arXiv subjects

Zhiyuan Cheng

Publications and source records attributed to Zhiyuan Cheng.

At least 19 recordsLinked to original sources

arXiv2026

AutoNeural: Co-Designing Vision-Language Models for NPU Inference

While Neural Processing Units (NPUs) offer high theoretical efficiency for edge AI, state-of-the-art Vision--Language Models (VLMs) tailored for GPUs often falter on these substrates. We attribute this hardware-model mismatch to two primary factors: the quantization brittleness of Vision Transformers (ViTs) and the I/O-bound nature of autoregressive attention mechanisms, which fail to utilize the high arithmetic throughput of NPUs. To bridge this gap, we propose AutoNeural, an NPU-native VLM architecture co-designed for integer-only inference. We replace the standard ViT encoder with a MobileNetV5-style backbone utilizing depthwise separable convolutions, which ensures bounded activation distributions for stable INT4/8/16 quantization. Complementing this, our language backbone integrates State-Space Model (SSM) principles with Transformer layers, employing efficient gated convolutions to achieve linear-time complexity. This hybrid design eliminates the heavy memory I/O overhead of Key-Value caching during generation. Our approach delivers substantial efficiency gains, reducing quantization error of vision encoder by up to 7x and end-to-end latency by 14x compared to conventional baselines. The AutoNeural also delivers 3x decoding speed and 4x longer context window than the baseline. We validate these improvements via a real-world automotive case study on the Qualcomm SA8295P SoC, demonstrating real-time performance for cockpit applications. Our results highlight that rethinking model topology specifically for NPU constraints is a prerequisite for robust multi-modal edge intelligence.

cs.CL↗

arXiv2026

Sustainable Hybrid Document-Routed Retrieval for Financial RAG: Resolving the Robustness-Precision Trade-off

Retrieval-Augmented Generation (RAG) systems for financial document QA typically follow a chunk-based paradigm: documents are split into fragments, embedded, and retrieved by similarity. In structurally homogeneous corpora such as regulatory filings, this suffers from cross-document chunk confusion. Semantic File Routing (SFR), which uses LLM structured output to route queries to whole documents, reduces catastrophic failures but sacrifices targeted-chunk precision. We identify this robustness-precision trade-off on the FinDER benchmark (1,500 queries across five groups): SFR achieves higher average scores (6.45 vs. 6.02) and fewer failures (10.3% vs. 22.5%), while chunk-based retrieval (CBR) yields more perfect answers (13.8% vs. 8.5%). To resolve it, we propose Hybrid Document-Routed Retrieval (HDRR), a two-stage architecture that uses SFR as a document filter followed by chunk retrieval scoped to the identified document(s), eliminating cross-document confusion while preserving chunk precision. HDRR achieves the best performance on every metric: an average score of 7.54 (25.2% above CBR, 16.9% above SFR), a 6.4% failure rate, 67.7% correctness (+18.7 pp over CBR), and a 20.1% perfect-answer rate (+6.3 pp over CBR, +11.6 pp over SFR), simultaneously attaining the lowest failure rate and highest precision across all five groups. Beyond accuracy, HDRR is also the most efficient of the high-quality systems: it preserves CBR's compact per-query token budget (~5K-15K, an order of magnitude below SFR's ~50K-200K), incurs no indexing-time LLM spend (versus the one-time ~$100 cost of contextual indexing), and uses fewer per-query LLM calls than self-correcting agentic baselines, translating directly to lower API spend and inference-time energy at deployment scale.

