Searcharxiv⌕ Search

arXiv subjects

Hu Liu

Publications and source records attributed to Hu Liu.

At least 19 recordsLinked to original sources
arXiv2026

TrainSDC: Characterizing and Mitigating Silent Data Corruption in Large Language Model Training

LLM training is increasingly vulnerable to silent data corruption (SDC), yet existing protection methods largely treat Transformer computations uniformly because their vulnerability remains poorly understood. We present the first systematic characterization of SDC vulnerability across major computation interfaces in both the forward and backward passes of Transformer training. Our analysis reveals two distinct error propagation mechanisms: forward-pass vulnerability is highly location dependent, with faults on the Q/K path producing persistent training deviations, whereas backward-pass vulnerability is largely governed by gradient exponent distributions rather than computation locations. Motivated by these observations, we propose TrainSDC, a characterization-guided protection framework consisting of Q/K-path recomputation, residual-gain monitoring, and exponent-aware gradient scaling. Experiments on Llama 3.2-1B and Qwen3-0.6B show that TrainSDC maintains training behavior close to fault-free execution under both sparse and dense fault injection while introducing only 1.65%-6.76% runtime overhead.

cs.LG↗
arXiv2026

VFACamou: View-Fused Adversarial Camouflage for Environment-Adaptive Physical Evasion

Adversarial camouflage in the physical world remains highly challenging, particularly under UAV reconnaissance where targets undergo continuous geometric changes and extreme illumination variations. Existing methods either optimize 2D digital perturbations that fail to generalize to dynamic viewpoints or produce visually unnatural textures that cannot be deployed in real scenarios. Therefore, we propose an end-to-end framework for adversarial camouflage generation that automatically produces wearable adversarial patterns and maintains stable attack performance in real physical environments with changing viewpoints, poses, and lighting conditions. Our method integrates UV-volume rendering with a diffusion-based texture generator, enabling consistent appearance under varying scales, poses, and lighting conditions. To ensure environmental realism, we propose an illumination color consistency estimator that extracts dominant background attributes and guides a natural texture loss to align the generated UV texture with the surrounding environment. A multi-scale dynamic training strategy further enhances robustness against viewpoint shifts and body deformation. Extensive experiments across multiple mainstream detectors demonstrate that our method achieves strong and stable physical attack performance while maintaining high perceptual naturalness, reducing human detection rates without introducing unnatural artifacts.

cs.CV↗
arXiv2026

ALM-MTA:Front-Door Causal Multi-Touch Attribution Method for Creator-Ecosystem Optimization

Consumption Drives Production (CDP) on social platforms aims to deliver interpretable incentive signals for creator ecosystem building and resource utilization improvement, which strongly relies on attribution. In large-scale and complex recommendation systems, the absence of accurate labels together with unobserved confounding renders backdoor adjustments alone insufficient for reliable attribution. To address these problems, we propose Adversarial Learning Mediator based Multi-Touch Attribution (ALM-MTA), an extensible causal framework that leverages front-door identification with an adversarially learned mediator: a proxy trained to distill outcome information to strengthen the causal pathway from treatment to outcome and eliminate shortcut leakage. We then introduce contrastive learning that conditions front-door marginalization on high-match consumption-upload pairs to ensure positivity in large treatment spaces. To assess causality from non-RCT logs, we also incorporate a non-personalized bucketed protocol, estimating grouped uplift and computing AUUC over treatment clusters. Finally, we evaluate ALM-MTA using a real-world recommendation system with 400 million DAU and 30 billion samples. ALM-MTA increases DAU by 0.04% and daily active creators by 0.6%, with unit exposure efficiency increased by 670%. On causal utility, ALM-MTA achieves higher grouped AUUC than the SOTA in every propensity bucket, with a maximum gain of 0.070. In terms of accuracy, ALM-MTA improves upload AUC by 40% compared to SOTA. These results demonstrate that front-door deconfounding with adversarial mediator learning provides accurate, personalized, and operationally efficient attribution for creator ecosystem optimization.

