arXiv subjects

Yuan Feng

Publications and source records attributed to Yuan Feng.

At least 37 records · Page 2Linked to original sources

arXiv2026

Milestones over Outcome: Unlocking Geometric Reasoning with Sub-Goal Verifiable Reward

Multimodal Large Language Models (MLLMs) struggle with complex geometric reasoning, largely because "black box" outcome-based supervision fails to distinguish between lucky guesses and rigorous deduction. To address this, we introduce a paradigm shift towards subgoal-level evaluation and learning. We first construct GeoGoal, a benchmark synthesized via a rigorous formal verification data engine, which converts abstract proofs into verifiable numeric subgoals. This structure reveals a critical divergence between reasoning quality and outcome accuracy. Leveraging this, we propose the Sub-Goal Verifiable Reward (SGVR) framework, which replaces sparse signals with dense rewards based on the Skeleton Rate. Experiments demonstrate that SGVR not only enhances geometric performance (+9.7%) but also exhibits strong generalization, transferring gains to general math (+8.0%) and other general reasoning tasks (+2.8%), demonstrating broad applicability across diverse domains.

cs.LG↗

arXiv2025

GeoBench: Rethinking Multimodal Geometric Problem-Solving via Hierarchical Evaluation

Geometric problem solving constitutes a critical branch of mathematical reasoning, requiring precise analysis of shapes and spatial relationships. Current evaluations of geometric reasoning in vision-language models (VLMs) face limitations, including the risk of test data contamination from textbook-based benchmarks, overemphasis on final answers over reasoning processes, and insufficient diagnostic granularity. To address these issues, we present GeoBench, a hierarchical benchmark featuring four reasoning levels in geometric problem-solving: Visual Perception, Goal-Oriented Planning, Rigorous Theorem Application, and Self-Reflective Backtracking. Through six formally verified tasks generated via TrustGeoGen, we systematically assess capabilities ranging from attribute extraction to logical error correction. Experiments reveal that while reasoning models like OpenAI-o3 outperform general MLLMs, performance declines significantly with increasing task complexity. Key findings demonstrate that sub-goal decomposition and irrelevant premise filtering critically influence final problem-solving accuracy, whereas Chain-of-Thought prompting unexpectedly degrades performance in some tasks. These findings establish GeoBench as a comprehensive benchmark while offering actionable guidelines for developing geometric problem-solving systems.

cs.CV↗

arXiv2025

Fine-grained Preference Optimization Improves Zero-shot Text-to-Speech

Integrating human feedback to align text-to-speech (TTS) system outputs with human preferences has proven to be an effective approach for enhancing the robustness of language model-based TTS systems. Current approaches primarily focus on using preference data annotated at the utterance level. However, frequent issues that affect the listening experience often only arise in specific segments of audio samples, while other segments are well-generated. In this study, we propose a fine-grained preference optimization approach (FPO) to enhance the robustness of TTS systems. FPO focuses on addressing localized issues in generated samples rather than uniformly optimizing the entire utterance. Specifically, we first analyze the types of issues in generated samples, categorize them into two groups, and propose a selective training loss strategy to optimize preferences based on fine-grained labels for each issue type. Experimental results show that FPO enhances the robustness of zero-shot TTS systems by effectively addressing local issues, significantly reducing the bad case ratio, and improving intelligibility. Furthermore, FPO exhibits superior data efficiency compared with baseline systems, achieving similar performance with fewer training samples.

