arXiv subjects

Zhenyu Song

Publications and source records attributed to Zhenyu Song.

10 recordsLinked to original sources

arXiv2026

CMNIE: An Information Extraction Benchmark for Chinese Military News

Structured extraction from Chinese military news supports intelligence analysis, decision-making, and knowledge base construction. However, existing resources provide limited support for joint informa?tion extraction in this domain, especially when events, event arguments, entities, and relations must be modeled together. We present CMNIE, an information extraction benchmark for Chinese military news. Extend?ing military-domain resources beyond document-level event annotations, CMNIE jointly annotates event triggers, event arguments, named enti?ties, and entity relations under a unified domain schema. The dataset contains 13,000 instances collected from public Chinese military news, with manual annotations for 7 event types, 10 argument roles, 7 entity types, and 8 relation types. We evaluate supervised IE models, zero-shot large language models, and fine-tuned LLM-based extraction methods on a shared test set. Experimental results show that CMNIE remains chal?lenging, especially for relation extraction and exact matching of event?argument spans; zero-shot LLMs often identify relevant semantic units but fail to match gold span boundaries exactly. CMNIE provides a stan?dardized benchmark for studying schema adherence, exact span match?ing, and joint structured extraction in specialized Chinese news.

cs.CL↗

arXiv2026

vCause: Efficient and Verifiable Causality Analysis for Cloud-based Endpoint Auditing

In cloud-based endpoint auditing, security administrators often rely on the cloud to perform causality analysis over log-derived versioned provenance graphs to investigate suspicious attack behaviors. However, the cloud may be distrusted or compromised by attackers, potentially manipulating the final causality analysis results. Consequently, administrators may not accurately understand attack behaviors and fail to implement effective countermeasures. This risk underscores the need for a defense scheme to ensure the integrity of causality analysis. While existing tamper-evident logging schemes and trusted execution environments show promise for this task, they are not specifically designed to support causality analysis and thus face inherent security and efficiency limitations. This paper presents vCause, an efficient and verifiable causality analysis system for cloud-based endpoint auditing. vCause integrates two authenticated data structures: a graph accumulator and a verifiable provenance graph. The data structures enable validation of two critical steps in causality analysis: (i) querying a point-of-interest node on a versioned provenance graph, and (ii) identifying its causally related components. Formal security analysis and experimental evaluation show that vCause can achieve secure and verifiable causality analysis with only <1% computational overhead on endpoints and 3.36% on the cloud.

cs.CR↗

arXiv2026

CRAFT: Fine-Grained Cost-Aware Expert Replication For Efficient Mixture-of-Experts Serving

Mixture-of-Experts (MoE) has recently emerged as the mainstream architecture for efficiently scaling large language models while maintaining near-constant computational cost. Expert parallelism distributes parameters by partitioning experts across devices, but this introduces token-level load imbalance during inference. Expert replication is a widely adopted load-balancing technique in serving frameworks that alleviates load imbalance in large-scale deployments by replicating experts with high loads. In this work, we demonstrate that existing replication schemes often over-replicate, with many replicas providing marginal improvement. Replicas consume substantial GPU memory, which may lead to resource contention and throughput degradation. We present CRAFT, an efficient expert replication framework that maximizes load balance under a given memory budget by performing fine-grained, per-layer replication based on the estimated replication benefit. CRAFT can be seamlessly integrated into existing serving frameworks without any additional training or model changes. Our evaluation shows that CRAFT increases end-to-end serving throughput by $1.14\times$ on average (up to $1.2\times$) over existing replication techniques in large-scale deployments with models ranging from hundreds of billions to a trillion parameters.

cs.DC↗

arXiv2025

AccelOpt: A Self-Improving LLM Agentic System for AI Accelerator Kernel Optimization