cs.SI↗
arXiv2026

HiFloat4 Format for Language Model Pre-training on Ascend NPUs

Large foundation models have become central to modern machine learning, with performance scaling predictably with model size and data. However, training and deploying such models incur substantial computational and memory costs, motivating the development of low-precision training techniques. Recent work has demonstrated that 4-bit floating-point (FP4) formats--such as MXFP4 and NVFP4--can be successfully applied to linear GEMM operations in large language models (LLMs), achieving up to 4x improvements in compute throughput and memory efficiency compared to higher-precision baselines. In this work, we investigate the recently proposed HiFloat4 FP4 format for Huawei Ascend NPUs and systematically compare it with MXFP4 in large-scale training settings. All experiments are conducted on Ascend NPU clusters, with linear and expert GEMM operations performed entirely in FP4 precision. We evaluate both dense architectures (e.g., Pangu and LLaMA-style models) and mixture-of-experts (MoE) models, where both standard linear layers and expert-specific GEMMs operate in FP4. Furthermore, we explore stabilization techniques tailored to FP4 training that significantly reduce numerical degradation, maintaining relative error within 1% of full-precision baselines while preserving the efficiency benefits of 4-bit computation. Our results provide a comprehensive empirical study of FP4 training on NPUs and highlight the practical trade-offs between FP4 formats in large-scale dense and MoE models.

cs.LG↗
arXiv2026

Superposition model for energy reconstruction and mass identification in cosmic ray spectra

The "knee" of cosmic ray spectra may reflect the maximum energy accelerated by galactic cosmic ray sources or the limit of the galaxy's ability to bind cosmic rays. Measurements of individual energy spectra are a crucial tool to understand the origin of the knee. Energy reconstruction and composition identification are foundations of the individual energy spectra measurements. One of the main scientific goals of Large High Altitude Air Shower Observatory (LHAASO) is measuring the cosmic ray energy spectra and composition from ~10 TeV to ~EeV. In this work, a novel method for reconstructing energy and logarithm mass (lnA) based on a superposition model is introduced. Energy and lnA are reconstructed using two universal, composition- and energy-independent calibration lines. For zenith angle below 40 degree, the energy and lnA biases are within +-5% and +-0.3, respectively, across all compositions. The method uses particle densities-measured by LHAASO's electromagnetic and muon detectors at a fixed distance from the shower axis-rather than integrated particle counts in annular bands. The density-based approach improves resolution for both energy and lnA, especially for heavy nuclei. The resulting energy resolution ranges from below 5% to ~15% above 1 PeV, the best mass resolution for iron achieved is below 25% above 10 PeV. The hadronic model dependencies of energy and lnA are also reported. These dependencies scale with lg(E/A) and are nearly independent of primary composition.

astro-ph.HE↗
arXiv2026

HiFloat4 Format for Language Model Inference

This paper introduces HiFloat4 (HiF4), a block floating-point data format tailored for deep learning. Each HiF4 unit packs 64 4-bit elements with 32 bits of shared scaling metadata, averaging 4.5 bits per value. The metadata specifies a three-level scaling hierarchy, capturing inter- and intra-group dynamic range while improving the utilization of the representational space. In addition, the large 64-element group size enables matrix multiplications to be executed in a highly fixed-point manner, significantly reducing hardware area and power consumption. To evaluate the proposed format, we conducted inference experiments on several language models, including LLaMA, Qwen, Mistral, DeepSeek-V3.1 and LongCat. Results show that HiF4 achieves higher average accuracy than the state-of-the-art NVFP4 format across multiple models and diverse downstream tasks.

cs.LG↗
arXiv2026

BAPS: A Fine-Grained Low-Precision Scheme for Softmax in Attention via Block-Aware Precision reScaling

As the performance gains from accelerating quantized matrix multiplication plateau, the softmax operation becomes the critical bottleneck in Transformer inference. This bottleneck stems from two hardware limitations: (1) limited data bandwidth between matrix and vector compute cores, and (2) the significant area cost of high-precision (FP32/16) exponentiation units (EXP2). To address these issues, we introduce a novel low-precision workflow that employs a specific 8-bit floating-point format (HiF8) and block-aware precision rescaling for softmax. Crucially, our algorithmic innovations make low-precision softmax feasible without the significant model accuracy loss that hampers direct low-precision approaches. Specifically, our design (i) halves the required data movement bandwidth by enabling matrix multiplication outputs constrained to 8-bit, and (ii) substantially reduces the EXP2 unit area by computing exponentiations in low (8-bit) precision. Extensive evaluation on language models and multi-modal models confirms the validity of our method. By alleviating the vector computation bottleneck, our work paves the way for doubling end-to-end inference throughput without increasing chip area, and offers a concrete co-design path for future low-precision hardware and software.