eess.AS↗

arXiv2025

Quantum Design Automation: Foundations, Challenges, and the Road Ahead

Quantum computing is transitioning from laboratory research to industrial deployment, yet significant challenges persist: system scalability and performance, fabrication yields, and the advancement of algorithms and applications. We emphasize that in building quantum computers -- spanning quantum chips, system integration, instruction sets, algorithms, and middleware such as quantum error correction schemes -- design is everywhere. In this paper, we advocate for a holistic design perspective in quantum computing, a perspective we argue is pivotal to unlocking innovative co-design opportunities and addressing the aforementioned key challenges. To equip readers with sufficient background for exploring co-optimization opportunities, we detail how interconnected computational methods and tools collaborate to enable end-to-end quantum computer design. This coverage encompasses critical stages -- such as chip layout design automation, high-fidelity system-level simulation, Hamiltonian derivation for quantum system modeling, control pulse simulation, decoherence analysis, and physical verification and testing -- followed by quantum instruction set design. We then proceed to quantum system and software development, including quantum circuit synthesis, quantum error correction and fault tolerance, and logic verification and testing. Through these discussions, we illustrate with concrete examples -- including co-optimizing quantum instruction sets with algorithmic considerations, customizing error correction circuits to hardware-specific constraints, and streamlining quantum chip design through tailored code design, among others. We hope that the detailed end-to-end design workflow as well as these examples will foster dialogue between the hardware and software communities, ultimately facilitating the translation of meaningful research findings into future quantum hardware implementations.

quant-ph↗

arXiv2025

AttnCache: Accelerating Self-Attention Inference for LLM Prefill via Attention Cache

Large Language Models (LLMs) are widely used in generative applications such as chatting, code generation, and reasoning. However, many realworld workloads such as classification, question answering, recommendation, and text embedding rely solely on the prefill stage of inference, where the model encodes input sequences without performing autoregressive decoding. In these prefill only scenarios, the self-attention computation becomes the primary performance bottleneck due to its quadratic complexity with respect to sequence length. In this paper, we observe that semantically different sentences often produce similar attention maps across layers and heads. Building on this insight, we propose AttnCache, a framework that accelerates the prefill stage of LLM inference by retrieving and reusing similar attention maps. Based on an attention map memorization database, AttnCache employs efficient caching and similarity search techniques to identify and reuse pre-cached attention maps during inference, thereby reducing the computational overhead of self-attention. Experimental results show that AttnCache achieves an average of 1.2x end-to-end and 2x attention speedup on CPU, and 1.6x end-to-end and 3x attention speedup on GPU, with negligible accuracy degradation.

cs.CL↗

arXiv2025

Ada-KV: Optimizing KV Cache Eviction by Adaptive Budget Allocation for Efficient LLM Inference

Large Language Models have excelled in various domains but face efficiency challenges due to the growing Key-Value (KV) cache required for long-sequence inference. Recent efforts aim to reduce KV cache size by evicting vast non-critical cache elements during runtime while preserving generation quality. However, these methods typically allocate compression budgets uniformly across all attention heads, ignoring the unique attention patterns of each head. In this paper, we establish a theoretical loss upper bound between pre- and post-eviction attention output, explaining the optimization target of prior cache eviction methods, while guiding the optimization of adaptive budget allocation. Base on this, we propose {\it Ada-KV}, the first head-wise adaptive budget allocation strategy. It offers plug-and-play benefits, enabling seamless integration with prior cache eviction methods. Extensive evaluations on 13 datasets from Ruler and 16 datasets from LongBench, all conducted under both question-aware and question-agnostic scenarios, demonstrate substantial quality improvements over existing methods. Our code is available at https://github.com/FFY0/AdaKV.

cs.CL↗

arXiv2025

Taming the Fragility of KV Cache Eviction in LLM Inference

Large language models have revolutionized natural language processing, yet their deployment remains hampered by the substantial memory and runtime overhead of the transformer's Key-Value cache. To mitigate this, recent methods employ a scoring-aggregation framework to evict unimportant cache entries, based on the stability assumption-that a fixed subset of entries remains consistently important during generation. However, prior work has largely focused on refining importance indicators for scoring, while defaulting to mean aggregation due to a faithful trust in the stability assumption. In this work, we argue that this underlying assumption is inherently fragile, making mean aggregation highly vulnerable in extreme cases. To counter this, we propose a simple yet elegant defensive aggregation strategy: a two-step, linear-time approach that controls worst-case risk, thereby defending against extreme cases with negligible computational overhead. Embodying this strategy, we propose a novel cache eviction method, DefensiveKV and its extension, Layer-DefensiveKV, which incorporates layer-wise budget allocation. Across seven task domains (18 datasets), our methods reduce generation quality loss by 2.3x and 4.3x respectively, versus the strongest baseline under a 20% cache size. These results set new performance benchmarks and pioneer a promising direction for optimizing cache eviction against underlying fragility through worst-case risk management. Our code is available at https://github.com/FFY0/DefensiveKV.

