arXiv subjects

Jiajun He

Publications and source records attributed to Jiajun He.

At least 37 records · Page 2Linked to original sources

arXiv2026

Evaluating Large Language Models in Scientific Discovery

Large language models (LLMs) are increasingly applied to scientific research, yet prevailing science benchmarks probe decontextualized knowledge and overlook the iterative reasoning, hypothesis generation, and observation interpretation that drive scientific discovery. We introduce a scenario-grounded benchmark that evaluates LLMs across biology, chemistry, materials, and physics, where domain experts define research projects of genuine interest and decompose them into modular research scenarios from which vetted questions are sampled. The framework assesses models at two levels: (i) question-level accuracy on scenario-tied items and (ii) project-level performance, where models must propose testable hypotheses, design simulations or experiments, and interpret results. Applying this two-phase scientific discovery evaluation (SDE) framework to state-of-the-art LLMs reveals a consistent performance gap relative to general science benchmarks, diminishing return of scaling up model sizes and reasoning, and systematic weaknesses shared across top-tier models from different providers. Large performance variation in research scenarios leads to changing choices of the best performing model on scientific discovery projects evaluated, suggesting all current LLMs are distant to general scientific "superintelligence". Nevertheless, LLMs already demonstrate promise in a great variety of scientific discovery projects, including cases where constituent scenario scores are low, highlighting the role of guided exploration and serendipity in discovery. This SDE framework offers a reproducible benchmark for discovery-relevant evaluation of LLMs and charts practical paths to advance their development toward scientific discovery.

cs.AI↗

arXiv2026

Multiresonant Membrane Metasurfaces for Multifunctional Fingerprint Recognition and Real-time Biochemical Tracking

Label-free identification and real-time tracking of biochemical substances became critical for molecular diagnostics and chemical analysis, yet conventional resonant terahertz metasurface sensing relies on a single resonance, limiting spectral selectivity and dynamic capability. Here, we suggest multiresonant membrane metasurfaces and implement them for simultaneous static molecular fingerprint retrieval and dynamic reaction monitoring within a single pixel. We consider a membrane metasurface supporting multiple quasi-bound states in the continuum designed at target frequencies and enabling the tailoring of the field enhancement and frequency-selective interaction with target analytes. As a proof-of-concept, we achieve label-free detection of the dual fingerprint absorption features of pefloxacin at 0.78 THz and 0.99 THz, and real-time tracking of vitamin C oxidation and denaturation under ambient conditions. The kinetic profiles extracted from the THz amplitude evolution show excellent agreement with nonlinear reaction models, demonstrating quantitative biochemical tracking capabilities. Our results establish a versatile and scalable THz photonic platform that unifies static fingerprint identification and dynamic reaction monitoring, paving the way toward integrated on-chip biochemical analytics and multifunctional metasurface sensors.

physics.optics↗

arXiv2026

Accelerated Parallel Tempering via Neural Transports

Markov Chain Monte Carlo (MCMC) algorithms are essential tools in computational statistics for sampling from unnormalised probability distributions, but can be fragile when targeting high-dimensional, multimodal, or complex target distributions. Parallel Tempering (PT) enhances MCMC's sample efficiency through annealing and parallel computation, propagating samples from tractable reference distributions to intractable targets via state swapping across interpolating distributions. The effectiveness of PT is limited by the often minimal overlap between adjacent distributions in challenging problems, which requires increasing the computational resources to compensate. We introduce a framework that accelerates PT by leveraging neural samplers -- including normalising flows, diffusion models, and controlled diffusions -- to reduce the required overlap. Our approach utilises neural samplers in parallel, circumventing the computational burden of neural samplers while preserving the asymptotic consistency of classical PT. We demonstrate theoretically and empirically on a variety of multimodal sampling problems that our method improves sample quality, reduces the computational cost compared to classical PT, and enables efficient free energy/normalising constant estimation.

stat.ML↗

arXiv2026

Assessing generative modeling approaches for free energy estimates in condensed matter

