arXiv subjects

Hengyu Zhang

Publications and source records attributed to Hengyu Zhang.

At least 19 recordsLinked to original sources

arXiv2026

CoRe-Gen: Robust Spectrum-to-Structure Generation under Imperfect Fingerprint Conditions

Molecular structure elucidation from tandem mass spectra (MS/MS) remains challenging, particularly for de novo generation beyond database coverage. A common approach decomposes the task into spectrum-to-fingerprint prediction followed by fingerprint-to-structure decoding, enabling the use of large-scale molecular corpora. However, at deployment, the decoder relies on predicted rather than oracle fingerprints, introducing structured errors that propagate into generation. Moreover, common autoregressive decoders threshold fingerprint probabilities and serialize active bits as discrete input tokens, blocking molecular generation gradients from the spectrum encoder. We present \textit{CoRe-Gen}, which improves the intermediate condition through synthetic-spectrum pretraining, matches deployment-time noise through frequency-aware corruption, and introduces a differentiable fingerprint-to-memory bridge for joint encoder--decoder finetuning. The bridge preserves soft bit confidences and conditions a structure-aware autoregressive decoder without discrete fingerprint tokenization. CoRe-Gen achieves 21.49\%/35.42\% Top-1/Top-10 exact-match accuracy on NPLIB1 and, under encoder-aligned evaluation, 20.02\%/22.45\% on MassSpecGym, while retaining efficient autoregressive inference.

cs.LG↗

arXiv2026

Pistis Technical Report

We introduce the Pistis model family, comprising 27B- and 9B-parameter multimodal large language models built on Qwen3.6 and Qwen3.5, respectively, and developed through a general and scalable post-training framework. The framework first establishes a strong foundation through large-scale multimodal supervised fine-tuning (SFT). Building on this SFT foundation, we propose Interleaved Distillation and Reinforcement Learning (IDRL), a novel post-training paradigm that tightly integrates on-policy distillation and reinforcement learning within a single training loop. By alternating between the two objectives, rather than optimizing either in isolation or combining them in a static joint loss, IDRL enables more effective knowledge transfer, greater optimization stability, and more precise credit assignment for long-horizon agentic trajectories, leading to stronger performance while mitigating common capability trade-offs. At both model scales, the framework produces two specialized variants: Pistis-Thinking, designed to strengthen deep multimodal reasoning, and Pistis-Agentic, which additionally incorporates agentic trajectory data to support long-horizon planning, iterative reasoning, and tool use. Pistis-Agentic is particularly strong in multimodal search. Both scales outperform their corresponding base models. Beyond model-parameter optimization, we further introduce Pistis-Auto-Harnessing (PAH), a system-level method that automatically improves the agent's inference harness through iterative optimization. Experiments demonstrate that PAH enhances the model performance without updating the model parameters or increasing the interaction budget.

cs.AI↗

arXiv2026

ReportQA: QA-Based Radiology Report Evaluation

Radiology report evaluation is essential for advancing automated report generation. Natural language generation metrics have limited clinical relevance. Clinical efficacy (CE) metrics evaluate important medical findings, but focus mainly on presence and cover only a limited set of entities. Due to heavy reliance on manual annotations, it is difficult for CE metrics to extend clinical entities or attributes. In clinical practice, radiology reports serve as a medium for information transfer. Clinicians use them to perform downstream diagnostic tasks without directly inspecting images. Based on this insight, we propose ReportQA, a clinical-related and flexible radiology report evaluation framework, supporting detailed quantitative analysis of radiology report generation systems. We first collect datasets covering multiple imaging modalities and anatomical regions. We then construct knowledge trees of clinical entities and attributes with radiologist guidance, and use large language models (LLMs) to extract structured information from raw reports. Next, we generate QA pairs from predefined templates and apply quality control through self-filtering and report-based filtering. During evaluation, the report is treated as context, and an LLM acts as a judge model to answer the QA pairs. Based on the resulting QA accuracy, we introduce QAScore metric. Compared with existing metrics, QAScore shows better alignment with radiologist judgments. Experiments on multiple state-of-the-art vision-language models reveal that current report-based inference paradigms struggle to learn fine-grained clinical representations and exhibit strong negative prior biases. In contrast, question-driven inference provides a more effective alternative. For reproducibility and extensibility, we release the knowledge trees, structured reports, and QA pairs, along with the pipeline code for QA construction and evaluation.

