arXiv subjects

Fan Yang

Publications and source records attributed to Fan Yang.

At least 109 records · Page 6Linked to original sources

arXiv2026

Completion-Path Credits: Multi-Resource Control for Scale-Up Fabrics

Scale-up fabrics connecting GPUs and AI accelerators carry tensor transfers together with remote reads, writes, atomics, and notifications over shared target-side receiver resources. Byte-denominated credits protect link buffers and streaming HBM traffic, but poorly represent small operations dominated by Atomic execution or response injection. This paper presents SemaCredit, a receiver controller that admits each remote-memory operation against a vector of target-resource demands and returns each component when its corresponding HBM, Atomic, or response stage completes. In a deterministic event simulator with multipath queues, eight HBM partitions, a serialized Atomic engine, and a response engine, SemaCredit matches a strong per-resource byte baseline on HBM-hotspot traffic while reducing small-operation P99 latency by 52.4% under Atomic contention and 10.2% under response incast. Application-shaped mixes show 57.7% and 14.5% P99 latency improvements for AllReduce-shaped and remote-read-shaped traffic while matching byte credits on HBM-dominated MoE traffic.

cs.NI↗

arXiv2026

Two-flag degenerations and real circles tangent to three conics

Three general plane conics admit $184$ complex tangent circles, and it had been conjectured that at most $136$ of them could be real. We construct an explicit strongly general triple of smooth conics over $\mathbb{Q}$ with exactly $160$ real tangent circles; the same count therefore occurs on a nonempty Euclidean chamber. The construction combines a fourfold splitting theorem for two flagged double-line degenerations with exact Sturm--Tarski, elimination, and interval certificates. We also show that the Grothendieck--Witt-valued count is $92\mathbb{H}$ and express its real local signs, up to a fixed orientation convention, in terms of curvature differences, residual intersection divisors, and contact normals. The exact-arithmetic code and certificate data are archived in the accompanying repository.

math.AG↗

arXiv2026

From Reasoning Depth to Reasoning Breadth: Evaluating Multi-Point Associative Reasoning in Large Language Models

Large language models (LLMs) have made substantial progress on reasoning tasks that require increasingly long and complex inferential chains. This progress primarily reflects reasoning depth. A complementary and comparatively unexamined capability is reasoning breadth: exploring multiple semantic directions in parallel and integrating the resulting clues into one coherent answer. We introduce MPAR-Bench, a bilingual English-Chinese benchmark that isolates reasoning breadth through multi-point associative reasoning. Inspired by the cooperative game Just One, each item asks a model to recover a hidden target from several independently generated, semantically diverse clues. We construct 1,000 items using a multi-agent clue-generation pipeline, embedding-based diversity filtering, and human verification. Only the answer space is drawn from public word lists, whereas every clue set is generated from scratch. Beyond exact-match accuracy, we evaluate models using accuracy, ANLS, embedding similarity, reasoning-trace verification, and four perturbations: clue masking, order shuffling, distractor injection, and multi-step clues. Across evaluated models, perturbations reduce accuracy by 9-18 percentage points in English and 5-12 percentage points in Chinese. Thinking mode improves standard-setting accuracy, especially in English, but does not consistently reduce sensitivity to perturbations. Case-level analysis also shows that extended reasoning can overturn an initially correct hypothesis. These results indicate that greater reasoning depth does not automatically confer robust reasoning breadth, and that reasoning breadth remains largely uncovered by current benchmarks.

cs.CL↗

arXiv2026

Agent Skills Can Be Harmful: An Empirical Study of Skill-Induced Failures in LLM Agents

