arXiv subjects

Yuhang Jiang

Publications and source records attributed to Yuhang Jiang.

At least 19 recordsLinked to original sources

arXiv2026

GeoSelect: Spatial-Program Execution for Training-Free Referring Remote Sensing Image Segmentation

Referring remote sensing image segmentation segments the object named by a natural-language expression in an aerial image. Existing training-free methods resolve the expression through implicit vision-language activations or region-text similarity, which gives weak control over the spatial, superlative, and ordinal relations that dominate aerial referring, such as the rightmost ship or the second court from the left. We propose GeoSelect, a training-free pipeline that reframes referring as the execution of a typed spatial program. A frozen, text-only language model synthesises the expression into a small domain-specific language, a well-formedness checker accepts the program, and a deterministic executor runs it. The central abstraction is a single scored candidate set type under which every operator composes: continuous geometric fields realise position and proximity, while discrete set and order operators add the extremum, ordinal, top-k, and relational constructions that fields alone cannot express. Execution is explicit, so every intermediate is inspectable, and a reliability ladder degrades any failing program to the field-only special case. GeoSelect achieves 58.86 mIoU on RRSIS-D test and 55.27 mIoU on RISBench test, more than twice the best prior training-free method on RRSIS-D, with no referring supervision and on a single GPU. Under a fixed detector and segmenter, explicit execution improves over implicit selectors under the same backbone; the best-box-IoU and outcome-partition diagnostics motivate complementary tests of proposal recall and program-path behaviour, with the program path the clearer priority on RISBench, and an exposure audit shows comparable accuracy on the audited unseen subset. Code and configurations are available at https://github.com/Avalon-S/GeoSelect.

cs.CV↗

arXiv2026

The Scissors Effect: When Resize-Based Input Diversity Helps or Hurts Transfer Attacks

Input Diversity (DI), a random resize and pad applied at each attack iteration, is a near-default ingredient of transfer-based attacks, widely assumed to improve transferability. We show this assumption is regime-dependent and, for adversarially trained surrogates, often reversed. Holding the attack fixed and varying only the surrogate, raising the DI probability improves transfer from standard surrogates but degrades it from robust ones: the two response curves separate like a pair of scissors, a pattern we call the Scissors Effect. On ImageNet, blind DI costs a robust source 10.3 percentage points of attack success across four architecturally diverse targets; the effect is several times smaller at 32x32. Direct measurement supports a bias-variance account: DI displaces the gradient by a comparable amount on both groups but reduces its variance only where the gradient is noisy, and robust surrogates have little noise left to average away. A gradient-consistency probe, frozen and hashed before the runs, predicts the sign of the effect on seven unseen surrogates, and we report where it fails alongside where it works. The practical consequence holds independently of the mechanism: leaving DI enabled by default understates the attack a robust surrogate can mount, and so overstates the robustness of the model being evaluated. Code: https://github.com/Avalon-S/ScissorsEffect.

cs.LG↗

arXiv2026

A Circuit, Not The Circuit: Non-Unique Causal Localisation of the Mamba-2 State Sink

Mechanistic interpretability routinely reads a probe and labels its top-activating units as the circuit executing the computation. We test the move in Mamba, on the state sink: the selective state-space analogue of the Transformer attention sink, where the Delta-gate fires disproportionately on boundary tokens such as BOS and newline. At Mamba-1 channel granularity the probe's units carry the causal effect. At Mamba-2 head granularity the label-to-locus link breaks in three ways. The causal set is not unique: a near-disjoint set of top-activation heads, sharing almost none of the specialists, matches or exceeds their ablation effect. Representation does not track function: dual heads, a quarter to over a third of all heads, respond alike to BOS and newline yet do not reproduce the specialists' causal profile. And the effect dissociates by intervention surface: perturbing the input-dependent Delta the probe reads moves the loss by at most 0.16 nats, while suppressing the same heads' output moves it by 1.5 to 7.3 nats. The heads still matter behaviourally, dropping needle-retrieval success on a RULER-style probe from 30/30 to 0/30 at 1024 context on both flagship checkpoints. A representational signature in a selective state-space model thus marks a behaviourally important head set without pinning down the circuit: a single probe recovers a circuit, not the circuit.

cs.CL↗

arXiv2026

TRACE: A Unified Rollout Budget Allocation Framework for Efficient Agentic Reinforcement Learning