cs.CL↗

arXiv2026

iFLYTEK-Embodied-Omni Technical Report

General-purpose embodied agents must understand multimodal instructions, anticipate how their environment will evolve, and produce precise control actions over extended horizons. Existing approaches typically specialize in visual-language reasoning, video-based world modeling, or action generation, while cascaded pipelines that first synthesize future observations and then infer actions can introduce interface bottlenecks and compound prediction errors. We present iFLYTEK-Embodied-Omni, a unified multimodal foundation model that jointly models vision(videos and images), language, and action within a single Omni framework. Its modality-specific visual-language, video-generation, and action-generation components communicate through shared multimodal self-attention. This design establishes brain-cerebellum collaboration: the vision-language modeland video generation model form a high-level brain for instruction understanding, task planning, progress tracking, and future visual-state prediction, whereas the action generation modelserves as a low-level cerebellum that directly converts planned subgoals and shared multimodal context into executable action chunks. To develop these capabilities, we combine action-annotated and action-free embodied videos from human demonstrations and robot interactions with embodied reasoning, embodied perception, and general-purpose image-text data to construct a comprehensive dataset. We further adopt a four-stage strategy that progressively trains the VLM, VGM, and AGM before jointly fine-tuning the complete model.

cs.AI↗

arXiv2026

GEAR-VLA: Learning Geometry-Aware Action Representations for Generalizable Robotic Manipulation

Vision-Language-Action (VLA) models achieve strong benchmark performance but still struggle in real-world deployment with unseen objects, background shifts, and different robot embodiments. We argue that this stems from the lack of a unified geometry-aware manipulation representation, leaving existing VLAs vulnerable to low-level trajectory supervision, misaligned 3D features, and embodiment differences. To address this, we propose GEAR-VLA, a VLA framework for learning unified geometry-aware action representations for generalizable robotic manipulation. GEAR-VLA adopts coarse-to-fine action learning, where multi-source embodied pretraining equips the VLM with embodied reasoning and discrete action understanding before latent action tokens connect action semantics to a gradient-decoupled DiT continuous action expert. It further performs semantic-aligned 3D integration by aligning a trainable 3D spatial backbone with the VLA representation while freezing the original VLM-aligned visual pathway. To share this representation across robots, GEAR-VLA uses embodiment canonicalization, where embodiment-aware states and embodiment-invariant actions confine robot differences to the low-level interface. Extensive simulation and real-world experiments demonstrate strong generalization: GEAR-VLA achieves state-of-the-art performance on LIBERO, zero-shot LIBERO-Plus, and RoboTwin 2.0, reaches 85.9% success on AgileX and 81.0% on the pretraining-unseen LDT-01 embodiment, and obtains 90.1% success on a 6,360-trial universal grasping benchmark with 212 unseen objects. Code and models will be released at https://github.com/babynabeauty/GEAR-VLA.

cs.RO↗

arXiv2026

Energy-Efficient On-Device RAG on a Mobile NPU: System Design and Benchmark on Snapdragon X Elite

Retrieval-Augmented Generation (RAG) pipelines are compute-intensive, combining embedding, retrieval, reranking, and large language model (LLM) generation. Running them entirely on-device benefits privacy, latency, and offline use, but the energy cost of CPU inference is a major barrier. We present what is, to our knowledge, the first end-to-end RAG pipeline that runs all neural stages -- embedding, reranking, and LLM generation -- on the Qualcomm Hexagon NPU of the Snapdragon X Elite. Profiling on a Dell XPS 13 laptop, we compare NPU-accelerated RAG against CPU and OpenCL/Adreno GPU baselines on indexing and query workloads. On indexing, the NPU achieves 9.1x higher embedding throughput and 12.3x less system energy. On a 120-query Wikipedia-passage benchmark, it delivers 18.1x faster LLM prefilling, 4.0x lower end-to-end query latency, and 4.0x less system energy than the CPU baseline; the same workload on the integrated GPU is 1.7x slower than CPU and uses 6.5x more energy than the NPU. A GPT-4.1 LLM-as-judge evaluation finds NPU answer quality on par with CPU and GPU within evaluator noise (mean 9.32 vs. 8.95 vs. 9.03 on a 1-10 rubric), with 86.7% of queries scoring identically across all three backends. On the Snapdragon X Elite / Hexagon class of laptop SoC, the NPU thus enables practical, energy-efficient on-device RAG without quality regression -- a sustainable path toward green edge intelligence that we expect to generalize to comparable mobile NPUs (Apple Neural Engine, Intel NPU, MediaTek APU) as their software stacks mature.