Agent skills are the de facto mechanism for extending LLM agents with reusable guidance. A skill can shape the agent's task execution, including planning, tool use, problem-solving, and validation. Prior work reported mixed results of agent skills: some skills improve task success rates, while others have no effect, increase token use and execution time, and even reduce success rates. This paper presents a comprehensive analysis of skill-induced agent failures by attributing task failures and cost regressions to specific loaded skills. We introduce a differential analysis framework that attributes a failure or regression to a skill by comparing a target skill-guided run against a no-skill or semantically matched skill reference run that solves the same task, or solves it more cheaply. We instantiate this framework on SkillsBench and SWE-Skills-Bench, yielding 307 skill-induced failures, including 125 functional failures and 182 efficiency regressions. We also build SkillTriage, a taxonomy-guided attribution tool that normalizes paired cases, extracts differential evidence, and produces triage reports. Our major findings include: (1) Skill induced functional failures are rarely caused by obviously irrelevant skills; instead, seemingly relevant skills often make the agent incorrectly implement or omit task-required implementation elements. (2) Skill-induced efficiency regressions are not explained by prompt length alone. (3) The largest sources within Excessive Procedure are excessive verification and heavy implementation pipelines, contributing 67 and 30 cases, respectively. This shows that skills often turn validation checklists and construction recipes into mandatory work. Based on our findings, we propose research topics and tooling improvements for safer and more cost-aware skill reuse.

cs.AI↗

arXiv2026

Physics-Constrained Co-Optimization and Data-Driven Layer-Resolved Classification of a Hybrid CZT/PIPS Detector for Mixed Radiation Fields

Compact mixed-radiation instruments must preserve a low-mass charged-particle entrance while providing enough high-Z depth for photon sensitivity. We first compare two detector heads within a 40 x 20 x 10 mm^3 design budget. S1 places bare CdZnTe (CZT) and passivated implanted planar silicon (PIPS) branches side by side and estimates three rates. S2 adds 0.50 mm of CZT behind PIPS to estimate X/gamma, low-beta, high-beta, and alpha rates. A hard-constraint search combines photon attenuation and deposition, Hecht charge collection, charged-particle energy loss, solid angle, resolution budgeting, timing, and response-matrix conditioning. Feasible screening points exist inside the initial envelope, but a fresh transport/electronics assessment gives only 25.04-25.07 cps/(uSv/h) under the robust H*(10) convention, and every conservative electronics draw exceeds 2.5% FWHM. We therefore retain the more informative S2 observation structure, relax only the 10 mm package-depth constraint, divide the bare CZT into independently biased layers, and add a low-noise sum channel for spectroscopy plus layer-resolved gradient-boosted classification for mixed-field analysis. The extension campaign contains 2.40 million Geant4 11.4.1 histories over three depths, five transport seeds, and 14 particle/energy cases. The final 40 x 20 x 12.570 mm^3 head uses 7.870 mm of bare CZT in five layers at 180 V per layer. Under the application-scenario U95 electronics profile, its live-time-corrected 662 keV sensitivity is 32.804 cps/(uSv/h), its independent-sum resolution is 2.303% FWHM at the 95th percentile, and the propagated beta/alpha absolute-efficiency lower bounds are 35.679%/39.832%. Additional endpoint transport covers 20 keV-3 MeV photons, 3-7 MeV alpha particles, and 155 keV-3.5 MeV beta spectra. The extended design therefore passes the original detector-performance criteria in the U95 model.

cs.CE↗

arXiv2026

When Agent Automation Becomes Profitable: Quantifying and Insuring Autonomous AI Risk through Trace-Economic Underwriting

AI agents can now take irreversible actions in operational systems, but agent-caused losses are still not clearly assigned, priced, or transferred. Providers often disclaim consequential damages, users are left with uncompensated losses, and default human review limits the efficiency gains of automation. We ask when autonomous AI deployment can become economically acceptable despite failure risk. Our answer is to quantify risk at the customer-task-trace episode level and transfer it through insurance. Automation is acceptable when its expected benefit exceeds the premium, control cost, and remaining risk. This requires a defined role with bounded permissions and comparable traces. We introduce trace-economic underwriting, which maps tool-use traces to customer exposure and claimable loss, then uses this representation for pricing, control, and risk transfer. It uses deterministic economic labels rather than an LLM judge. In our trace-to-loss testbed, trace-economic pricing reduces pricing MAE from $17.7K to $569 and removes regressive cross-subsidy. A 300-trace expert audit accepts 295 labels unchanged. On 1,000 real SWE-smith traces, trace-conditioned controls reduce CVaR95 by 72%. Theorem~1 gives a finite-sample scope condition. We release code, labels, and audit sheets.

cs.AI↗

arXiv2026

Extreme Value Alpha and Crash Risk: Separating Structural Tails from Lottery Tails with LLM-Extracted Disclosure Networks