Reinforcement learning with verifiable rewards (RLVR) is a promising approach for enhancing reasoning and agentic behavior in large language models. However, rollout-intensive policy optimization is often limited by insufficient reward contrast, arising when overly simple or complex prompts generate low-variance feedback and when outcome-only rewards assign the same terminal assessment to every decision in a multi-turn rollout. Past efforts have focused on allocating available rollout resources to promising prompts, yet they only leverage sample informativeness at the prompt level and neglect variation in prefix-level informativeness across turns within the same rollout. This work targets multi-turn agentic RL by modeling each ReAct-style thought-action-observation turn as a semantically distinct node, allowing budget allocation to extend from prompt roots to turn-level prefixes with further continuations, which naturally forms tree-structured rollouts. We introduce Tree Rollout Allocation for Contrastive Exploration (TRACE), a unified rollout allocation framework that enhances reward contrast within a fixed sampling budget. Technically, TRACE allocates rollout budget to both prompt roots and intermediate prefixes that are most likely to yield mixed terminal rewards. A shared generalizable predictor estimates conditional success probability at these anchors from prefix histories to guide this allocation. The resulting adaptive tree structure enriches outcome-only feedback and amplifies the policy-update signal. Empirically, TRACE achieves competitive performance and efficiency gains on typical agentic benchmarks, e.g., improving Qwen3-14B Multi-Hop QA average accuracy by 2.8 points over competitive baselines at equal sampling cost.

cs.LG↗

arXiv2026

Chaos Is a LADDER: Domain Generalization Beyond Invariance via Reweighting

Domain generalization (DG) aims to learn from multiple source domains and generalize to unseen target domains. Most DG methods pursue invariance: they seek a causal representation whose prediction rule is invariant across domains. This principle is effective when the causal mechanism is stable, but becomes restrictive when the domain itself modulates how causal content maps to the response. In this case, directly feeding domain style into the predictor can create misleading shortcuts, since style does not by itself cause the response. Yet the apparent chaos of multiple styles can become a ladder: style can locate the unseen target domain among source domains and guide which domain-dependent prediction rules should be trusted. We propose \emph{Latent Adaptive Domain Disentanglement and Environment Reweighting} (LADDER), a fixed-model DG pipeline that learns causal/style representations, freezes the encoders, fits source-specific classifiers, and uses an unlabeled target-domain covariate set only at inference to compute weights over these fixed classifiers, with no target labels or model-state updates. We establish theoretical guarantees for source reweighting and validate LADDER on simulations, FMoW, and a location-grouped iWildCam protocol, with gains in overall and group-averaged accuracy.

stat.ML↗

arXiv2026

Task Structure Reverses Layerwise State Encoding in Sequence Models

Mechanistic studies of sequence models often treat layerwise state encodings as architectural traits: recurrent models concentrate readable state, attention-based models distribute it. We find that the same architecture reverses this profile when the task changes. Across Transformers, Mamba, Mamba-2, LSTMs, and GRUs, Parity is concentrated late in Mamba and the recurrent baselines and built gradually by Transformer; on bounded-depth Dyck-k the pattern flips. The same flip appears in fine-tuned Mamba-130M and Pythia-160M, and the Pythia Dyck bottleneck persists at 410M. Two explanations are conflated in the literature: algebraic structure (commutativity) versus computational structure (prefix update vs. stack). To separate them we add a third task: non-commutative S_3 permutation composition. S_3 groups with Parity, not Dyck, on layerwise probing across all five architectures and on Mamba-specific Conv1D attribution, so the grouping tracks computational structure rather than commutativity. Causal interventions show that, in the 4-layer formal models, linearly readable directions are often functionally necessary and can remain important at out-of-distribution lengths on Parity and Dyck. At pretrained scale the picture splits. Fine-tuned Pythia Dyck has a strong middle-layer bottleneck (L6-L7 ablation drops accuracy by roughly 81% at 160M; broader L4-L18 plateau at 410M), far weaker at the best-probe layer. Pretrained Mamba shows the complementary failure mode: its final layer is highly readable, no single probe direction breaks the task on Parity, Dyck, or S_3, yet mid-position activation patching there recovers about 97-98% of the clean-corrupted logit gap. Probing localizes where state is linearly available, not always where the computation is bottlenecked. Mechanistic signatures are properties of architecture and task together.

cs.LG↗

arXiv2026

Stop Wandering, Find the Keys: LLMs Discriminate Key States for Efficient Multi-Agent Exploration