cs.CL↗

arXiv2026

EviCare: Enhancing Diagnosis Prediction with Deep Model-Guided Evidence for In-Context Reasoning

Recent advances in large language models (LLMs) have enabled promising progress in diagnosis prediction from electronic health records (EHRs). However, existing LLM-based approaches tend to overfit to historically observed diagnoses, often overlooking novel yet clinically important conditions that are critical for early intervention. To address this, we propose EviCare, an in-context reasoning framework that integrates deep model guidance into LLM-based diagnosis prediction. Rather than prompting LLMs directly with raw EHR inputs, EviCare performs (1) deep model inference for candidate selection, (2) evidential prioritization for set-based EHRs, and (3) relational evidence construction for novel diagnosis prediction. These signals are then composed into an adaptive in-context prompt to guide LLM reasoning in an accurate and interpretable manner. Extensive experiments on two real-world EHR benchmarks (MIMIC-III and MIMIC-IV) demonstrate that EviCare achieves significant performance gains, which consistently outperforms both LLM-only and deep model-only baselines by an average of 20.65\% across precision and accuracy metrics. The improvements are particularly notable in challenging novel diagnosis prediction, yielding average improvements of 30.97\%.

cs.CL↗

arXiv2026

Doppler Shift Keying Modulation for Uplink Multiple Access over Doubly-Dispersive Channels

The delay-Doppler (DD) domain modulation has been regarded as one of the most competitive candidates to support wireless communications for emerging high-mobility applications in the sixth-generation mobile networks. Unfortunately, most of the existing designs for DD domain modulation suffer from high peak-to-average power ratio (PAPR) and unbearable detection complexity under uplink transmission since large time duration and bandwidth are required to guarantee high DD resolutions. To address these issues, the Doppler shift keying (DSK) modulation based on the orthogonal delay Doppler division multiplexing modulator is proposed in this paper, where the input-output characterization in the DD domain is fully exploited. The principle of the DSK transceiver is first established with the one-hot mapper and low-complexity iterative successive interference cancellation-maximum ratio combining detector for point-to-point scenarios. The proposed scheme is then generalized to the zero auto-correlation sequence-based implementation, which benefits the extension of multi-user (MU) uplink DSK frameworks. For uplink DSK transmission, Zadoff-Chu (ZC) sequences are adopted as the basis sequences. We optimize the assignment of ZC roots to different user equipments (UEs) by minimizing the maximum inter-user interference. This optimization process, which analyzes the root allocation, directly assigns a specific ZC sequence to each UE. The PAPR and bit error rate performance of the proposed DSK modulation with the low-complexity detector is finally verified by extensive simulation results under doubly-dispersive channels, which demonstrates the superiority of DSK modulation especially for uplink multiple access over doubly dispersive channels.

eess.SP↗

arXiv2025

RSMA-Assited and Transceiver-Coordinated ICI Management for MIMO-OFDM System

High-mobility scenarios are becoming increasingly critical in next-generation communication systems. While multiple-input multiple-output orthogonal frequency division multiplexing (MIMO-OFDM) stands as a prominent technology, its performance in such scenarios is fundamentally limited by Doppler-induced inter-carrier interference (ICI). Rate splitting multiple access (RSMA), recognized as a key multiple access technique for future communications, demonstrates superior interference management capabilities that we leverage to address this challenge. In specific, we propose a novel RSMA-assisted and transceiver-coordinated transmission scheme for ICI management in MIMO-OFDM system: (1) At the receiver side, we develop a hybrid successive interference cancellation (SIC) architecture with dynamic subcarrier clustering, which enables parallel intra-cluster and serial inter-cluster processing to balance complexity and performance. (2) At the transmitter~side, we design a matched hybrid precoding through formulated sum-rate maximization, solved via our proposed augmented boundary-compressed particle swarm optimization (ABC-PSO) algorithm for analog phase optimization and weighted minimum mean-square error (WMMSE)-based digital precoding iteration. Simulation results show that our scheme brings effective ICI suppression and enhanced system capacity with controlled complexity.