A heavy upper tail in a stock's returns is ambiguous: it can be a lottery tail, transient jump risk that investors overpay for (the MAX discount), or a structural tail, the statistical shadow of an economic reconfiguration that precedes extreme winners. Returns alone cannot separate them, so tail heat alone is not an alpha signal. Our discriminator is the firm's disclosure-measured network: a directed, span-grounded graph from 10-K filings via an auditable LLM pipeline, whose rewiring decomposes into edge birth, death, and drift. The central sign pattern: tail heat with network death is the crash side; tail heat with an intact or forming network is where structural tails and historical winners live. Pilot evidence from 24 technology firms (2014-2025) supports the crash side: upper-tail heat interacted with death mass predicts negative forward abnormal returns (monthly t = -2.9; firm-vintage t = -3.9; wild-cluster p = 0.04; robust to two-way clustering and controls), and the same configuration preceded NVIDIA's 2018 and 2022 drawdowns. This death-side signal is immediately useful as a risk-monitoring danger flag. The alpha side is directionally positive but not significant in the pilot and awaits the confirmatory test. A pre-registered replication on 50 random S&P 500 firms failed, defining the boundary: outside coherent ecosystems the disclosure graph nearly vanishes (83% of firm-vintages have zero death mass), so the discriminator exists only where firms densely document counterparties. The confirmatory design is fully pre-specified, with a gatekept primary pair, archived power simulations, positive-only winner labels, selection-corrected benchmarks, and a frozen ecosystem-coherent universe with a density gate; it activates only if the gate passes. If confirmed, tail heat becomes a conditional signal separating crash risk from structural winners.

stat.AP↗

arXiv2026

Rethinking Self-Evolving Agents: Do We Still Need Prescribed Optimization Pipelines?

Self-evolving agents are usually built around prescribed optimization pipelines: the framework decides how to gather evidence, revise a persistent artifact, select candidates, and stop. We ask whether this task-specific procedure remains necessary when a frontier model acts as the optimizer. We introduce Open-Ended Optimization (OEO), which keeps the objective, permitted interactions, resource budget, data boundary, and evaluation fixed while allowing the optimizer to compose the improvement process online. We compare OEO with two complementary prescribed approaches: SkillOpt, a staged pipeline with bounded edits, and GEPA, a reflective evolutionary search. Across 14 head-to-head comparisons over 8 benchmark-target-model settings, GPT-5.5-driven OEO records 12 wins, 1 tie, and 1 narrow loss of 0.21 percentage points. It uses a median 34.3 percent of SkillOpt's configured target-interaction token budget. A one-shot, zero-interaction control shows that the gains are not explained by a single prior-driven rewrite. However, delegation has a capability boundary: SkillOpt outperforms OEO with a medium optimizer, and a weak optimizer cannot operate through the unchanged OEO interface. In the fully instrumented OEO-SkillOpt pair, trajectory analysis further shows that prescription changes how optimization proceeds more consistently than it changes final behavior. Together, these findings recast prescribed pipelines as capability-dependent scaffolding: essential constraints remain external, but a sufficiently capable optimizer can compose the route from measurable feedback to persistent improvement.

cs.AI↗

arXiv2026

AirFlow: Context Preserving and Multi-Rate State Modeling for Air Quality Forecasting

Accurate air quality forecasting is essential for public health and urban environmental management, but remains challenging because pollutant channels differ in periodicity and distribution drift, while their concentration trajectories contain both multi-scale dependencies and rapid changes. Recent methods have improved spatial dependency learning and meteorological covariate modeling. However, pollutant channels are still passed through the same normalization rule and temporal backbone, using a shared latent representation for channel-specific distributions and changes at different rates. To address this limitation, we propose AirFlow, a pollutant-aware dual-stream framework that operates on station multivariate observations without additional graph propagation or predefined signal decomposition. Specifically, AirFlow designs two novel blocks: (1) a statistic-guided normalization routing mechanism that selects a normalization path for each pollutant according to its 24-hour autocorrelation and distribution drift; and (2) a hierarchical dual-stream state model that combines multi-scale state space propagation with learnable response coefficients, where gated bidirectional cross-attention exchanges information and adaptively fuses the resulting representations. Experiments on real-world data from multiple cities show that AirFlow achieves the best performance in 34 of 36 metrics comparisons, with reductions of up to 11.11% root mean square error over the state-of-the-art baseline. AirFlow also requires only 0.0483M parameters and 0.0215G FLOPs, achieving high forecasting accuracy with low computational overhead.