With expansive state-action spaces, efficient multi-agent exploration remains a longstanding challenge in reinforcement learning. Although pursuing novelty, diversity, or uncertainty attracts increasing attention, redundant efforts brought by exploration without proper guidance choices poses a practical issue for the community. This paper introduces a systematic approach, termed LEMAE, choosing to channel informative task-relevant guidance from a knowledgeable Large Language Model (LLM) for Efficient Multi-Agent Exploration. Specifically, we ground linguistic knowledge from LLM into symbolic key states, that are critical for task fulfillment, in a discriminative manner at low LLM inference costs. To unleash the power of key states, we design Subspace-based Hindsight Intrinsic Reward (SHIR) to guide agents toward key states by increasing reward density. Additionally, we build the Key State Memory Tree (KSMT) to track transitions between key states in a specific task for organized exploration. Benefiting from diminishing redundant explorations, LEMAE outperforms existing SOTA approaches on the challenging benchmarks (e.g., SMAC and MPE) by a large margin, achieving a 10x acceleration in certain scenarios.

cs.AI↗

arXiv2026

PInVerify: An Offline Embodied Benchmark for Active Instance Verification

Embodied agents have made strong progress in navigating to target objects, but reaching the goal vicinity does not guarantee that the agent has found the correct instance: subtle attribute differences (e.g., "white floral" vs. "white striped") often require close-range, multi-view inspection. We address this gap with Active Instance Verification (AIV), a task in which an agent actively selects viewpoints around a candidate object to decide whether it matches a fine-grained natural-language description. We formalize AIV as a finite-horizon decision process and introduce PInVerify, an offline embodied benchmark for AIV: 3,000 evaluation episodes across 18 object categories, delivered as multi-view captures with a 6-sector navigation topology that exposes trap views (navigable but uninformative) and unreachable sectors. As reference baselines we build a training-free pipeline and a LoRA-fine-tuned end-to-end agent around open-source multimodal large language models (MLLMs) at on-device scale ($\leq$8B parameters), with attribute decomposition, a visibility-weighted multi-view tracker, and three next-best-view (NBV) strategies. In our evaluation across Qwen3-VL (4B/8B), SenseNova-SI-1.2-InternVL3-8B, CLIP, and SigLIP2, the best MLLM-based baseline exceeds the best embedding baseline by 4.9 pp; GT-box ablations show a +3.1 pp detection gap; and we do not observe reliable gains from active viewpoint selection within the tested NBV strategies. A LoRA-fine-tuned agent (SFT+GSPO) reaches 85.6%. PInVerify aims to support further work on active, fine-grained semantic verification in embodied AI. Code: https://github.com/Avalon-S/PInVerify.

cs.CV↗

arXiv2026

Listwise Policy Optimization: Group-based RLVR as Target-Projection on the LLM Response Simplex

Reinforcement learning with verifiable rewards (RLVR) has become a standard approach for large language models (LLMs) post-training to incentivize reasoning capacity. Among existing recipes, group-based policy gradient is prevalent, which samples a group of responses per prompt and updates the policy via group-relative advantage signals. This work reveals that these optimization strategies share a common geometric structure: each implicitly defines a target distribution on the response simplex and projects toward it via first-order approximation. Building on this insight, we propose Listwise Policy Optimization (LPO) to explicitly conduct the target-projection, which demystifies the implicit target by restricting the proximal RL objective to the response simplex, and then projects the policy via exact divergence minimization. This framework provides (i) monotonic improvement on the listwise objective with bounded, zero-sum, and self-correcting projection gradients, and (ii) flexibility in divergence selection with distinct structural properties through the decoupled projection step. On diverse reasoning tasks and LLM backbones, LPO consistently improves training performance over typical policy gradient baselines under matched targets, while intrinsically preserving optimization stability and response diversity.

cs.LG↗

arXiv2026

Small Generalizable Prompt Predictive Models Can Steer Efficient RL Post-Training of Large Reasoning Models

Reinforcement learning enhances the reasoning capabilities of large language models but often involves high computational costs due to rollout-intensive optimization. Online prompt selection presents a plausible solution by prioritizing informative prompts to improve training efficiency. However, current methods either depend on costly, exact evaluations or construct prompt-specific predictive models lacking generalization across prompts. This study introduces Generalizable Predictive Prompt Selection (GPS), which performs Bayesian inference towards prompt difficulty using a lightweight generative model trained on the shared optimization history. Intermediate-difficulty prioritization and history-anchored diversity are incorporated into the batch acquisition principle to select informative prompt batches. The small predictive model also generalizes at test-time for efficient computational allocation. Experiments across varied reasoning benchmarks indicate GPS's substantial improvements in training efficiency, final performance, and test-time efficiency over superior baseline methods.