The accurate estimation of free energy differences between two states is a long-standing challenge in molecular simulations. Traditional approaches generally rely on sampling multiple intermediate states to ensure sufficient overlap in phase space and are, consequently, computationally expensive. Boltzmann Generators and related generative-model-based methods have recently addressed this challenge by learning a direct probability density transform between two states. However, it remains unclear which approach provides the best trade-off between efficiency, accuracy, and scalability. In this work, we review and benchmark selected generative approaches for condensed-matter systems, including discrete and continuous normalizing flows for targeted free energy perturbation and FEAT (Free Energy Estimators with Adaptive Transport) combined with the escorted Jarzynski equality, using coarse-grained monatomic ice and Lennard-Jones solids as benchmark systems. All models yield highly accurate free energy estimates and, depending on the system, may require fewer energy evaluations than traditional methods. Continuous flows and FEAT are most efficient in energy evaluations, whereas discrete flows have substantially lower inference cost. By releasing all data together with our results, we enable future benchmarking of free energy estimation methods in condensed-phase systems.

cond-mat.stat-mech↗

arXiv2026

VisionCreator: A Native Visual-Generation Agentic Model with Understanding, Thinking, Planning and Creation

Visual content creation tasks demand a nuanced understanding of design conventions and creative workflows-capabilities challenging for general models, while workflow-based agents lack specialized knowledge for autonomous creative planning. To overcome these challenges, we propose VisionCreator, a native visual-generation agentic model that unifies Understanding, Thinking, Planning, and Creation (UTPC) capabilities within an end-to-end learnable framework. Our work introduces four key contributions: (i) VisGenData-4k and its construction methodology using metacognition-based VisionAgent to generate high-quality creation trajectories with explicit UTPC structures; (ii) The VisionCreator agentic model, optimized through Progressive Specialization Training (PST) and Virtual Reinforcement Learning (VRL) within a high-fidelity simulated environment, enabling stable and efficient acquisition of UTPC capabilities for complex creation tasks; (iii) VisGenBench, a comprehensive benchmark featuring 1.2k test samples across diverse scenarios for standardized evaluation of multi-step visual creation capabilities; (iv) Remarkably, our VisionCreator-8B/32B models demonstrate superior performance over larger closed-source models across multiple evaluation dimensions. Overall, this work provides a foundation for future research in visual-generation agentic systems.

cs.CV↗

arXiv2026

CREPE: Controlling Diffusion with Replica Exchange

Inference-time control of diffusion models aims to steer model outputs to satisfy new constraints without retraining. Previous approaches have mostly relied on heuristic guidance or have been coupled with Sequential Monte Carlo (SMC) for bias correction. In this paper, we propose a flexible alternative based on replica exchange, an algorithm designed initially for sampling problems. We refer to this method as CREPE (Controlling with REPlica Exchange). Unlike SMC, CREPE: (1) generates particles sequentially, (2) maintains high diversity in the generated samples after a burn-in period, and (3) enables online refinement or early termination. We demonstrate its versatility across various tasks, including temperature annealing, reward-tilting, model composition and classifier-free guidance debiasing, with competitive performance compared to prior SMC methods.

cs.LG↗

arXiv2026

Topology-Aware Coordination for Multi-Functional Low-Altitude Wireless Networks

Low-altitude wireless networks (LAWNs) are expected to consist of multi-tier, heterogeneous terrestrial and non-terrestrial devices, where effective coordination is essential to fully unlock the complementary capabilities of diverse systems from different vendors. To address this issue, we propose a novel multi-functional coordination framework that enables seamless cooperation within the LAWN while supporting efficient execution of diverse network functions. In the proposed architecture, each device or infrastructure element is assigned to a specific functional role, namely, edge mobile terminal (E-MT), distributed MT (D-MT), or computing center. E-MTs are equipped with lightweight, independent signal processing and computing capabilities, while D-MTs and the computing center handle regional and global coordination, respectively. To enhance the overall network efficiency, we model the LAWN as a sparse graph, where nodes represent network nodes and edges are defined according to a set of controllable connection rules. This topology-aware (TA) representation allows for efficiently solving various coordination tasks across the network. Numerical results show that the proposed TA coordination framework outperforms baseline approaches that lack topological insights, achieving higher efficiency in multi-task coordination. Finally, we discuss key technical challenges and outline potential solutions for future deployment.