cs.CL↗

arXiv2026

Enhancing Financial Report Question-Answering: A Retrieval-Augmented Generation System with Reranking Analysis

Financial analysts face significant challenges extracting information from lengthy 10-K reports, which often exceed 100 pages. This paper presents a Retrieval-Augmented Generation (RAG) system designed to answer questions about S&P 500 financial reports and evaluates the impact of neural reranking on system performance. Our pipeline employs hybrid search combining full-text and semantic retrieval, followed by an optional reranking stage using a cross-encoder model. We conduct systematic evaluation using the FinDER benchmark dataset, comprising 1,500 queries across five experimental groups. Results demonstrate that reranking significantly improves answer quality, achieving 49.0 percent correctness for scores of 8 or above compared to 33.5 percent without reranking, representing a 15.5 percentage point improvement. Additionally, the error rate for completely incorrect answers decreases from 35.3 percent to 22.5 percent. Our findings emphasize the critical role of reranking in financial RAG systems and demonstrate performance improvements over baseline methods through modern language models and refined retrieval strategies.

cs.CL↗

arXiv2026

Improving the Completeness and Comparability of Segment Disclosures: A Large Language Model Approach

Segment-level disclosures are a central component of financial reporting, providing insight into firms' internal organization and the allocation of economic activities across operating units. However, segment information is often presented in both qualitative and quantitative forms, dispersed across tables and narrative sections of Form 10-K filings. Empirical research relying on structured databases faces both completeness and comparability challenges, as some firm-year observations may be missing, nested segment disclosures are not captured, and support for longitudinal and cross-firm comparability is limited. This study develops a large language model-based framework to extract segment disclosures directly from Form 10-K filings and to preserve both reportable and nested segment information. We further design a retrieval augmented system that incorporates information across multiple filings to support comparability. We use two representative settings to demonstrate its application: longitudinal analysis within a firm to interpret segment changes over time, and cross firm alignment of geographic segments across firms with different reporting structures. The results indicate that the artifact accurately extracts segment-level information and effectively addresses questions that require cross-period knowledge, demonstrating the potential of LLM-based approaches to enhance the measurement and interpretation of segment disclosures.

cs.CL↗

arXiv2025

iFlyBot-VLM Technical Report

We introduce iFlyBot-VLM, a general-purpose Vision-Language Model (VLM) used to improve the domain of Embodied Intelligence. The central objective of iFlyBot-VLM is to bridge the cross-modal semantic gap between high-dimensional environmental perception and low-level robotic motion control. To this end, the model abstracts complex visual and spatial information into a body-agnostic and transferable Operational Language, thereby enabling seamless perception-action closed-loop coordination across diverse robotic platforms. The architecture of iFlyBot-VLM is systematically designed to realize four key functional capabilities essential for embodied intelligence: 1) Spatial Understanding and Metric Reasoning; 2) Interactive Target Grounding; 3) Action Abstraction and Control Parameter Generation; 4) Task Planning and Skill Sequencing. We envision iFlyBot-VLM as a scalable and generalizable foundation model for embodied AI, facilitating the progression from specialized task-oriented systems toward generalist, cognitively capable agents. We conducted evaluations on 10 current mainstream embodied intelligence-related VLM benchmark datasets, such as Blink and Where2Place, and achieved optimal performance while preserving the model's general capabilities. We will publicly release both the training data and model weights to foster further research and development in the field of Embodied Intelligence.

cs.RO↗

arXiv2025

EffiTune: Diagnosing and Mitigating Training Inefficiency for Parameter Tuner in Robot Navigation System