eess.SP↗

arXiv2025

From User Interface to Agent Interface: Efficiency Optimization of UI Representations for LLM Agents

While Large Language Model (LLM) agents show great potential for automated UI navigation such as automated UI testing and AI assistants, their efficiency has been largely overlooked. Our motivating study reveals that inefficient UI representation creates a critical performance bottleneck. However, UI representation optimization, formulated as the task of automatically generating programs that transform UI representations, faces two unique challenges. First, the lack of Boolean oracles, which traditional program synthesis uses to decisively validate semantic correctness, poses a fundamental challenge to co-optimization of token efficiency and completeness. Second, the need to process large, complex UI trees as input while generating long, compositional transformation programs, making the search space vast and error-prone. Toward addressing the preceding limitations, we present UIFormer, the first automated optimization framework that synthesizes UI transformation programs by conducting constraint-based optimization with structured decomposition of the complex synthesis task. First, UIFormer restricts the program space using a domain-specific language (DSL) that captures UI-specific operations. Second, UIFormer conducts LLM-based iterative refinement with correctness and efficiency rewards, providing guidance for achieving the efficiency-completeness co-optimization. UIFormer operates as a lightweight plugin that applies transformation programs for seamless integration with existing LLM agents, requiring minimal modifications to their core logic. Evaluations across three UI navigation benchmarks spanning Android and Web platforms with five LLMs demonstrate that UIFormer achieves 48.7% to 55.8% token reduction with minimal runtime overhead while maintaining or improving agent performance. Real-world industry deployment at WeChat further validates the practical impact of UIFormer.

cs.SE↗

arXiv2025

Moyun: A Diffusion-Based Model for Style-Specific Chinese Calligraphy Generation

Although Chinese calligraphy generation has achieved style transfer, generating calligraphy by specifying the calligrapher, font, and character style remains challenging. To address this, we propose a new Chinese calligraphy generation model 'Moyun' , which replaces the Unet in the Diffusion model with Vision Mamba and introduces the TripleLabel control mechanism to achieve controllable calligraphy generation. The model was tested on our large-scale dataset 'Mobao' of over 1.9 million images, and the results demonstrate that 'Moyun' can effectively control the generation process and produce calligraphy in the specified style. Even for calligraphy the calligrapher has not written, 'Moyun' can generate calligraphy that matches the style of the calligrapher.

cs.CV↗

arXiv2025

Leveraging Channel Knowledge Map for Multi-User Hierarchical Beam Training Under Position Uncertainty

Channel knowledge map (CKM) emerges as a promising framework to acquire location-specific channel information without consuming wireless resources, creating new horizons for advanced wireless network design and optimization. Despite its potential, the practical application of CKM in beam training faces several challenges. On one hand, the user's precise location is typically unavailable prior to beam training, which limits the utility of CKM since its effectiveness relies heavily on accurate input of position data. On the other hand, the intricate interplay among CKM, real-time observations, and training strategies has not been thoroughly studied, leading to suboptimal performance and difficulties in practical implementation. In this paper, we present a framework for CKM-aided beam training that addresses these limitations. For single-user scenario, we propose a reward-motivated beam-potential hierarchical strategy which integrates partial position information and CKM. This strategy models the user equipment (UE) position uncertainty and formulates the hierarchical searching process as a pruned binary search tree. An optimal hierarchical searching strategy with minimal overhead is derived by evaluating the weights and rewards of potential codewords. Furthermore, a low-complexity two-layer lookahead scheme is designed to balance overhead and computational demands. For multi-user scenario, we develop a correlation-driven position-pruning training scheme, where sidelobe gains from inter-user interference are exploited to provide additional side information for overhead reduction, allowing all users to be simultaneously assigned their respective supportive beams. Simulations validate the superior performances of proposed approaches in advancing 6G beam training.

