arXiv subjects

Zhengjie Wang

Publications and source records attributed to Zhengjie Wang.

8 recordsLinked to original sources

arXiv2026

Combining Hierarchical Cognitive Process with Process Supervision for Interpretable Scene Safety Understanding

Scene safety understanding plays a life-or-death role in situational awareness in various critical domains. Traditional methods that rely on learning direct mappings between scenes and safety levels often lack interpretability, limiting their reliability in critical applications. An effective approach to overcoming this challenge lies in interpreting human cognitive processes and equipping machine models with analogous cognitive capabilities. This work explores an effective way of integrating scene safety cognitive process modeling and process supervision. Specifically, we first construct a hierarchical cognitive safety structure, which motivates the development of a novel, high-quality scene safety understanding dataset based on multi-step reasoning with process labels. This dataset serves both as a benchmark and a resource to improve the safety reasoning capabilities of Large Language Models (LLMs), while also enabling a granular analysis of intermediate reasoning steps through information flow and saliency-based techniques. Building upon this foundation, we introduce a modular and flexible process supervision framework that reflects the hierarchical nature of human cognition. This framework leverages LLMs as the core architecture and incorporates Low-Rank Adaptation(LoRA) and Mixture-of-Experts (MoE) strategies to enable specialization and collaboration among expert modules, each tasked with specific sub-processes of the overall reasoning chain. Systematic experimental evaluations and analyses confirm that our framework exhibits superior interpretability and performance characteristics compared to traditional approaches.

cs.CL↗

arXiv2026

Structure for Reading, Prose for Writing: Asymmetric Structural Conditioning in Multi-Agent Document Authoring

Multi-agent pipelines that author formal documents must both read a requester's forms and write against them. We report a deployed tender-response system, running an open-weights model under sovereignty constraints, and evaluate it against human-written bids the same organisation actually submitted. On a blind comparison where the system had no worked example available, an LLM judge rated its answers at least as good as the human-submitted answer on $40$ of $55$ ground-truth sections, better on $4$, missing on none, and flagged one unsupported claim in total. Classifying every gap the judge identified shows that $68\%$ were content absent from the system's own sources -- knowledge the human author held and the pipeline was never given -- so only $6$ of the $15$ adverse verdicts involve a deficiency the system could have avoided. A divergence from ground truth is more often an information-availability result than a writing-quality one, and evaluations that do not separate the two understate such systems. Against this backdrop we report a conditioning asymmetry. It is well established that rendering documents as structural markup rather than flat prose improves extraction, and we reproduce that on three reading tasks. The benefit does not transfer to conditioning: converting a bid's \emph{instruction} material from prose to nested XML dropped answer quality from $74\%$ to $48\%$ under a paired comparison. We further find that naming a forbidden construction concentrates rather than removes it -- $96\%$ of surviving defects fall in the two forms the prompt explicitly names -- and that coupling a stochastic annotation to a deterministic windowing function moves the extracted requirement count from $68$ to $51$ on a byte-identical file. Structure belongs where the model reads; prose and self-applied tests belong where it writes.

cs.AI↗

arXiv2026

Electron-like high-temperature superconductivity induced by compressive strain in La2PrNi2O7 thin films

The realization of high-temperature superconductivity in bilayer nickelates under epitaxial compressive strain is widely interpreted as mimicking the effects of high hydrostatic pressure. To test the equivalence of these mechanisms, we investigated a comprehensive strain continuum ranging from compressive (-2.14%) to tensile (+0.91%). Crucially, via ozone-assisted atomic-layer epitaxy, we realized high-temperature superconductivity in as-grown La2PrNi2O7 films on NdAlO3 substrates, which induce the most extreme compressive strain in this material system. Under extreme compression (-2.14%), these films exhibit a Tc_onset of 60 K, zero resistance at 33 K, and a diamagnetic response at 20 K, with magnetotransport measurements confirming a quasi-two-dimensional superconducting nature. Comparing our phase diagram with reported data reveals distinct lattice responses: unlike in pressurized crystals, the superconducting window in epitaxial films diverges significantly in the out-of-plane parameter c (or c/ap ratio) but remains consistent with the bulk regarding the in-plane parameter ap. Crucially, while superconductivity in both systems emerges from the suppression of spin-density waves (SDW), Hall measurements reveal a fundamental electronic dichotomy: optimal superconducting films are intrinsically electron-like (exhibiting a negative Hall coefficient), in stark contrast to the hole-like nature (positive Hall coefficient) of high-pressure bulk crystals and non-superconducting tensile films. Ultimately, both tuning strategies effectively modulate the underlying correlation landscape - the true driver of superconductivity - transcending the constraints of specific Fermi surface topologies. This work establishes a macroscopic platform for probing the multi-orbital physics of nickelates, offering a new dimension for investigating high-temperature superconductivity.