cs.LG↗
arXiv2025

The adaptive EM schemes for McKean-Vlasov SDEs with common noise in finite and infinite horizons

This paper is dedicated to investigating the adaptive Euler-Maruyama (EM) schemes for the approximation of McKean-Vlasov stochastic differential equations (SDEs) with common noise. When the drift and diffusion coefficients both satisfy the superlinear growth conditions, the $L^p$ convergence rates in finite and infinite horizons are revealed, which reacts to the particle number and step size. Subsequently, there is an illustration of the theory results by means of two numerical examples.

math.NA↗
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

Study of Stability and Consistency of EAS Thermal Neutron Detection at ENDA-64

Introduction:Electron-Neutron Detector Array (ENDA) is designed to measure thermal neutrons produced by hadronic interactions between cosmic ray extensive air showers (EAS) and the surrounding environment as well as electrons around the cores of EAS. ENDA is located within Large High Altitude Air Shower Observatory (LHAASO). ENDA was expanded from an initial 16 detectors to 64 detectors in April 2023, so called ENDA-64, and has been running alongside LHAASO. The stability and consistency of neutron detection are crucial for laying a solid foundation for subsequent data analysis and physical results. Methods:We obtain the stability by studying variations of event rate and thermal neutron rate in each cluster and the consistency by comparing distribution of number of thermal neutrons between clusters. Additionally, we investigate the specific influences of the rainy and dry seasons, as well as the presence or absence of sand cubes under the detectors, to examine the environmental factors affecting neutron measurement performance. Results:The calibration results indicate good consistency in thermal neutron detection across the clusters, with the maximum inconsistency of 6.85%. The maximum instability of event rate and thermal neutron rate over time are 4.68% and 11.0% respectively. The maximum inconsistency between the clusters without the sand cubes is 18%. The use of sand cubes is effective in protecting the target material from rainwater, and the sand cubes help the cluster to increase collection of neutrons generated by EAS events.

hep-ex↗
arXiv2025

Unconstrained Monotonic Calibration of Predictions in Deep Ranking Systems

Ranking models primarily focus on modeling the relative order of predictions while often neglecting the significance of the accuracy of their absolute values. However, accurate absolute values are essential for certain downstream tasks, necessitating the calibration of the original predictions. To address this, existing calibration approaches typically employ predefined transformation functions with order-preserving properties to adjust the original predictions. Unfortunately, these functions often adhere to fixed forms, such as piece-wise linear functions, which exhibit limited expressiveness and flexibility, thereby constraining their effectiveness in complex calibration scenarios. To mitigate this issue, we propose implementing a calibrator using an Unconstrained Monotonic Neural Network (UMNN), which can learn arbitrary monotonic functions with great modeling power. This approach significantly relaxes the constraints on the calibrator, improving its flexibility and expressiveness while avoiding excessively distorting the original predictions by requiring monotonicity. Furthermore, to optimize this highly flexible network for calibration, we introduce a novel additional loss function termed Smooth Calibration Loss (SCLoss), which aims to fulfill a necessary condition for achieving the ideal calibration state. Extensive offline experiments confirm the effectiveness of our method in achieving superior calibration performance. Moreover, deployment in Kuaishou's large-scale online video ranking system demonstrates that the method's calibration improvements translate into enhanced business metrics. The source code is available at https://github.com/baiyimeng/UMC.

cs.IR↗
arXiv2024

Many-Objective Search-Based Coverage-Guided Automatic Test Generation for Deep Neural Networks