cs.IT↗

arXiv2025

BeamCKM: A Framework of Channel Knowledge Map Construction for Multi-Antenna Systems

The channel knowledge map (CKM) enables efficient construction of high-fidelity mapping between spatial environments and channel parameters via electromagnetic information analysis. Nevertheless, existing studies are largely confined to single-antenna systems, failing to offer dedicated guidance for multi-antenna communication scenarios. To address the inherent conflict between traditional real-value pathloss map and multi-degree-of-freedom (DoF) coherent beamforming in B5G/6G systems, this paper proposes a novel concept of BeamCKM and CKMTransUNet architecture. The CKMTransUNet approach combines a UNet backbone for multi-scale feature extraction with a vision transformer (ViT) module to capture global dependencies among encoded linear vectors, utilizing a composite loss function to characterize the beam propagation characteristics. Furthermore, based on the CKMTransUNet backbone, this paper presents a methodology named M3ChanNet. It leverages the multi-modal learning technique and cross-attention mechanisms to extract intrinsic side information from environmental profiles and real-time multi-beam observations, thereby further improving the map construction accuracy. Simulation results demonstrate that the proposed method consistently outperforms state-of-the-art (SOTA) interpolation methods and deep learning (DL) approaches, delivering superior performance even when environmental contours are inaccurate. For reproducibility, the code is publicly accessible at https://github.com/github-whh/BeamCKM.

eess.SP↗

arXiv2025

Hierarchical Graph Information Bottleneck for Multi-Behavior Recommendation

In real-world recommendation scenarios, users typically engage with platforms through multiple types of behavioral interactions. Multi-behavior recommendation algorithms aim to leverage various auxiliary user behaviors to enhance prediction for target behaviors of primary interest (e.g., buy), thereby overcoming performance limitations caused by data sparsity in target behavior records. Current state-of-the-art approaches typically employ hierarchical design following either cascading (e.g., view$\rightarrow$cart$\rightarrow$buy) or parallel (unified$\rightarrow$behavior$\rightarrow$specific components) paradigms, to capture behavioral relationships. However, these methods still face two critical challenges: (1) severe distribution disparities across behaviors, and (2) negative transfer effects caused by noise in auxiliary behaviors. In this paper, we propose a novel model-agnostic Hierarchical Graph Information Bottleneck (HGIB) framework for multi-behavior recommendation to effectively address these challenges. Following information bottleneck principles, our framework optimizes the learning of compact yet sufficient representations that preserve essential information for target behavior prediction while eliminating task-irrelevant redundancies. To further mitigate interaction noise, we introduce a Graph Refinement Encoder (GRE) that dynamically prunes redundant edges through learnable edge dropout mechanisms. We conduct comprehensive experiments on three real-world public datasets, which demonstrate the superior effectiveness of our framework. Beyond these widely used datasets in the academic community, we further expand our evaluation on several real industrial scenarios and conduct an online A/B testing, showing again a significant improvement in multi-behavior recommendations. The source code of our proposed HGIB is available at https://github.com/zhy99426/HGIB.

cs.IR↗

arXiv2025

Adaptive Graph Integration for Cross-Domain Recommendation via Heterogeneous Graph Coordinators

In the digital era, users typically interact with diverse items across multiple domains (e.g., e-commerce, streaming platforms, and social networks), generating intricate heterogeneous interaction graphs. Leveraging multi-domain data can improve recommendation systems by enriching user insights and mitigating data sparsity in individual domains. However, integrating such multi-domain knowledge for cross-domain recommendation remains challenging due to inherent disparities in user behavior and item characteristics and the risk of negative transfer, where irrelevant or conflicting information from the source domains adversely impacts the target domain's performance. To tackle these challenges, we propose HAGO, a novel framework with \textbf{H}eterogeneous \textbf{A}daptive \textbf{G}raph co\textbf{O}rdinators, which dynamically integrates multi-domain graphs into a cohesive structure. HAGO adaptively adjusts the connections between coordinators and multi-domain graph nodes to enhance beneficial inter-domain interactions while alleviating negative transfer. Furthermore, we introduce a universal multi-domain graph pre-training strategy alongside HAGO to collaboratively learn high-quality node representations across domains. Being compatible with various graph-based models and pre-training techniques, HAGO demonstrates broad applicability and effectiveness. Extensive experiments show that our framework outperforms state-of-the-art methods in cross-domain recommendation scenarios, underscoring its potential for real-world applications. The source code is available at https://github.com/zhy99426/HAGO.