cs.AI↗

arXiv2026

Perpendicular electric field induced competing $s^\pm$- and $d$-wave pairings in La$_3$Ni$_2$O$_7$ thin film

Inspired by the possibility that superconducting properties may be altered by applying a perpendicular electric field in the Ruddlesden-Popper (RP) bilayer nickelate La$_3$Ni$_2$O$_7$ thin film, we investigated the imbalanced two-orbital bilayer Hubbard model with a layer potential bias using dynamical cluster quantum Monte Carlo calculations. Focusing on the pairing symmetry evolution with the potential bias between layers for the undoped, hole-doped, and electron-doped regimes, we found that the $s^\pm$-wave pairing, originating from the $d_{z^2}$ orbital at high temperature regime as well as $d_{x^2-y^2}$ orbital at lower temperatures, is suppressed by the potential bias; while a possible pairing symmetry transition from $s^\pm$-wave to $d$-wave pairing occurs, driven by the interlayer orbital mismatch and the transfer of electrons into the $d_{x^2-y^2}$ orbitals. Our large-scale many-body calculations align with the previous expectation from weak-coupling methods and provide further insight into the superconducting mechanism in RP nickelates.

cond-mat.supr-con↗

arXiv2026

SkillReason: Reasoning-Enhanced Agent Skill Retrieval for Implicit User Requests

Large language model agents increasingly rely on reusable skills to extend their capabilities beyond parametric knowl- edge. However, retrieving the appropriate skill from a large- scale library remains challenging because realistic user re- quests are often concise and underspecified, stating only the task goal while leaving the required capabilities and execu- tion steps implicit. Existing benchmarks provide limited cov- erage of such requests. To address this gap, we introduce SkillReason-Bench, a large-scale cross-domain benchmark containing 3,729 queries and a retrieval corpus of 61,228 skills spanning nine domains. We further propose SkillRea- son, a two-stage framework that uses chain-of-thought rea- soning as training-time supervision for skill retrieval. In Stage I, capability reasoning traces generated by a stronger teacher provide explicit supervision through contrastive learning, re- trieval distribution alignment, and language modeling, en- couraging the retriever to internalize capability reasoning in its query representation. In Stage II, a retrieval-guided GRPO objective encourages the model to explore reasoning trajecto- ries better suited to its own capabilities and more effective for retrieval. At inference, SkillReason directly encodes the orig- inal query without autoregressive CoT generation, preserv- ing efficient query-only retrieval. Extensive experiments on SkillReason-Bench, SkillRet, and SRA-Bench show that Skill- Reason achieves state-of-the-art performance across all three benchmarks, demonstrating that reasoning-enhanced training better bridges the semantic gap between high-level task goals and skill capabilities.

cs.AI↗

arXiv2026

Sharp $\ell^p$-Improving Estimates for Fixed-Radius Discrete Spherical Averages

Let $d\geq 4$ and let $R>0$. When $d=4$, assume that $R^2\in\mathbb{N}\setminus 4\mathbb{N}$; when $d\geq 5$, let $R^2\in\mathbb{N}$ be arbitrary. We prove the fixed-radius estimate $$\|A_R f\|_{\ell^{p'}(\mathbb{Z}^d)}\leq C_{d,p,\varepsilon}R^{-d(2/p-1)+\varepsilon}\|f\|_{\ell^p(\mathbb{Z}^d)}$$ for $(d+2)/d\leq p\leq 2$, where $p'$ is the Hölder conjugate exponent of $p$ and $A_R$ is the probability average over the lattice sphere of radius $R$. This extends the fixed-radius estimates of Kesler--Lacey and Hughes to the sharp lower endpoint $p=(d+2)/d$.

math.CA↗

arXiv2026

PushDualGen: Enabling LLMs to Generate Semantic IDs with Interpretable Copy for Industrial Push Recommendation