Robot navigation systems are critical for various real-world applications such as delivery services, hospital logistics, and warehouse management. Although classical navigation methods provide interpretability, they rely heavily on expert manual tuning, limiting their adaptability. Conversely, purely learning-based methods offer adaptability but often lead to instability and erratic robot behaviors. Recently introduced parameter tuners aim to balance these approaches by integrating data-driven adaptability into classical navigation frameworks. However, the parameter tuning process currently suffers from training inefficiencies and redundant sampling, with critical regions in environment often underrepresented in training data. In this paper, we propose EffiTune, a novel framework designed to diagnose and mitigate training inefficiency for parameter tuners in robot navigation systems. EffiTune first performs robot-behavior-guided diagnostics to pinpoint critical bottlenecks and underrepresented regions. It then employs a targeted up-sampling strategy to enrich the training dataset with critical samples, significantly reducing redundancy and enhancing training efficiency. Our comprehensive evaluation demonstrates that EffiTune achieves more than a 13.5% improvement in navigation performance, enhanced robustness in out-of-distribution scenarios, and a 4x improvement in training efficiency within the same computational budget.

cs.RO↗

arXiv2025

CloneShield: A Framework for Universal Perturbation Against Zero-Shot Voice Cloning

Recent breakthroughs in text-to-speech (TTS) voice cloning have raised serious privacy concerns, allowing highly accurate vocal identity replication from just a few seconds of reference audio, while retaining the speaker's vocal authenticity. In this paper, we introduce CloneShield, a universal time-domain adversarial perturbation framework specifically designed to defend against zero-shot voice cloning. Our method provides protection that is robust across speakers and utterances, without requiring any prior knowledge of the synthesized text. We formulate perturbation generation as a multi-objective optimization problem, and propose Multi-Gradient Descent Algorithm (MGDA) to ensure the robust protection across diverse utterances. To preserve natural auditory perception for users, we decompose the adversarial perturbation via Mel-spectrogram representations and fine-tune it for each sample. This design ensures imperceptibility while maintaining strong degradation effects on zero-shot cloned outputs. Experiments on three state-of-the-art zero-shot TTS systems, five benchmark datasets and evaluations from 60 human listeners demonstrate that our method preserves near-original audio quality in protected inputs (PESQ = 3.90, SRS = 0.93) while substantially degrading both speaker similarity and speech quality in cloned samples (PESQ = 1.07, SRS = 0.08).

cs.SD↗

arXiv2024

Roadmap on Quantum Magnetic Materials

Fundamental research on two-dimensional (2D) magnetic systems based on van der Waals materials has been gaining traction rapidly since their recent discovery. With the increase of recent knowledge, it has become clear that such materials have also a strong potential for applications in devices that combine magnetism with electronics, optics, and nanomechanics. Nonetheless, many challenges still lay ahead. Several fundamental aspects of 2D magnetic materials are still unknown or poorly understood, such as their often-complicated electronic structure, optical properties, and magnetization dynamics, and their magnon spectrum. To elucidate their properties and facilitate integration in devices, advanced characterization techniques and theoretical frameworks need to be developed or adapted. Moreover, developing synthesis methods which increase critical temperatures and achieve large-scale, high-quality homogeneous thin films is crucial before these materials can be used for real-world applications. Therefore, the field of 2D magnetic materials provides many challenges and opportunities for the discovery and exploration of new phenomena, as well as the development of new applications. This Roadmap presents the background, challenges, and potential research directions for various relevant topics in the field on the fundamentals, synthesis, characterization, and applications. We hope that this work can provide a strong starting point for young researchers in the field and provide a general overview of the key challenges for more experienced researchers.

cond-mat.mtrl-sci↗

arXiv2024

GPTuner: A Manual-Reading Database Tuning System via GPT-Guided Bayesian Optimization