cs.IR↗

arXiv2025

Data-Driven Modulation Optimization with LMMSE Equalization for Reliability Enhancement in Underwater Acoustic Communications

Ultra-reliable underwater acoustic (UWA) communications serve as one of the key enabling technologies for future space-air-ground-underwater integrated networks. However, the reliability of current UWA transmission is still insufficient since severe performance degradation occurs for conventional multicarrier systems in UWA channels with severe delay-scale spread. To solve this problem, we exploit learning-inspired approaches to optimize the modulation scheme under the assumption of linear minimum mean square error (LMMSE) equalization, where the discrete representation of waveforms is adopted by utilizing Nyquist filters. The optimization problem is first transferred into maximizing the fairness of estimation mean square error (MSE) for each data symbol since the total MSE is invariant considering the property of orthogonal modulation. The Siamese architecture is then adopted to obtain consistent optimization results across various channel conditions, which avoids the overhead of online feedback, cooperation, and deployment of neural networks and guarantees generalization. The overall scheme including the loss function, neural network structure, and training process is also investigated in depth in this paper. The excellent performance and robustness of the proposed modulation scheme are verified by carrying out the bit error rate test over various UWA channels with severe delay-scale spread.

eess.SP↗

arXiv2025

Beamforming-Codebook-Aware Channel Knowledge Map Construction for Multi-Antenna Systems

Channel knowledge map (CKM) has emerged as a crucial technology for next-generation communication, enabling the construction of high-fidelity mappings between spatial environments and channel parameters via electromagnetic information analysis. Traditional CKM construction methods like ray tracing are computationally intensive. Recent studies utilizing neural networks (NNs) have achieved efficient CKM generation with reduced computational complexity and real-time processing capabilities. Nevertheless, existing research predominantly focuses on single-antenna systems, failing to address the beamforming requirements inherent to MIMO configurations. Given that appropriate precoding vector selection in MIMO systems can substantially enhance user communication rates, this paper presents a TransUNet-based framework for constructing CKM, which effectively incorporates discrete Fourier transform (DFT) precoding vectors. The proposed architecture combines a UNet backbone for multiscale feature extraction with a Transformer module to capture global dependencies among encoded linear vectors. Experimental results demonstrate that the proposed method outperforms state-of-the-art (SOTA) deep learning (DL) approaches, yielding a 17\% improvement in RMSE compared to RadioWNet. The code is publicly accessible at https://github.com/github-whh/TransUNet.

eess.SP↗

arXiv2024

Universal Modem Generation with Inherent Adaptability to Variant Underwater Acoustic Channels: a Data-Driven Perspective

In underwater acoustic (UWA) communication, orthogonal frequency division multiplexing (OFDM) is commonly employed to mitigate the inter-symbol interference (ISI) caused by delay spread. However, path-specific Doppler effects in UWA channels could result in significant inter-carrier interference (ICI) in the OFDM system. To address this problem, we introduce a multi-resolution convolutional neural network (CNN) named UWAModNet in this paper, designed to optimize the modem structure, specifically modulation and demodulation matrices. Based on a trade-off between the minimum and the average equivalent sub-channel rate, we propose an optimization criterion suitable to evaluate the performance of our learned modem. Additionally, a two-stage training strategy is developed to achieve quasi-optimal results. Simulations indicate that the learned modem outperforms zero-padded OFDM (ZP-OFDM) in terms of equivalent sub-channel rate and bit error rate, even under more severe Doppler effects during testing compared to training.

eess.SP↗

arXiv2024

SinkLoRA: Enhanced Efficiency and Chat Capabilities for Long-Context Large Language Models