Push recommendation in KuaiShou proactively delivers personalized content to nearly one billion users to facilitate their engagement. Recently, generative recommendation has achieved end-to-end user personalization through semantic ID. However, their black- box characteristics make recommendation logics difficult to trace, hindering their deployment. OneRec-Thinking addresses this by incorporating CoT before generating SIDs, but this significantly increases inference cost. To support large-scale industrial applications, we propose PushDualGen, a lightweight generator, which first generates the SID and then produces a copy as a skippable explanation. PushDualGen has been deployed in Kuaishou's push recommendation system. Online A/B tests demonstrate the effectiveness of PushDualGen, delivering significant improvements in both user attraction and satisfaction. The effective play rate for videos recommended to users has relatively increased by 8.50%, while the dissatisfaction rate has relatively fallen by 37.70%. In the long term, PushDualGen optimises the content ecosystem, providing more exposure for long-tail videos.

cs.IR↗

arXiv2026

Symmetries of Spin-Splitting Induced by Spin-Orbit Coupling in Non-magnetic Crystals

Spin-orbit coupling (SOC) leads to splitting of otherwise spin-degenerate bands in noncentrosymmetric materials, even if time-reversal symmetry is present. While this gives rise to well-known phenomena such as the Rashba and Dresselhaus effects, various other terms are allowed based on the point group of the crystal and the electronic Hamiltonian. In this study, we utilize point group representations to illustrate that four different types of SOC terms (Rashba, Dresselhaus, Weyl, and Ising) can emerge in periodic solids. We construct reciprocal-space energy expressions for each type of SOC-induced splitting of opposite-spin bands, and follow a similar procedure to also obtain minimal tight-binding models that capture all types of spin-splittings for subgroups of two high-symmetry groups $m\bar{3}m$ and $6/mmm$. Furthermore, we also obtain a complete list of nodal features in the electronic band structure in these systems, distinguishing between crystallographic-symmetry-imposed nodal lines and those imposed by time-reversal-symmetry only. Finally, we conclude by presenting a list of materials that host each type of inversion-breaking SOC effects. Our classification of the spin-splitting symmetries in non-magnetic systems with SOC is the counterpart of the recent classification of spin-splitting symmetries in unconventional magnetic systems without SOC, such as altermagnets and odd-parity magnets. More broadly, our work provides a basis for studying superconductivity and other collective electronic phenomena that are impacted by SOC-induced band splittings in noncentrosymmetric materials.

cond-mat.mtrl-sci↗

arXiv2026

Skill-RAG: Failure-State-Aware Retrieval Augmentation via Hidden-State Probing and Skill Routing

Retrieval-Augmented Generation (RAG) has emerged as a foundational paradigm for grounding large language models in external knowledge. While adaptive retrieval mechanisms have improved retrieval efficiency, existing approaches treat post-retrieval failure as a signal to retry rather than to diagnose -- leaving the structural causes of query-evidence misalignment unaddressed. We observe that a significant portion of persistent retrieval failures stem not from the absence of relevant evidence but from an alignment gap between the query and the evidence space. We propose Skill-RAG, a failure-aware RAG framework that couples a lightweight hidden-state prober with a prompt-based skill router. The prober gates retrieval at two pipeline stages; upon detecting a failure state, the skill router diagnoses the underlying cause and selects among four retrieval skills -- query rewriting, question decomposition, evidence focusing, and an exit skill for truly irreducible cases -- to correct misalignment before the next generation attempt. Experiments across multiple open-domain QA and complex reasoning benchmarks show that Skill-RAG substantially improves accuracy on hard cases persisting after multi-turn retrieval, with particularly strong gains on out-of-distribution datasets. Representation-space analyses further reveal that the proposed skills occupy structured, separable regions of the failure state space, supporting the view that query-evidence misalignment is a typed rather than monolithic phenomenon.