Modern database management systems (DBMS) expose hundreds of configurable knobs to control system behaviours. Determining the appropriate values for these knobs to improve DBMS performance is a long-standing problem in the database community. As there is an increasing number of knobs to tune and each knob could be in continuous or categorical values, manual tuning becomes impractical. Recently, automatic tuning systems using machine learning methods have shown great potentials. However, existing approaches still incur significant tuning costs or only yield sub-optimal performance. This is because they either ignore the extensive domain knowledge available (e.g., DBMS manuals and forum discussions) and only rely on the runtime feedback of benchmark evaluations to guide the optimization, or they utilize the domain knowledge in a limited way. Hence, we propose GPTuner, a manual-reading database tuning system. Firstly, we develop a Large Language Model (LLM)-based pipeline to collect and refine heterogeneous knowledge, and propose a prompt ensemble algorithm to unify a structured view of the refined knowledge. Secondly, using the structured knowledge, we (1) design a workload-aware and training-free knob selection strategy, (2) develop a search space optimization technique considering the value range of each knob, and (3) propose a Coarse-to-Fine Bayesian Optimization Framework to explore the optimized space. Finally, we evaluate GPTuner under different benchmarks (TPC-C and TPC-H), metrics (throughput and latency) as well as DBMS (PostgreSQL and MySQL). Compared to the state-of-the-art approaches, GPTuner identifies better configurations in 16x less time on average. Moreover, GPTuner achieves up to 30% performance improvement (higher throughput or lower latency) over the best-performing alternative.

cs.DB↗

arXiv2024

Pressure-dependent magnetotransport measurement in Kagome metal Yb$_{0.5}$Co_3Ge$_3$

Kagome materials are known to be an ideal platform that hosts a plethora of interesting phases such as topological states, electronic correlation, and magnetism, owing to their unique band structure and geometry. We report magnetotransport measurement in Kagome metal Yb$_{0.5}$Co_3Ge$_3$ as a function of pressure. Below $\sim25^\circ$ K the temperature dependence of resistance shows an upturn that is accompanied by a strong negative magnetoresistance, which could be attributed to Kondo effect. Upon pressurization above 1 GPa the resistance shows a reduction as a function of temperature below $4^\circ$ K, along with a further enhanced negative magnetoresistance. This might indicate an onset of a pressure-induced Kondo coherence effect.

cond-mat.str-el↗

arXiv2024

Attack as Defense: Run-time Backdoor Implantation for Image Content Protection

As generative models achieve great success, tampering and modifying the sensitive image contents (i.e., human faces, artist signatures, commercial logos, etc.) have induced a significant threat with social impact. The backdoor attack is a method that implants vulnerabilities in a target model, which can be activated through a trigger. In this work, we innovatively prevent the abuse of image content modification by implanting the backdoor into image-editing models. Once the protected sensitive content on an image is modified by an editing model, the backdoor will be triggered, making the editing fail. Unlike traditional backdoor attacks that use data poisoning, to enable protection on individual images and eliminate the need for model training, we developed the first framework for run-time backdoor implantation, which is both time- and resource- efficient. We generate imperceptible perturbations on the images to inject the backdoor and define the protected area as the only backdoor trigger. Editing other unprotected insensitive areas will not trigger the backdoor, which minimizes the negative impact on legal image modifications. Evaluations with state-of-the-art image editing models show that our protective method can increase the CLIP-FID of generated images from 12.72 to 39.91, or reduce the SSIM from 0.503 to 0.167 when subjected to malicious editing. At the same time, our method exhibits minimal impact on benign editing, which demonstrates the efficacy of our proposed framework. The proposed run-time backdoor can also achieve effective protection on the latest diffusion models. Code are available.

cs.CR↗

arXiv2024

ROCAS: Root Cause Analysis of Autonomous Driving Accidents via Cyber-Physical Co-mutation

As Autonomous driving systems (ADS) have transformed our daily life, safety of ADS is of growing significance. While various testing approaches have emerged to enhance the ADS reliability, a crucial gap remains in understanding the accidents causes. Such post-accident analysis is paramount and beneficial for enhancing ADS safety and reliability. Existing cyber-physical system (CPS) root cause analysis techniques are mainly designed for drones and cannot handle the unique challenges introduced by more complex physical environments and deep learning models deployed in ADS. In this paper, we address the gap by offering a formal definition of ADS root cause analysis problem and introducing ROCAS, a novel ADS root cause analysis framework featuring cyber-physical co-mutation. Our technique uniquely leverages both physical and cyber mutation that can precisely identify the accident-trigger entity and pinpoint the misconfiguration of the target ADS responsible for an accident. We further design a differential analysis to identify the responsible module to reduce search space for the misconfiguration. We study 12 categories of ADS accidents and demonstrate the effectiveness and efficiency of ROCAS in narrowing down search space and pinpointing the misconfiguration. We also show detailed case studies on how the identified misconfiguration helps understand rationale behind accidents.