cs.IT↗

arXiv2026

Delay Alignment Modulation for Secure ISAC Systems

This paper introduces delay-alignment modulation (DAM) for secure integrated sensing and communication (ISAC). Due to the broadcast nature of multi-user downlinks, communications are vulnerable to eavesdropping. DAM applies controlled per-path symbol delays at the transmitter to coherently align the multipath components at the intended user, enhancing the received signal power, while simultaneously creating delay misalignment at the eavesdropper (Eve). To mitigate sensing degradation caused by multipath propagation, we propose a two-stage protocol that first estimates the angle and then the delay of the line-of-sight (LoS) path after suppressing multipath interference. We derive the secrecy spectral efficiency (SSE) and the Cramer-Rao (CRB) of the target delay. Finally, we develop a path-based zero-forcing (ZF) precoding framework and formulate a max-min SSE design under CRB and power constraints. Simulation results show DAM significantly outperforms the strongest-path (SP) benchmark in terms of SSE, while meeting sensing requirements, since intentional delay alignment at legitimate users degrades reception at Eve.

cs.IT↗

arXiv2026

DiffRatio: Training One-Step Diffusion Models Without Teacher Supervision

Score-based distillation methods (e.g., variational score distillation) train one-step diffusion models by first pre-training a teacher score model and then distilling it into a one-step student model. However, the gradient estimator in the distillation stage usually suffers from two sources of bias: (1) biased teacher supervision due to score estimation error incurred during pre-training, and (2) the student model's score estimation error during distillation. These biases can degrade the quality of the resulting one-step diffusion model. To address this, we propose DiffRatio, a new framework for training one-step diffusion models: instead of estimating the teacher and student scores independently and then taking their difference, we directly estimate the score difference as the gradient of a learned log density ratio between the student and data distributions across diffusion time steps. This approach greatly simplifies the training pipeline, significantly reduces gradient estimation bias, and improves one-step generation quality. Additionally, it also reduces auxiliary network size by using a lightweight density-ratio network instead of two full score networks, which improves computational and memory efficiency. DiffRatio achieves competitive one-step generation results on CIFAR-10 and ImageNet (64x64 and 512x512), outperforming most teacher-supervised distillation methods. Moreover, the learned density ratio naturally serves as a verifier, enabling a principled inference-time parallel scaling scheme that further improves the generation quality without external rewards or additional sequential computation.

cs.LG↗

arXiv2026

A Comprehensive Study on the Effectiveness of ASR Representations for Noise-Robust Speech Emotion Recognition

This paper proposes an efficient attempt to noisy speech emotion recognition (NSER). Conventional NSER approaches have proven effective in mitigating the impact of artificial noise sources, such as white Gaussian noise, but are limited to non-stationary noises in real-world environments due to their complexity and uncertainty. To overcome this limitation, we introduce a new method for NSER by adopting the automatic speech recognition (ASR) model as a noise-robust feature extractor to eliminate non-vocal information in noisy speech. We first obtain intermediate layer information from the ASR model as a feature representation for emotional speech and then apply this representation for the downstream NSER task. Our experimental results show that 1) the proposed method achieves better NSER performance compared with the conventional noise reduction method, 2) outperforms self-supervised learning approaches, and 3) even outperforms text-based approaches using ASR transcription or the ground truth transcription of noisy speech.

