questions from guiding code implementation. Experiments with nine frontier models show that even GPT-6 Astra, which has the highest observed implementation score, achieves a task pass rate of only 67%, underscoring the challenge of independent software delivery in this setting. The dataset and evaluation code are available at https://github.com/upwork/SWE-Clar.

cs.SE↗

arXiv2026

Altermagnetism in Folded Iron Chain Systems

Altermagnetism, a moment-compensated magnetism with momentum-dependent spin splitting, has so far been realized primarily in Néel-type antiferromagnets. Here, we demonstrate altermagnetism in folded iron-chain systems with block-type magnetic order (e.g., $\uparrow \uparrow \downarrow \downarrow$). In the strictly one-dimensional (1D) case, altermagnetism is forbidden by the $[C_2||{\cal T}]$ symmetry, but it can emerge when such chains are stacked into a quasi-1D array. To generate altermagnetic spin splitting via this chain-stacking mechanism, two essential ingredients are required: a block-type spin order that spontaneously breaks the combined parity-time reversal symmetry, and finite inter-chain hopping. Using first-principles calculations, we identify the K$_3$Fe$_2$Se$_4$ family as a promising platform that realizes this mechanism, hosting a $d$-wave altermagnet even without lattice $C_{4}$ symmetry. Moreover, owing to its lower symmetry, the block-type folded chain exhibits an isotropic altermagnetic spin splitting, which may manifest as an altermagnetic wrapping effect. Our findings unify real-space block magnetism and reciprocal-space altermagnetism, establishing folded iron chains as a tunable platform for zero-moment spintronic applications.

cond-mat.mtrl-sci↗

arXiv2026

Can Native Safety Travel? Reusing Safety Pretraining Across Models via On-Policy Distillation

Safety pretraining has emerged as a promising paradigm for embedding safety deeply into a model during pretraining, yielding robustness to subsequent fine-tuning that conventional safety post-training often lacks. We refer to this safety acquired during model pretraining as native safety. Yet native safety is expensive to acquire, confined to the model in which it was pretrained, and often accompanied by lower general utility. In this work, we rethink the practical value of safety-pretrained models and introduce, for the first time, the native safety reuse problem: Can native safety be transferred from a specialized, potentially smaller, safety-pretrained "teacher" to standard-pretrained "students," enabling them to inherit its robustness while preserving their general utility? We show that native safety can "travel" through on-policy distillation. A standard-pretrained student generates responses to safety-pretraining prompts, while a small safety-pretrained teacher supervises the student's token-level predictions along its own generated trajectories; cross-tokenizer alignment further enables transfer across model families. The resulting students exhibit substantially greater safety robustness than their safety-post-trained counterparts while retaining higher general utility than the teacher: subsequent fine-tuning raises the average attack success rate (ASR) of the 1.7B student by only 5.00%, compared with 11.58% for safety post-training, and its average utility of 48.20% far exceeds the aligned teacher's 27.42%. Unlike conventional large-to-small knowledge distillation, our framework enables selective, small-to-large safety transfer: it reuses the robust safety expertise of a smaller pretrained teacher without inheriting its utility limitations. These findings establish native safety as a reusable model property rather than one permanently tied to the originally pretrained model.

cs.LG↗

arXiv2026

SWE-Residual: Evaluating Starting-State Robustness of Coding Agents

Software engineering (SWE) benchmarks are the primary way to measure the capabilities of coding agents: an agent is placed in a code repository or machine environment, given a request, and judged by hidden tests. These benchmarks almost always start the agent from a clean state that contains no earlier work on the task. In practice, however, agents often continue work that a developer or another agent has already done and reported as complete, with no indication of what remains incorrect or missing. We introduce SWE-Residual, a benchmark for Starting-State Robustness: the ability to maintain performance on a task when the starting state already contains earlier work on that task. SWE-Residual contains 198 Residual States drawn from 111 tasks in DeepSWE and Terminal-Bench, where the earlier work modifies a code repository or a whole machine environment. Each Residual State is the state a predecessor agent left behind after submitting its work as complete, although the hidden tests still fail. The evaluated agent receives only the original request and this state, without the predecessor's trajectory or any sign that the attempt failed; because the request and hidden tests are unchanged, the starting state is the only difference from the source benchmark. Across five frontier agents, success drops by 20.6--53.5 percentage points on DeepSWE and 20.6--34.9 points on Terminal-Bench relative to a Clean Start, and agent rankings on DeepSWE reverse. Failed continuations show a pattern we call acceptance bias: agents treat the inherited work as the definition of what remains to be done instead of checking it against the request. On DeepSWE, in 58.9--88.2% of failed runs, they submit with exactly the same tests failing, although they pass most of these tests from a Clean Start. SWE-Residual thus exposes a capability gap that clean-start evaluation does not capture.