To ensure the reliability of DNN systems and address the test generation problem for neural networks, this paper proposes a fuzzing test generation technique based on many-objective optimization algorithms. Traditional fuzz testing employs random search, leading to lower testing efficiency and tends to generate numerous invalid test cases. By utilizing many-objective optimization techniques, effective test cases can be generated. To achieve high test coverage, this paper proposes several improvement strategies. The frequency-based fuzz sampling strategy assigns priorities based on the frequency of selection of initial data, avoiding the repetitive selection of the same data and enhancing the quality of initial data better than random sampling strategies. To address the issue that global search may yield test not satisfying semantic constraints, a local search strategy based on the Monte Carlo tree search is proposed to enhance the algorithm's local search capabilities. Furthermore, we improve the diversity of the population and the algorithm's global search capability by updating SPEA2's external archive based on a decomposition-based archiving strategy. To validate the effectiveness of the proposed approach, experiments were conducted on several public datasets and various neural network models. The results reveal that, compared to random and clustering-based sampling, the frequency-based fuzz sampling strategy provides a greater improvement in coverage rate in the later stages of iterations. On complex networks like VGG16, the improved SPEA2 algorithm increased the coverage rate by about 12% across several coverage metrics, and by approximately 40% on LeNet series networks. The experimental results also indicates that the newly generated test cases not only exhibit higher coverage rates but also generate adversarial samples that reveal model errors.

cs.SE↗
arXiv2024

Ascend HiFloat8 Format for Deep Learning

This preliminary white paper proposes a novel 8-bit floating-point data format HiFloat8 (abbreviated as HiF8) for deep learning. HiF8 features tapered precision. For normal value encoding, it provides 7 exponent values with 3-bit mantissa, 8 exponent values with 2-bit mantissa, and 16 exponent values with 1-bit mantissa. For denormal value encoding, it extends the dynamic range by 7 extra powers of 2, from 31 to 38 binades (notice that FP16 covers 40 binades). Meanwhile, HiF8 encodes all the special values except that positive zero and negative zero are represented by only one bit-pattern. Thanks to the better balance between precision and dynamic range, HiF8 can be simultaneously used in both forward and backward passes of AI training. In this paper, we will describe the definition and rounding methods of HiF8, as well as the tentative training and inference solutions. To demonstrate the efficacy of HiF8, massive simulation results on various neural networks, including traditional neural networks and large language models (LLMs), will also be presented.

cs.LG↗
arXiv2024

Beyond Relevance: Improving User Engagement by Personalization for Short-Video Search

Personalized search has been extensively studied in various applications, including web search, e-commerce, social networks, etc. With the soaring popularity of short-video platforms, exemplified by TikTok and Kuaishou, the question arises: can personalization elevate the realm of short-video search, and if so, which techniques hold the key? In this work, we introduce $\text{PR}^2$, a novel and comprehensive solution for personalizing short-video search, where $\text{PR}^2$ stands for the Personalized Retrieval and Ranking augmented search system. Specifically, $\text{PR}^2$ leverages query-relevant collaborative filtering and personalized dense retrieval to extract relevant and individually tailored content from a large-scale video corpus. Furthermore, it utilizes the QIN (Query-Dominate User Interest Network) ranking model, to effectively harness user long-term preferences and real-time behaviors, and efficiently learn from user various implicit feedback through a multi-task learning framework. By deploying the $\text{PR}^2$ in production system, we have achieved the most remarkable user engagement improvements in recent years: a 10.2% increase in CTR@10, a notable 20% surge in video watch time, and a 1.6% uplift of search DAU. We believe the practical insights presented in this work are valuable especially for building and improving personalized search systems for the short video platforms.

cs.IR↗
arXiv2024

A Self-boosted Framework for Calibrated Ranking

Scale-calibrated ranking systems are ubiquitous in real-world applications nowadays, which pursue accurate ranking quality and calibrated probabilistic predictions simultaneously. For instance, in the advertising ranking system, the predicted click-through rate (CTR) is utilized for ranking and required to be calibrated for the downstream cost-per-click ads bidding. Recently, multi-objective based methods have been wildly adopted as a standard approach for Calibrated Ranking, which incorporates the combination of two loss functions: a pointwise loss that focuses on calibrated absolute values and a ranking loss that emphasizes relative orderings. However, when applied to industrial online applications, existing multi-objective CR approaches still suffer from two crucial limitations. First, previous methods need to aggregate the full candidate list within a single mini-batch to compute the ranking loss. Such aggregation strategy violates extensive data shuffling which has long been proven beneficial for preventing overfitting, and thus degrades the training effectiveness. Second, existing multi-objective methods apply the two inherently conflicting loss functions on a single probabilistic prediction, which results in a sub-optimal trade-off between calibration and ranking. To tackle the two limitations, we propose a Self-Boosted framework for Calibrated Ranking (SBCR).