cs.AI↗

arXiv2026

Discovery of a hybridization-wave electronic order in a van der Waals Kondo lattice

Kondo lattice systems, in which localized magnetic moments coherently hybridize with itinerant electrons, exhibit a rich landscape of emergent quantum phenomena. Within this framework, the hybridization strength itself has been theoretically proposed as a spatially modulated order parameter, giving rise to a so-called hybridization wave. However, direct experimental evidence of this quantum state has remained an outstanding challenge. Here, we report the direct observation of a hybridization wave in the layered transition metal dichalcogenide 6R-TaS2, a naturally occurring heterostructure composed of alternating 1T- and 1H-TaS2 layers. Using scanning tunneling microscopy and spectroscopy (STM/STS), we identify the hybridization gap in 1T layer, demonstrating the establishment of a coherent Kondo lattice. Notably, we discover that the hybridization gap present a uniaxial unit-cell doubling modulation, which breaks the both translational and rotational symmetries of the underlying Star-of-David superlattice. Such unit-cell doubling is not caused by structural topography, and therefore, constitutes the real-space visualization of the hybridization-wave order. Furthermore, the hybridization wave correlates with an energy-dependent nematic order that shares the same periodicity and orientation, revealing intertwined electronic instabilities. Our findings not only validate a long-standing prediction but also establish layer-engineered van der Waals materials as a versatile platform for exploring and controlling hybridization-driven quantum phases.

cond-mat.str-el↗

arXiv2026

Closing Reasoning Gaps in Clinical Agents with Differential Reasoning Learning

Clinical decision support requires not only correct answers but also clinically valid reasoning. We propose Differential Reasoning Learning (DRL), a framework that improves clinical agents by learning from reasoning discrepancies. From reference reasoning rationales (e.g., physician-authored clinical rationale, clinical guidelines, or outputs from more capable models) and the agent's free-form chain-of-thought (CoT), DRL extracts reasoning graphs as directed acyclic graphs (DAGs) and performs a clinically weighted graph edit distance (GED)-based discrepancy analysis. An LLM-as-a-judge aligns semantically equivalent nodes and diagnoses discrepancies between graphs. These graph-level discrepancy diagnostics are converted into natural-language instructions and stored in a Differential Reasoning Knowledge Base (DR-KB). At inference, we retrieve top-$k$ instructions via Retrieval-Augmented Generation (RAG) to augment the agent prompt and patch likely logic gaps. Evaluation on open medical question answering (QA) benchmarks and a Return Visit Admissions (RVA) prediction task from internal clinical data demonstrates gains over baselines, improving both final-answer accuracy and reasoning fidelity. Ablation studies confirm gains from infusing reference reasoning rationales and the top-$k$ retrieval strategy. Clinicians' review of the output provides further assurance of the approach. Together, results suggest that DRL supports more reliable clinical decision-making in complex reasoning scenarios and offers a practical mechanism for deployment under limited token budgets.

cs.AI↗

arXiv2026

A Benchmark for End-to-End Zero-Shot Biomedical Relation Extraction with LLMs: Experiments with OpenAI Models

Extracting relations from scientific literature is a fundamental task in biomedical NLP because entities and relations among them drive hypothesis generation and knowledge discovery. As literature grows rapidly, relation extraction (RE) is indispensable to curate knowledge graphs to be used as computable structured and symbolic representations. With the rise of LLMs, it is pertinent to examine if it is better to skip tailoring supervised RE methods, save annotation burden, and just use zero shot RE (ZSRE) via LLM API calls. In this paper, we propose a benchmark with seven biomedical RE datasets with interesting characteristics and evaluate three Open AI models (GPT-4, o1, and GPT-OSS-120B) for end-to-end ZSRE. We show that LLM-based ZSRE is inching closer to supervised methods in performances on some datasets but still struggles on complex inputs expressing multiple relations with different predicates. Our error analysis reveals scope for improvements.