We present AccelOpt, a self-improving large language model (LLM) agentic system that autonomously optimizes kernels for emerging AI acclerators, eliminating the need for expert-provided hardware-specific optimization knowledge. AccelOpt explores the kernel optimization space through iterative generation, informed by an optimization memory that curates experiences and insights from previously encountered slow-fast kernel pairs. We build NKIBench, a new benchmark suite of AWS Trainium accelerator kernels with varying complexity extracted from real-world LLM workloads to evaluate the effectiveness of AccelOpt. Our evaluation confirms that AccelOpt's capability improves over time, boosting the average percentage of peak throughput from $49\%$ to $61\%$ on Trainium 1 and from $45\%$ to $59\%$ on Trainium 2 for NKIBench kernels. Moreover, AccelOpt is highly cost-effective: using open-source models, it matches the kernel improvements of Claude Sonnet 4 while being $26\times$ cheaper. The code is open-sourced at https://github.com/zhang677/AccelOpt.

cs.LG↗

arXiv2025

TTrace: Lightweight Error Checking and Diagnosis for Distributed Training

Distributed training is essential for scaling the training of large neural network models, such as large language models (LLMs), across thousands of GPUs. However, the complexity of distributed training programs makes them particularly prone to silent bugs, which do not produce explicit error signals but lead to incorrect training outcomes. Effectively detecting and localizing such silent bugs in distributed training is challenging. Common debugging practices based on monitoring training loss or gradient norm curves are indirect, inefficient, and provide no way to localize bugs. To address those challenges, we design and implement TTrace, the first systematic differential testing system for detecting and localizing silent bugs in distributed training. TTrace aligns intermediate tensors from distributed training with those from a trusted reference implementation. To properly compare the floating-point values in the corresponding tensors, we propose a novel mathematical analysis that provides a guideline for setting tolerances, enabling TTrace to distinguish bug-induced errors from numerical errors. Experimental results demonstrate that TTrace effectively detects 11 existing bugs and 3 new bugs in the widely used Megatron-LM framework, while requiring fewer than 10 lines of code changes. TTrace is effective in various training recipes, including low-precision recipes involving BF16 and FP8. Notably, a popular open-source training framework has already adopted the method proposed by TTrace in its development workflow.

cs.DC↗

arXiv2024

Cooperative Reward Shaping for Multi-Agent Pathfinding

The primary objective of Multi-Agent Pathfinding (MAPF) is to plan efficient and conflict-free paths for all agents. Traditional multi-agent path planning algorithms struggle to achieve efficient distributed path planning for multiple agents. In contrast, Multi-Agent Reinforcement Learning (MARL) has been demonstrated as an effective approach to achieve this objective. By modeling the MAPF problem as a MARL problem, agents can achieve efficient path planning and collision avoidance through distributed strategies under partial observation. However, MARL strategies often lack cooperation among agents due to the absence of global information, which subsequently leads to reduced MAPF efficiency. To address this challenge, this letter introduces a unique reward shaping technique based on Independent Q-Learning (IQL). The aim of this method is to evaluate the influence of one agent on its neighbors and integrate such an interaction into the reward function, leading to active cooperation among agents. This reward shaping method facilitates cooperation among agents while operating in a distributed manner. The proposed approach has been evaluated through experiments across various scenarios with different scales and agent counts. The results are compared with those from other state-of-the-art (SOTA) planners. The evidence suggests that the approach proposed in this letter parallels other planners in numerous aspects, and outperforms them in scenarios featuring a large number of agents.

cs.AI↗

arXiv2024

Faster Post-Quantum TLS 1.3 Based on ML-KEM: Implementation and Assessment

TLS is extensively utilized for secure data transmission over networks. However, with the advent of quantum computers, the security of TLS based on traditional public-key cryptography is under threat. To counter quantum threats, it is imperative to integrate post-quantum algorithms into TLS. Most PQ-TLS research focuses on integration and evaluation, but few studies address the improvement of PQ-TLS performance by optimizing PQC implementation. For the TLS protocol, handshake performance is crucial, and for post-quantum TLS (PQ-TLS) the performance of post-quantum key encapsulation mechanisms (KEMs) directly impacts handshake performance. In this work, we explore the impact of post-quantum KEMs on PQ-TLS performance. We explore how to improve ML-KEM performance using the latest Intel's Advanced Vector Extensions instruction set AVX-512. We detail a spectrum of techniques devised to parallelize polynomial multiplication, modular reduction, and other computationally intensive modules within ML-KEM. Our optimized ML-KEM implementation achieves up to 1.64x speedup compared to the latest AVX2 implementation. Furthermore, we introduce a novel batch key generation method for ML-KEM that can seamlessly integrate into the TLS protocols. The batch method accelerates the key generation procedure by 3.5x to 4.9x. We integrate the optimized AVX-512 implementation of ML-KEM into TLS 1.3, and assess handshake performance under both PQ-only and hybrid modes. The assessment demonstrates that our faster ML-KEM implementation results in a higher number of TLS 1.3 handshakes per second under both modes. Additionally, we revisit two IND-1-CCA KEM constructions discussed in Eurocrypt22 and Asiacrypt23. Besides, we implement them based on ML-KEM and integrate the one of better performance into TLS 1.3 with benchmarks.