cs.CL↗

arXiv2026

Argus: A General-Purpose Agentic Reasoning Runtime for Long-Horizon Tasks

Long-horizon reasoning requires an agentic runtime that can persist when evidence supports its current approach and pivot when measurements reveal failure, hidden constraints, or a misspecified objective. We present Argus, a persistent, self-evolving runtime in which Manager, Planner, Engineer, and Reviewer execute bounded missions over durable project state. Argus separates stable user intent from operational objectives, constraints, and verification criteria, and admits memories, skills, procedures, verifiers, routing decisions, and rejected routes only after role-owned review and, when available, task-native verification. Model weights remain fixed; self-evolution occurs through persistent runtime state and control policy, with autonomous execution between operator-owned escalation points. Across seven GPT-5.5 benchmark arenas, Argus achieves about 78% on SWE-Bench Pro versus 59% for Direct Copilot while using 1.41 times the aggregate tokens. After verification-gated self-evolution, mature SWE-Bench waves use 21% fewer solve-input tokens and 15% less active workflow time per task than startup waves, while recording 34 verifier recoveries and 22 strict review-loop rescues. Argus also reaches 76.8% on AARRI-Bench and a 28.0-point gap on mathematical data synthesis, with competitive GPU-kernel and language-model-training results. Beyond benchmarks, an optimized RWKV6 kernel was merged upstream; a multi-day mathematics campaign retained falsified routes and proof-backed frontier updates; and six paper pipelines completed 254 missions with 16 stage rollbacks. These results show that a fixed-weight, self-evolving harness can revise, recover, and accumulate verified approaches while producing structured trajectories for future supervised and reinforcement learning.

cs.AI↗

arXiv2026

Impact of Interacting Dark Energy on the Growth of Matter Density Perturbations: Observational Constraints from DESI and Multi-Probe Data

We investigate the impact of a non-gravitational dark sector interaction on the growth of matter density perturbations within both the interacting $w$CDM and the dynamical Chevallier-Polarski-Linder (CPL) scenarios. For $w$CDM model, we develop a parameterization for the growth rate based on a second-order approximation for the growth index $γ$ that explicitly includes the coupling constant $α$. Our analysis reveals a theoretical degeneracy: the coupling induces a correction $Δγ\simeq 1.1α$ in both models, allowing an interacting dark energy model to mimic the growth index predicted by certain modified gravity theories. Then, we confront the models with the latest multi-probe observations, including the Pantheon+ sample of Type Ia supernovae, Baryon Acoustic Oscillation (BAO) data from the Sloan Digital Sky Survey (SDSS) and the second data release (DR2) of the Dark Energy Spectroscopic Instrument (DESI), Cosmic Microwave Background (CMB) measurements, Hubble parameter $H(z)$ data, and redshift-space distortion (RSD) measurements. Our analysis finds that the coupling constant is consistent with zero at approximately the $3σ$ and $2σ$ confidence levels for $w$CDM and CPL models, respectively, showing no definitive statistical evidence for a departure from the standard $Λ$CDM cosmology. The observational constraints strongly disfavor the region of parameter space where interacting dark energy can mimic modified gravity, restricting the growth index to a common approximate interval of $0.53 \lesssim γ\lesssim 0.60$ for both models. This reinforces the growth index as a robust diagnostic for distinguishing between a non-minimal interaction in the dark sector and a genuine modification of gravity with current data.

astro-ph.CO↗

arXiv2026

CADRE: Dynamic Catching via Implicit Contact Descriptors and Task-Appropriate Recovery Affordances

Real-world dexterous manipulation often encounters unexpected errors and disturbances, which can lead to catastrophic failures, such as dropping the manipulated object. To address this challenge, we focus on the problem of catching a falling object while it remains within grasping range and, importantly, resetting the system to a configuration favorable for resuming the primary manipulation task. We propose Contact-Aware Dynamic Recovery (CADRE), a reinforcement learning framework that incorporates a Neural Descriptor Field (NDF)-inspired module to extract implicit contact features. Building on these contact features, we introduce an Implicit Recovery Affordance function to encourage recovery to task-appropriate states. Compared to methods that rely solely on object pose or point cloud input, NDFs can directly reason about finger-object correspondence and better establish a recovery target for RL training. Our experiments show that incorporating contact features improves training efficiency, enhances convergence performance for RL training, and ultimately leads to more successful recoveries. Additionally, we demonstrate that CADRE can generalize zero-shot to unseen objects with different geometries.

cs.RO↗