cond-mat.supr-con↗

arXiv2026

Directional-dependent Berezinskii-Kosterlitz-Thouless transition at EuO/KTaO$_3$(111) interfaces

In two dimensions, a phase-coherent superconducting state is established via a Berezinskii-Kosterlitz-Thouless (BKT) transition, whose critical temperature $T_{\rm BKT}$ is determined by the global superfluid stiffness in uniform superconducting systems. We report that at the interface between (111)-oriented KTaO$_3$ and ferromagnetic EuO, the two-dimensional superconducting state exhibits a BKT transition relying on the direction of in-plane bias current. The highest $T_{\rm BKT}$ occurs when current is applied along one of the [11$\bar{2}$] axes of KTaO$_3$, underscoring a spontaneous breaking of the threefold lattice rotational symmetry. Such directional dependence of $T_{\rm BKT}$ is consistently reflected in the nonreciprocal signals stemming from superconducting fluctuations above the transition. We attribute this phenomenon to an interfacial phase segregation; the phase with higher $T_{\rm BKT}$ self-organizes into quasi-one-dimensional textures that stretch along one of the [11$\bar{2}$] directions. Our results point toward the emergence of exotic phases of matter beyond the description of conventional BKT physics at a superconducting interface that is subjected to ferromagnetic proximity.

cond-mat.supr-con↗

arXiv2026

UNeMo: Collaborative Visual-Language Reasoning and Navigation via a Multimodal World Model

Vision-and-Language Navigation (VLN) requires agents to autonomously navigate complex environments via visual images and natural language instructions--remains highly challenging. Recent research on enhancing language-guided navigation reasoning using pre-trained large language models (LLMs) has shown promising prospects. However, the reasoning of such methods is limited to the linguistic modality, lacking visual reasoning capabilities. Moreover, existing reasoning modules are optimized separately from navigation policies, leading to incompatibility and potential conflicts in optimization objectives.To tackle these challenges, we introduce UNeMo, a novel framework designed for the collaborative optimization of visual state reasoning and navigational decision-making. It introduces a Multimodal World Model (MWM) that takes visual features, language instructions, and navigational actions as inputs to jointly predict subsequent visual states, enabling cross-modal reasoning. Via a Hierarchical Prediction-Feedback (HPN) mechanism, MWM collaborates with navigation policies: the first layer generates actions using current vision-and-language features; MWM then infers post-action visual states to guide the second layer's fine-grained decisions. This forms a dynamic bidirectional promotion mechanism where MWM reasoning optimizes navigation policies, while policy decisions feedback to improve MWM's reasoning accuracy. Experiments on R2R and REVERIE datasets show UNeMo outperforms state-of-the-art methods by 2.1% and 0.7% in navigation accuracy for unseen scenes, validating its effectiveness.

cs.AI↗

arXiv2025

Data-driven robust UAV position estimation in GPS signal-challenged environment

In this paper, we consider a position estimation problem for an unmanned aerial vehicle (UAV) equipped with both proprioceptive sensors, i.e. IMU, and exteroceptive sensors, i.e. GPS and a barometer. We propose a data-driven position estimation approach based on a robust estimator which takes into account that the UAV model is affected by uncertainties and thus it belongs to an ambiguity set. We propose an approach to learn this ambiguity set from the data.

math.OC↗

arXiv2024

Dynamic Depth Decoding: Faster Speculative Decoding for LLMs

The acceleration of Large Language Models (LLMs) with speculative decoding provides a significant runtime improvement without any loss of accuracy. Currently, EAGLE-2 is the state-of-the-art speculative decoding method, improving on EAGLE with a dynamic draft tree. We introduce Dynamic Depth Decoding (DDD), which optimises EAGLE-2's tree drafting method using a dynamic depth. This extends the average speedup that EAGLE-2 achieves over EAGLE by $44\%$, giving DDD an average speedup of $3.16$x.

cs.CL↗

arXiv2023

Using fine-tuning and min lookahead beam search to improve Whisper

The performance of Whisper in low-resource languages is still far from perfect. In addition to a lack of training data on low-resource languages, we identify some limitations in the beam search algorithm used in Whisper. To address these issues, we fine-tune Whisper on additional data and propose an improved decoding algorithm. On the Vietnamese language, fine-tuning Whisper-Tiny with LoRA leads to an improvement of 38.49 in WER over the zero-shot Whisper-Tiny setting which is a further reduction of 1.45 compared to full-parameter fine-tuning. Additionally, by using Filter-Ends and Min Lookahead decoding algorithms, the WER reduces by 2.26 on average over a range of languages compared to standard beam search. These results generalise to larger Whisper model sizes. We also prove a theorem that Min Lookahead outperforms the standard beam search algorithm used in Whisper.

eess.AS↗