cs.CL↗

arXiv2025

SkewRoute: Training-Free LLM Routing for Knowledge Graph Retrieval-Augmented Generation via Score Skewness of Retrieved Context

Large language models excel at many tasks but often incur high inference costs during deployment. To mitigate hallucination, many systems use a knowledge graph to enhance retrieval-augmented generation (KG-RAG). However, the large amount of retrieved knowledge contexts increase these inference costs further. A promising solution to balance performance and cost is LLM routing, which directs simple queries to smaller LLMs and complex ones to larger LLMs. However, no dedicated routing methods currently exist for RAG, and existing training-based routers face challenges scaling to this domain due to the need for extensive training data. We observe that the score distributions produced by the retrieval scorer strongly correlate with query difficulty. Based on this, we propose an extremely simple yet effective routing framework, the first specifically designed for KG-RAG that efficiently balances performance and cost in a plug-and-play manner. It delivers over 3x higher routing effectiveness while reducing runtime to less than 0.001x compared to existing methods. Our code is available at https://github.com/hrwang00/SkewRoute.

cs.IR↗

arXiv2025

Audiobook-CC: Controllable Long-context Speech Generation for Multicast Audiobook

Existing text-to-speech systems predominantly focus on single-sentence synthesis and lack adequate contextual modeling as well as fine-grained performance control capabilities for generating coherent multicast audiobooks. To address these limitations, we propose a context-aware and emotion controllable speech synthesis framework specifically engineered for multicast audiobooks with three key innovations: a context mechanism for contextual consistency, a disentanglement paradigm to decouple style control from speech prompts for semantic consistency, and self-distillation to boost emotional expressiveness and instruction controllability. Experimental results show superior performance across the generation of narration, dialogue, and the whole chapter, significantly outperforming existing baselines. Ablation studies are conducted to validate the effectiveness of our proposed methods. Demo samples can be found in https://everest-ai.github.io/.

eess.AS↗

arXiv2025

Probing the Strong Gravity Region of Black Holes with eXTP

We present the novel capabilities of the enhanced X-ray Timing and Polarimetry (eXTP) mission to study the strong gravity region around stellar-mass black holes in X-ray binary systems and supermassive black holes in active galactic nuclei. eXTP can combine X-ray spectral, timing, and polarimetric techniques to study the accretion process near black holes, measure black hole masses and spins, and test Einstein's theory of General Relativity in the strong field regime. We show how eXTP can improve the current measurements of black holes of existing X-ray missions and we discuss the scientific questions that can be addressed.

astro-ph.HE↗

arXiv2025

Borrowing Dirty Qubits in Quantum Programs

Dirty qubits are ancillary qubits that can be borrowed from idle parts of a computation, enabling qubit reuse and reducing the demand for fresh, clean qubits-a resource that is typically scarce in practice. For such reuse to be valid, the initial states of the dirty qubits must not affect the functionality of the quantum circuits in which they are employed. Moreover, their original states, including any entanglement they possess, must be fully restored after use-a requirement commonly known as safe uncomputation. In this paper, we formally define the semantics of dirty-qubit borrowing as a feature in quantum programming languages, and introduce a notion of safe uncomputation for dirty qubits in quantum programs. We also present an efficient algorithm, along with experimental results, for verifying safe uncomputation of dirty qubits in certain quantum circuits.

quant-ph↗

arXiv2025

A Tensor Network based Decision Diagram for Representation of Quantum Circuits

Tensor networks have been successfully applied in simulation of quantum physical systems for decades. Recently, they have also been employed in classical simulation of quantum computing, in particular, random quantum circuits. This paper proposes a decision diagram style data structure, called TDD (Tensor Decision Diagram), for more principled and convenient applications of tensor networks. This new data structure provides a compact and canonical representation for quantum circuits. By exploiting circuit partition, the TDD of a quantum circuit can be computed efficiently. Furthermore, we show that the operations of tensor networks essential in their applications (e.g., addition and contraction), can also be implemented efficiently in TDDs. A proof-of-concept implementation of TDDs is presented and its efficiency is evaluated on a set of benchmark quantum circuits. It is expected that TDDs will play an important role in various design automation tasks related to quantum circuits, including but not limited to equivalence checking, error detection, synthesis, simulation, and verification.