cs.SE↗

arXiv2026

PartiGen: Part-Aware Representations for Language-Conditioned Humanoid Motion Generation

Translating language into humanoid motion requires coordinating body parts, determining motion duration, and producing references suitable for robot execution. We introduce PartiGen, a part-aware latent diffusion framework that incorporates anatomical structure into a shared whole-body motion representation. Part-specific attention processes anatomically masked frame features, and its outputs are fused to combine local limb information with whole-body context. Conditioned on text and motion history, the model generates motion primitives autoregressively. Forward-kinematics supervision encourages kinematic consistency, while a learned end-of-sequence predictor selects a frame-level endpoint after generation. On HumanML3D retargeted to the Unitree G1, PartiGen achieves a segment-level FID of 3.165, a 16.8% reduction over TextOp, and improves R@1 by 1.4 percentage points. In the duration study with 16-frame primitives, endpoint prediction reduces mean absolute duration error from 12.471 to 3.080 frames compared with a fixed-length cutoff. We further evaluate robot execution using a pretrained, frozen GRIT tracker: 409 of 500 generated references (81.8%) pass the posture-based stability screen in MuJoCo. Following simulation screening and manual selection, 26 of 27 attempted motions are judged successful on the real G1.

cs.RO↗

arXiv2026

BeJEPA: A Goal-Conditioned Humanoid Behavior World Model via Unsupervised RL

Humanoid control requires coordinating whole-body posture with position and heading in the world frame. Mainstream controllers imitate trajectories in the root frame, but struggle to regain the target world pose after displacement. We introduce BeJEPA, a goal-conditioned Behavior World Model learned through unsupervised reinforcement learning from human motion priors and robot interaction. BeJEPA learns a shared representation of body configuration and world placement through a joint-embedding predictive architecture (JEPA). A Dynamic Predictor predicts the next state embedding under an action, while a Successor Predictor summarizes a goal-conditioned policy's expected discounted future state embeddings in a single vector. During joint training, Dynamic grounds the initial representation in observed physical transitions, while Successor increasingly organizes it around long-term behavior. Their composition guides policy improvement through one physical prediction and one behavioral summary, without long imagined rollouts. In 20-second motion-tracking experiments, BeJEPA with online action refinement using Dynamic outperforms all evaluated baselines in global body-position error, reducing it on AMASS by 61.7% relative to BFM-Zero with world-pose inputs and 58.7% relative to SONIC. Physical deployment on a Unitree G1 demonstrates joint attainment of posture, world position, and heading goals, recovery from the ground, and closed-loop return after disturbances and goal changes. Project page: https://ylyem9x.github.io/BeJEPA-project.

cs.RO↗

arXiv2026

PTagent: An LLM-assisted tool for finite-temperature model construction in cosmological phase transitions

Cosmological phase-transition calculations require well-defined finite-temperature numerical models, whose construction involves assembling model information under consistent theoretical prescriptions. In this work we present \texttt{PTagent}, a large language model (LLM) agent-based framework for constructing such models from SARAH model files, research papers, and DRalgo sources. The LLM agent interprets and organizes model information and identifies unresolved physical choices, while dedicated backends perform the model-dependent calculations. After review and approval of the resulting physical model specification, the backends generate the corresponding CosmoTransitions-style or PhaseTracer-style numerical potentials. By separating model interpretation from numerical implementation, PTagent provides an inspectable and reproducible approach to finite-temperature model construction for cosmological phase-transition studies.

hep-ph↗