Extending the functionality of the Transformer model to accommodate longer sequence lengths has become a critical challenge. This extension is crucial not only for improving tasks such as language translation and long-context processing but also for enabling novel applications like chatbots, code generation, and multimedia content creation. The primary obstacle is the self-attention mechanism, which scales quadratically with sequence length in terms of computation time and memory requirements. LongLoRA proposed shifted sparse attention (S$^2$-Attn), effectively enabling context extension and leading to non-trivial computation savings with similar performance to fine-tuning with vanilla attention. However, LongLoRA is still not as efficient as vanilla attention, reaching only 39\% of the perplexity improvement compared to full attention. This inefficiency is due to the cyclic shift applied within different attention head patterns, causing either chaos in the attention head structure or unnecessary information exchange between token groups. To address these issues, We propose \textbf{SinkLoRA}, which features better work partitioning. Specifically, (1) we developed SF-Attn with a segmentation and reassembly algorithm to proportionally return cyclically shifted groups of attention heads to their un-shifted state together with global attention of "sink attention tokens", achieving 92\% of the perplexity improvement compared to full attention after fine tuning, and (2) applied a SOTA KV cache compression algorithm H$_2$O to accelerate inference. Furthermore, We conducted supervised fine-tuning with SinkLoRA using a self collected LongAlpaca-plus dataset. All our code, models, datasets, and demos are available at \url{https://github.com/Dexter-GT-86/SinkLoRA}.

cs.CL↗

arXiv2024

Deep Pattern Network for Click-Through Rate Prediction

Click-through rate (CTR) prediction tasks play a pivotal role in real-world applications, particularly in recommendation systems and online advertising. A significant research branch in this domain focuses on user behavior modeling. Current research predominantly centers on modeling co-occurrence relationships between the target item and items previously interacted with by users in their historical data. However, this focus neglects the intricate modeling of user behavior patterns. In reality, the abundance of user interaction records encompasses diverse behavior patterns, indicative of a spectrum of habitual paradigms. These patterns harbor substantial potential to significantly enhance CTR prediction performance. To harness the informational potential within user behavior patterns, we extend Target Attention (TA) to Target Pattern Attention (TPA) to model pattern-level dependencies. Furthermore, three critical challenges demand attention: the inclusion of unrelated items within behavior patterns, data sparsity in behavior patterns, and computational complexity arising from numerous patterns. To address these challenges, we introduce the Deep Pattern Network (DPN), designed to comprehensively leverage information from user behavior patterns. DPN efficiently retrieves target-related user behavior patterns using a target-aware attention mechanism. Additionally, it contributes to refining user behavior patterns through a pre-training paradigm based on self-supervised learning while promoting dependency learning within sparse patterns. Our comprehensive experiments, conducted across three public datasets, substantiate the superior performance and broad compatibility of DPN.

cs.IR↗

arXiv2024

Modem Optimization of High-Mobility Scenarios: A Deep-Learning-Inspired Approach

The next generation wireless communication networks are required to support high-mobility scenarios, such as reliable data transmission for high-speed railways. Nevertheless, widely utilized multi-carrier modulation, the orthogonal frequency division multiplex (OFDM), cannot deal with the severe Doppler spread brought by high mobility. To address this problem, some new modulation schemes, e.g. orthogonal time frequency space and affine frequency division multiplexing, have been proposed with different design criteria from OFDM, which promote reliability with the cost of extremely high implementation complexity. On the other hand, end-to-end systems achieve excellent gains by exploiting neural networks to replace traditional transmitters and receivers, but have to retrain and update continually with channel varying. In this paper, we propose the Modem Network (ModNet) to design a novel modem scheme. Compared with end-to-end systems, channels are directly fed into the network and we can directly get a modem scheme through ModNet. Then, the Tri-Phase training strategy is proposed, which mainly utilizes the siamese structure to unify the learned modem scheme without retraining frequently faced up with time-varying channels. Simulation results show the proposed modem scheme outperforms OFDM systems under different highmobility channel statistics.

eess.SP↗