quant-ph↗

arXiv2025

DESign: Dynamic Context-Aware Convolution and Efficient Subnet Regularization for Continuous Sign Language Recognition

Current continuous sign language recognition (CSLR) methods struggle with handling diverse samples. Although dynamic convolutions are ideal for this task, they mainly focus on spatial modeling and fail to capture the temporal dynamics and contextual dependencies. To address this, we propose DESign, a novel framework that incorporates Dynamic Context-Aware Convolution (DCAC) and Subnet Regularization Connectionist Temporal Classification (SR-CTC). DCAC dynamically captures the inter-frame motion cues that constitute signs and uniquely adapts convolutional weights in a fine-grained manner based on contextual information, enabling the model to better generalize across diverse signing behaviors and boost recognition accuracy. Furthermore, we observe that existing methods still rely on only a limited number of frames for parameter updates during training, indicating that CTC learning overfits to a dominant path. To address this, SR-CTC regularizes training by applying supervision to subnetworks, encouraging the model to explore diverse CTC alignment paths and effectively preventing overfitting. A classifier-sharing strategy in SR-CTC further strengthens multi-scale consistency. Notably, SR-CTC introduces no inference overhead and can be seamlessly integrated into existing CSLR models to boost performance. Extensive ablations and visualizations further validate the effectiveness of the proposed methods. Results on mainstream CSLR datasets (i.e., PHOENIX14, PHOENIX14-T, CSL-Daily) demonstrate that DESign achieves state-of-the-art performance.

cs.CV↗

arXiv2025

Serving Large Language Models on Huawei CloudMatrix384

The rapid evolution of large language models (LLMs), driven by growing parameter scales, adoption of mixture-of-experts (MoE) architectures, and expanding context lengths, imposes unprecedented demands on AI infrastructure. Traditional AI clusters face limitations in compute intensity, memory bandwidth, inter-chip communication, and latency, compounded by variable workloads and strict service-level objectives. Addressing these issues requires fundamentally redesigned hardware-software integration. This paper introduces Huawei CloudMatrix, a next-generation AI datacenter architecture, realized in the production-grade CloudMatrix384 supernode. It integrates 384 Ascend 910 NPUs and 192 Kunpeng CPUs interconnected via an ultra-high-bandwidth Unified Bus (UB) network, enabling direct all-to-all communication and dynamic pooling of resources. These features optimize performance for communication-intensive operations, such as large-scale MoE expert parallelism and distributed key-value cache access. To fully leverage CloudMatrix384, we propose CloudMatrix-Infer, an advanced LLM serving solution incorporating three core innovations: a peer-to-peer serving architecture that independently scales prefill, decode, and caching; a large-scale expert parallelism strategy supporting EP320 via efficient UB-based token dispatch; and hardware-aware optimizations including specialized operators, microbatch-based pipelining, and INT8 quantization. Evaluation with the DeepSeek-R1 model shows CloudMatrix-Infer achieves state-of-the-art efficiency: prefill throughput of 6,688 tokens/s per NPU and decode throughput of 1,943 tokens/s per NPU (<50 ms TPOT). It effectively balances throughput and latency, sustaining 538 tokens/s per NPU even under stringent 15 ms latency constraints, while INT8 quantization maintains model accuracy across benchmarks.

cs.DC↗

arXiv2025

Today's Cat Is Tomorrow's Dog: Accounting for Time-Based Changes in the Labels of ML Vulnerability Detection Approaches