cs.SE↗

arXiv2024

MixLoRA: Enhancing Large Language Models Fine-Tuning with LoRA-based Mixture of Experts

Fine-tuning Large Language Models (LLMs) is a common practice to adapt pre-trained models for specific applications. While methods like LoRA have effectively addressed GPU memory constraints during fine-tuning, their performance often falls short, especially in multi-task scenarios. In contrast, Mixture-of-Expert (MoE) models, such as Mixtral 8x7B, demonstrate remarkable performance in multi-task learning scenarios while maintaining a reduced parameter count. However, the resource requirements of these MoEs remain challenging, particularly for consumer-grade GPUs with less than 24GB memory. To tackle these challenges, we propose MixLoRA, an approach to construct a resource-efficient sparse MoE model based on LoRA. MixLoRA inserts multiple LoRA-based experts within the feed-forward network block of a frozen pre-trained dense model and employs a commonly used top-k router. Unlike other LoRA-based MoE methods, MixLoRA enhances model performance by utilizing independent attention-layer LoRA adapters. Additionally, an auxiliary load balance loss is employed to address the imbalance problem of the router. Our evaluations show that MixLoRA improves about 9% accuracy compared to state-of-the-art PEFT methods in multi-task learning scenarios. We also propose a new high-throughput framework to alleviate the computation and memory bottlenecks during the training and inference of MOE models. This framework reduces GPU memory consumption by 40% and token computation latency by 30% during both training and inference.

cs.CL↗

arXiv2024

Self-supervised Adversarial Training of Monocular Depth Estimation against Physical-World Attacks

Monocular Depth Estimation (MDE) plays a vital role in applications such as autonomous driving. However, various attacks target MDE models, with physical attacks posing significant threats to system security. Traditional adversarial training methods, which require ground-truth labels, are not directly applicable to MDE models that lack ground-truth depth. Some self-supervised model hardening techniques (e.g., contrastive learning) overlook the domain knowledge of MDE, resulting in suboptimal performance. In this work, we introduce a novel self-supervised adversarial training approach for MDE models, leveraging view synthesis without the need for ground-truth depth. We enhance adversarial robustness against real-world attacks by incorporating L_0-norm-bounded perturbation during training. We evaluate our method against supervised learning-based and contrastive learning-based approaches specifically designed for MDE. Our experiments with two representative MDE networks demonstrate improved robustness against various adversarial attacks, with minimal impact on benign performance.

cs.CV↗

arXiv2024

BadPart: Unified Black-box Adversarial Patch Attacks against Pixel-wise Regression Tasks

Pixel-wise regression tasks (e.g., monocular depth estimation (MDE) and optical flow estimation (OFE)) have been widely involved in our daily life in applications like autonomous driving, augmented reality and video composition. Although certain applications are security-critical or bear societal significance, the adversarial robustness of such models are not sufficiently studied, especially in the black-box scenario. In this work, we introduce the first unified black-box adversarial patch attack framework against pixel-wise regression tasks, aiming to identify the vulnerabilities of these models under query-based black-box attacks. We propose a novel square-based adversarial patch optimization framework and employ probabilistic square sampling and score-based gradient estimation techniques to generate the patch effectively and efficiently, overcoming the scalability problem of previous black-box patch attacks. Our attack prototype, named BadPart, is evaluated on both MDE and OFE tasks, utilizing a total of 7 models. BadPart surpasses 3 baseline methods in terms of both attack performance and efficiency. We also apply BadPart on the Google online service for portrait depth estimation, causing 43.5% relative distance error with 50K queries. State-of-the-art (SOTA) countermeasures cannot defend our attack effectively.

cs.CV↗