cs.CL↗

arXiv2025

A Riemannian Accelerated Proximal Gradient Method

Riemannian accelerated gradient methods have been well studied for smooth optimization, typically treating geodesically convex and geodesically strongly convex cases separately. However, their extension to nonsmooth problems on manifolds with theoretical acceleration remains underexplored. To address this issue, we propose a unified Riemannian accelerated proximal gradient method for problems of the form $F(x) = f(x) + h(x)$ on manifolds, where $f$ can be either geodesically convex or geodesically strongly convex, and $h$ is $ρ$-retraction-convex, possibly nonsmooth. We rigorously establish accelerated convergence rate under reasonable conditions. Additionally, we introduce a safeguard mechanism to ensure global convergence in non-convex settings. Numerical results validate the theoretical acceleration of the proposed method.

math.OC↗

arXiv2025

McKean-Vlasov SPDEs driven by Poisson random measure: Well-posedness and large deviation principle

In this work, we investigate the McKean-Vlasov stochastic partial differential equations driven by Poisson random measure. By adapting the variational framework, we prove the well-posedness and large deviation principle for a class of McKean-Vlasov stochastic partial differential equations with monotone coefficients. The main results can be applied to quasi-linear McKean-Vlasov equations such as distribution dependent stochastic porous media equation and stochastic p-Laplace equation. Our proof is based on the weak convergence approach introduced by Budhiraja et al. for Poisson random measures, the time discretization procedure and relative entropy estimates. In particular, we succeed in dropping the compactness assumption of embedding in the Gelfand triple in order to deal with the case of bounded and unbounded domains in applications.

math.PR↗

arXiv2025

Relation Extraction with Instance-Adapted Predicate Descriptions

Relation extraction (RE) is a standard information extraction task playing a major role in downstream applications such as knowledge discovery and question answering. Although decoder-only large language models are excelling in generative tasks, smaller encoder models are still the go to architecture for RE. In this paper, we revisit fine-tuning such smaller models using a novel dual-encoder architecture with a joint contrastive and cross-entropy loss. Unlike previous methods that employ a fixed linear layer for predicate representations, our approach uses a second encoder to compute instance-specific predicate representations by infusing them with real entity spans from corresponding input instances. We conducted experiments on two biomedical RE datasets and two general domain datasets. Our approach achieved F1 score improvements ranging from 1% to 2% over state-of-the-art methods with a simple but elegant formulation. Ablation studies justify the importance of various components built into the proposed architecture.

cs.CL↗

arXiv2025

Signature of gate tunable superconducting network in twisted bilayer graphene

Twisted van der Waals materials provide a tunable platform for investigating two-dimensional superconductivity and quantum phases. Using spectra-imaging scanning tunneling microscopy, we study the superconducting states in twisted bilayer graphene and track their evolution from insulating phases. Gate-dependent spectroscopic measurements reveal two distinct regimes: under-doped (ν = -2.3) and optimally doped (ν = -2.6). In the under-doped regime, partial superconductivity arises, forming a network interspersed with non-gapped regions. At optimal doping, the entire unit cell demonstrates superconductivity, with gap size modulation showing an anti-correlation with the local density of states. This gate-dependent transition from an insulating phase to a modulated superconductor uncovers an unexpected spatial hierarchy in pairing behavior and offers direct microscopic insights to constrain theories of superconductivity in moiré systems.

cond-mat.supr-con↗

arXiv2025

Unsupervised Data Generation for Offline Reinforcement Learning: A Perspective from Model

Offline reinforcement learning (RL) recently gains growing interests from RL researchers. However, the performance of offline RL suffers from the out-of-distribution problem, which can be corrected by feedback in online RL. Previous offline RL research focuses on restricting the offline algorithm in in-distribution even in-sample action sampling. In contrast, fewer work pays attention to the influence of the batch data. In this paper, we first build a bridge over the batch data and the performance of offline RL algorithms theoretically, from the perspective of model-based offline RL optimization. We draw a conclusion that, with mild assumptions, the distance between the state-action pair distribution generated by the behavioural policy and the distribution generated by the optimal policy, accounts for the performance gap between the policy learned by model-based offline RL and the optimal policy. Secondly, we reveal that in task-agnostic settings, a series of policies trained by unsupervised RL can minimize the worst-case regret in the performance gap. Inspired by the theoretical conclusions, UDG (Unsupervised Data Generation) is devised to generate data and select proper data for offline training under tasks-agnostic settings. Empirical results demonstrate that UDG can outperform supervised data generation on solving unknown tasks.

cs.LG↗