cs.IR↗
arXiv2024

Research on the knee region of cosmic ray by using a novel type of electron-neutron detector array

By accurately measuring composition and energy spectrum of cosmic ray, the origin problem of so called "keen" region (energy > 1 PeV) can be solved. However, up to the present, the results of the spectrum in the knee region obtained by several previous experiments have shown obvious differences, so they cannot give effective evidence for judging the theoretical models on the origin of the knee. Recently, the Large High Altitude Air Shower Observatory (LHAASO) has reported several major breakthroughs and important results in astro-particle physics field. Relying on its advantages of wide-sky survey, high altitude location and large area detector arrays, the research content of LHAASO experiment mainly includes ultra high-energy gamma-ray astronomy, measurement of cosmic ray spectra in the knee region, searching for dark matter and new phenomena of particle physics at higher energy. The electron and Thermal Neutron detector (EN-Detector) is a new scintillator detector which applies thermal neutron detection technology to measure cosmic ray extensive air shower (EAS). This technology is an extension of LHAASO. The EN-Detector Array (ENDA) can highly efficiently measure thermal neutrons generated by secondary hadrons so called "skeleton" of EAS. In this paper, we perform the optimization of ENDA configuration, and obtain expectations on the ENDA results, including thermal neutron distribution, trigger efficiency and capability of cosmic ray composition separation. The obtained real data results are consistent with those by the Monte Carlo simulation.

astro-ph.HE↗
arXiv2023

Study of longitudinal development of air showers in the knee energy range

Ground-based cosmic ray experiments detect cosmic ray mainly by measuring the longitudinal and lateral distribution of secondary particles produced in the extensive air shower (EAS). The EAS of cosmic ray in the knee energy region is simulated via CORSIKA software. Several simulation samples with different energy, composition and zenith angles were carried out to understand the longitudinal development of electron, muon and Cherenkov light in EAS. All the results presented were obtained assuming an observation plane at an altitude of 4400 m a.s.l. The differences of longitudinal development between electron and Cherenkov light were studied, and the reconstruction uncertainty of shower maximum for electron from Cherenkov light was estimated to be 10-15g/cm$^2$ for nuclei above 1 PeV. The performances of energy measurement and the composition discrimination ability based on longitudinal development were studied and compared with that from lateral distribution. It was found that number of electron per depth at its shower maximum has the smallest shower-to-shower fluctuations, but the shower-to-shower fluctuations of electron density measured at observation level was very close to it when the appropriate zenith angle was employed. The shower-to-shower fluctuations of shower maximum for electron is 50-55 g/cm$^2$ for proton, and 20-25 g/cm$^2$ for iron, but the composition discrimination ability between nuclei from muon density measured at observation level is much better than the shower maximum variable from longitudinal development. The hadronic model dependencies of the longitudinal development and lateral distribution were also discussed.

astro-ph.HE↗
arXiv2023

Properties of secondary components in extensive air shower of cosmic rays in knee energy region

The knee of cosmic ray spectra reflects the maximum energy accelerated by galactic cosmic ray sources or the limit to the ability of galaxy to bind cosmic rays. The measuring of individual energy spectra is a crucial tool to ascertain the origin of the knee. The Extensive Air Shower of cosmic rays in the knee energy region is simulated via CORSIKA software. The energy resolution for different secondary components and primary nuclei identification capability are studied. The energy reconstruction by using electromagnetic particles in the energy around knee is better than by using other secondary particles. The resolution is 10-19 percent for proton, and 4-8 percent for iron. For the case of primary nuclei identification capability, the discriminability of density of muons is best both at low (around 100 TeV) and high (around 10 PeV) energy, the discriminability of the shape of lateral distribution of electron and gamma-rays are good at low energy and the discriminability of density of neutrons is good at high energy. The differences between the lateral distributions of secondary particles simulated by EPOS-LHC and QGSJet-II-04 hadronic model are also studied. The results in this work can provide important information for selecting the secondary components and detector type during energy reconstruction and identifying the primary nuclei of cosmic rays in the knee region.

astro-ph.HE↗