cs.CR↗

arXiv2023

Optimized Vectorization Implementation of CRYSTALS-Dilithium

CRYSTALS-Dilithium is a lattice-based signature scheme to be standardized by NIST as the primary post-quantum signature algorithm. In this work, we make a thorough study of optimizing the implementations of Dilithium by utilizing the Advanced Vector Extension (AVX) instructions, specifically AVX2 and the latest AVX-512. We first present an improved parallel small polynomial multiplication with tailored early evaluation (PSPM-TEE) to further speed up the signing procedure. Our PSPM algorithm outperform the NTT by 47%-66% in AVX2 and AVX-512 implementation. We then present a tailored reduction method that is simpler and faster than Montgomery reduction. We minimize the CPU cycles of tailored reduction AVX-512 implementation by using AVX-512IFMA. Finally, we propose a fully and highly vectorized implementation of Dilithium using AVX-512. This is achieved by carefully vectorizing most of Dilithium functions with the AVX-512 instructions in order to improve efficiency both for time and for space simultaneously. With all the optimization efforts, our AVX-512 implementation improves the performance by 43.2%/39.3%/45.6% in key generation, 36.6%/41.6%/43.7% in signing, and 45.3%/46.5%/47.4% in verification for the parameter sets of Dilithium2/3/5 respectively. To the best of our knowledge, our AVX-512 implementation has the best performance for Dilithium on the Intel x86-64 CPU platform to date.

cs.CR↗

arXiv2016

PIWD: A Plugin-based Framework for Well-Designed SPARQL

In the real world datasets (e.g.,DBpedia query log), queries built on well-designed patterns containing only AND and OPT operators (for short, WDAO-patterns) account for a large proportion among all SPARQL queries. In this paper, we present a plugin-based framework for all SELECT queries built on WDAO-patterns, named PIWD. The framework is based on a parse tree called \emph{well-designed AND-OPT tree} (for short, WDAO-tree) whose leaves are basic graph patterns (BGP) and inner nodes are the OPT operators. We prove that for any WDAO-pattern, its parse tree can be equivalently transformed into a WDAO-tree. Based on the proposed framework, we can employ any query engine to evaluate BGP for evaluating queries built on WDAO-patterns in a convenient way. Theoretically, we can reduce the query evaluation of WDAO-patterns to subgraph homomorphism as well as BGP since the query evaluation of BGP is equivalent to subgraph homomorphism. Finally, our preliminary experiments on gStore and RDF-3X show that PIWD can answer all queries built on WDAO-patterns effectively and efficiently.

cs.DB↗

arXiv2016

Efficient Approximation of Well-Designed SPARQL Queries

Query response time often influences user experience in the real world. However, it possibly takes more time to answer a query with its all exact solutions, especially when it contains the OPT operations since the OPT operation is the least conventional operator in SPARQL. So it becomes essential to make a trade-off between the query response time and the accuracy of their solutions. In this paper, based on the depth of the OPT operation occurring in a query, we propose an approach to obtain its all approximate queries with less depth of the OPT operation. This paper mainly discusses those queries with well-designed patterns since the OPT operation in a well-designed pattern is really "optional". Firstly, we transform a well-designed pattern in OPT normal form into a well-designed tree, whose inner nodes are labeled by OPT operation and leaf nodes are labeled by patterns containing other operations such as the AND operation and the FILTER operation. Secondly, based on this well-designed tree, we remove "optional" well-designed subtrees with less depth of the OPT operation and then obtain approximate queries with different depths of the OPT operation. Finally, we evaluate the approximate query efficiency with the degree of approximation.

cs.DB↗