cs.SD↗

arXiv2025

Pinching-Antenna System-Assisted Localization: A Stochastic Geometry Perspective

This paper proposes a novel localization framework underpinned by a pinching-antenna (PA) system, in which the target location is estimated using received signal strength (RSS) measurements obtained from downlink signals transmitted by the PAs. To develop a comprehensive analytical framework, we employ stochastic geometry to model the spatial distribution of the PAs, enabling tractable and insightful network-level performance analysis. Closed-form expressions for target localizability and the Cramer-Rao lower bound (CRLB) distribution are analytically derived, enabling the evaluation of the fundamental limits of PA-assisted localization systems without extensive simulations. Furthermore, the proposed framework provides practical guidance for selecting the optimal waveguide number to maximize localization performance. Numerical results also highlight the superiority of the PA-assisted approach over conventional fixed-antenna systems in terms of the CRLB.

eess.SP↗

arXiv2025

Adversarial Learning-Based Radio Map Reconstruction for Fingerprinting Localization

This letter presents a feature-guided adversarial framework, namely ComGAN, which is designed to reconstruct an incomplete fingerprint database by inferring missing received signal strength (RSS) values at unmeasured reference points (RPs). An auxiliary subnetwork is integrated into a conditional generative adversarial network (cGAN) to enable spatial feature learning. An optimization method is then developed to refine the RSS predictions by aggregating multiple prediction sets, achieving an improved localization performance. Experimental results demonstrate that the proposed scheme achieves a root mean squared error (RMSE) comparable to the ground-truth measurements while outperforming state-of-the-art reconstruction methods. When the reconstructed fingerprint is combined with measured data for training, the fingerprinting localization achieves accuracy comparable to models trained on fully measured datasets.

eess.SP↗

arXiv2025

BudgetLeak: Membership Inference Attacks on RAG Systems via the Generation Budget Side Channel

Retrieval-Augmented Generation (RAG) enhances large language models by integrating external knowledge, but reliance on proprietary or sensitive corpora poses various data risks, including privacy leakage and unauthorized data usage. Membership inference attacks (MIAs) are a common technique to assess such risks, yet existing approaches underperform in RAG due to black-box constraints and the absence of strong membership signals. In this paper, we identify a previously unexplored side channel in RAG systems: the generation budget, which controls the maximum number of tokens allowed in a generated response. Varying this budget reveals observable behavioral patterns between member and non-member queries, as members gain quality more rapidly with larger budgets. Building on this insight, we propose BudgetLeak, a novel membership inference attack that probes responses under different budgets and analyzes metric evolution via sequence modeling or clustering. Extensive experiments across four datasets, three LLM generators, and two retrievers demonstrate that BudgetLeak consistently outperforms existing baselines, while maintaining high efficiency and practical viability. Our findings reveal a previously overlooked data risk in RAG systems and highlight the need for new defenses.

cs.CR↗

arXiv2025

Deep-Learning-Empowered Programmable Topolectrical Circuits

Topolectrical circuits provide a versatile platform for exploring and simulating modern physical models. However, existing approaches suffer from incomplete programmability and ineffective feature prediction and control mechanisms, hindering the investigation of physical phenomena on an integrated platform and limiting their translation into practical applications. Here, we present a deep learning empowered programmable topolectrical circuits (DLPTCs) platform for physical modeling and analysis. By integrating fully independent, continuous tuning of both on site and off site terms of the lattice Hamiltonian, physics graph informed inverse state design, and immediate hardware verification, our system bridges the gap between theoretical modeling and practical realization. Through flexible control and adiabatic path engineering, we experimentally observe the boundary states without global symmetry in higher order topological systems, their adiabatic phase transitions, and the flat band like characteristic corresponding to Landau levels in the circuit. Incorporating a physics graph informed mechanism with a generative AI model for physics exploration, we realize arbitrary, position controllable on board Anderson localization, surpassing conventional random localization. Utilizing this unique capability with high fidelity hardware implementation, we further demonstrate a compelling cryptographic application: hash based probabilistic information encryption by leveraging Anderson localization with extensive disorder configurations, enabling secure delivery of full ASCII messages.