Vulnerability datasets used for ML testing implicitly contain retrospective information. When tested on the field, one can only use the labels available at the time of training and testing (e.g. seen and assumed negatives). As vulnerabilities are discovered across calendar time, labels change and past performance is not necessarily aligned with future performance. Past works only considered the slices of the whole history (e.g. DiverseVUl) or individual differences between releases (e.g. Jimenez et al. ESEC/FSE 2019). Such approaches are either too optimistic in training (e.g. the whole history) or too conservative (e.g. consecutive releases). We propose a method to restructure a dataset into a series of datasets in which both training and testing labels change to account for the knowledge available at the time. If the model is actually learning, it should improve its performance over time as more data becomes available and data becomes more stable, an effect that can be checked with the Mann-Kendall test. We validate our methodology for vulnerability detection with 4 time-based datasets (3 projects from BigVul dataset + Vuldeepecker's NVD) and 5 ML models (Code2Vec, CodeBERT, LineVul, ReGVD, and Vuldeepecker). In contrast to the intuitive expectation (more retrospective information, better performance), the trend results show that performance changes inconsistently across the years, showing that most models are not learning.

cs.CR↗

arXiv2025

Reflection Spectra of Accretion Disks Illuminated by an Off-Axis Corona

Relativistic reflection features in the X-ray spectra of accreting black holes are considered to be generated by the illumination of the accretion disk by the hot corona. In this work, we present a numerical method for the emission line profile and the reflection spectrum produced by an off-axis X-ray source. The X-ray source is considered as a point source, as in the lamppost scenario, except that it is located off-axis and moves at arbitrary velocity. The observed flux for the distant observer is calculated directly without priority evaluation of the emissivity on the accretion disk, which allows our model to be applicable to the point source that deviates from the axis of the black hole spins and moves with a velocity. To study the impact of the off-axis geometry on the measurement of source properties, we simulate observations for a black hole binary with NuSTAR and eXTP. We compare the simulation with the observation of the phase-resolved spectra of the low-frequency quasiperiodic oscillation observed by the Insight Hard X-ray Modulation Telescope. Due to the nonaxisymmetric illumination on the accretion disk, parameters of the model are not reproduced by the lamppost model, including the corona height, radial velocity, and the reflection fraction. On the other hand, all the model parameters are recovered through the off-axis model.

astro-ph.HE↗

arXiv2025

Path Pooling: Training-Free Structure Enhancement for Efficient Knowledge Graph Retrieval-Augmented Generation

Although Large Language Models achieve strong success in many tasks, they still suffer from hallucinations and knowledge deficiencies in real-world applications. Many knowledge graph-based retrieval-augmented generation (KG-RAG) methods enhance the quality and credibility of LLMs by leveraging structure and semantic information in KGs as external knowledge bases. However, these methods struggle to effectively incorporate structure information, either incurring high computational costs or underutilizing available knowledge. Inspired by smoothing operations in graph representation learning, we propose path pooling, a simple, training-free strategy that introduces structure information through a novel path-centric pooling operation. It seamlessly integrates into existing KG-RAG methods in a plug-and-play manner, enabling richer structure information utilization. Extensive experiments demonstrate that incorporating the path pooling into the state-of-the-art KG-RAG method consistently improves performance across various settings while introducing negligible additional cost.

cs.AI↗

arXiv2025

Lego Sketch: A Scalable Memory-augmented Neural Network for Sketching Data Streams

Sketches, probabilistic structures for estimating item frequencies in infinite data streams with limited space, are widely used across various domains. Recent studies have shifted the focus from handcrafted sketches to neural sketches, leveraging memory-augmented neural networks (MANNs) to enhance the streaming compression capabilities and achieve better space-accuracy trade-offs.However, existing neural sketches struggle to scale across different data domains and space budgets due to inflexible MANN configurations. In this paper, we introduce a scalable MANN architecture that brings to life the {\it Lego sketch}, a novel sketch with superior scalability and accuracy. Much like assembling creations with modular Lego bricks, the Lego sketch dynamically coordinates multiple memory bricks to adapt to various space budgets and diverse data domains. Our theoretical analysis guarantees its high scalability and provides the first error bound for neural sketch. Furthermore, extensive experimental evaluations demonstrate that the Lego sketch exhibits superior space-accuracy trade-offs, outperforming existing handcrafted and neural sketches. Our code is available at https://github.com/FFY0/LegoSketch_ICML.

cs.LG↗