cond-mat.dis-nn↗

arXiv2025

Progressive Tempering Sampler with Diffusion

Recent research has focused on designing neural samplers that amortize the process of sampling from unnormalized densities. However, despite significant advancements, they still fall short of the state-of-the-art MCMC approach, Parallel Tempering (PT), when it comes to the efficiency of target evaluations. On the other hand, unlike a well-trained neural sampler, PT yields only dependent samples and needs to be rerun -- at considerable computational cost -- whenever new samples are required. To address these weaknesses, we propose the Progressive Tempering Sampler with Diffusion (PTSD), which trains diffusion models sequentially across temperatures, leveraging the advantages of PT to improve the training of neural samplers. We also introduce a novel method to combine high-temperature diffusion models to generate approximate lower-temperature samples, which are minimally refined using MCMC and used to train the next diffusion model. PTSD enables efficient reuse of sample information across temperature levels while generating well-mixed, uncorrelated samples. Our method significantly improves target evaluation efficiency, outperforming diffusion-based neural samplers.

cs.LG↗

arXiv2025

FinGAN: An Interpretable RSS Generation Network for Scalable Fingerprint Localization

This work introduces FinGAN, a robust received signal strength (RSS) data generator designed to expand RSS fingerprint datasets. Compared to existing generative adversarial models that either rely on known reference positions (RPs) or depend on predefined priors, FinGAN learns the latent information between RPs and RSS values by maximizing the mutual information between the generated RSS data and the RPs, enabling an end-to-end RSS generation directly from RPs. This allows us to accurately generate RSS data for previously unmeasured RPs. Both quantitative and qualitative evaluations demonstrate that FinGAN produces synthetic RSS data closely aligned with real RSS sample collected from the on-site experiment, preserving localization performance comparable to that achieved with complete real-world datasets. To further validate its generalizability, FinGAN is also trained and evaluated on open-source datasets from three typical office environments,and the results demonstrate consistent performance across different scenarios.

eess.SP↗

arXiv2025

M4SER: Multimodal, Multirepresentation, Multitask, and Multistrategy Learning for Speech Emotion Recognition

Multimodal speech emotion recognition (SER) has emerged as pivotal for improving human-machine interaction. Researchers are increasingly leveraging both speech and textual information obtained through automatic speech recognition (ASR) to comprehensively recognize emotional states from speakers. Although this approach reduces reliance on human-annotated text data, ASR errors possibly degrade emotion recognition performance. To address this challenge, in our previous work, we introduced two auxiliary tasks, namely, ASR error detection and ASR error correction, and we proposed a novel multimodal fusion (MF) method for learning modality-specific and modality-invariant representations across different modalities. Building on this foundation, in this paper, we introduce two additional training strategies. First, we propose an adversarial network to enhance the diversity of modality-specific representations. Second, we introduce a label-based contrastive learning strategy to better capture emotional features. We refer to our proposed method as M4SER and validate its superiority over state-of-the-art methods through extensive experiments using IEMOCAP and MELD datasets.

cs.HC↗

arXiv2025

PARCO: Phoneme-Augmented Robust Contextual ASR via Contrastive Entity Disambiguation

Automatic speech recognition (ASR) systems struggle with domain-specific named entities, especially homophones. Contextual ASR improves recognition but often fails to capture fine-grained phoneme variations due to limited entity diversity. Moreover, prior methods treat entities as independent tokens, leading to incomplete multi-token biasing. To address these issues, we propose Phoneme-Augmented Robust Contextual ASR via COntrastive entity disambiguation (PARCO), which integrates phoneme-aware encoding, contrastive entity disambiguation, entity-level supervision, and hierarchical entity filtering. These components enhance phonetic discrimination, ensure complete entity retrieval, and reduce false positives under uncertainty. Experiments show that PARCO achieves CER of 4.22% on Chinese AISHELL-1 and WER of 11.14% on English DATA2 under 1,000 distractors, significantly outperforming baselines. PARCO also demonstrates robust gains on out-of-domain datasets like THCHS-30 and LibriSpeech.

cs.CL↗