Searcharxiv⌕ Search

arXiv subjects

Yifan Huang

Publications and source records attributed to Yifan Huang.

At least 19 recordsLinked to original sources
arXiv2026

Mechanism-Driven Monitors for Preemptive Detection of LLM Training Instability

Frontier large language model training consumes massive accelerator fleets and long wall-clock computation, making stability failures costly when they occur. After a numerical or a hyperparameter fault has already destabilized the training dynamics, it may continue for thousands of steps while loss and gradient norms still appear normal. We study mechanism-driven detection of training instability by deriving internal monitors from the functional role of each critical module and from the earliest computational sites where failures are expected to produce measurable signatures. For low-precision flash attention, we monitor the spectral entropy of a QK bilinear decomposition, whose first-order term becomes abnormal before the loss fully collapses. For MoE routers, we derive indicators from their role in expert selection. Our fault-injection experiments on low-precision attention, large learning-rate, and combined faults show that these signals provide distinct signatures for different failures, triggering thousands of steps before loss divergence.

cs.CL↗
arXiv2026

Efficiently Training Time-to-First-Spike Spiking Neural Networks from Scratch

Spiking Neural Networks (SNNs), with their event-driven and biologically inspired mechanisms, are well-suited for energy-efficient neuromorphic hardware. Neural coding, which is critical to SNNs, determines how information is represented via spikes. While Time-to-First-Spike (TTFS) coding uses a single spike per neuron to offer extreme sparsity and energy efficiency, it often suffers from unstable training and low accuracy due to its sparse firing. To address these challenges, we propose a training framework that incorporates parameter initialization, training normalization, a temporal output decoder, and a re-evaluation of the pooling layer. The proposed parameter initialization and training normalization mitigate signal diminishing and gradient vanishing, which helps stabilize training. Our output decoder aggregates temporal spikes to encourage earlier firing, thereby reducing latency. The re-evaluation of the pooling layer demonstrates that max-pooling violates single-spike constraints, which should be avoided, whereas average-pooling preserves them. Experiments show that our framework stabilizes and accelerates training, reduces latency, and achieves state-of-the-art accuracy for step-by-step TTFS SNNs on MNIST (99.48%), Fashion-MNIST (92.90%), CIFAR10 (90.56%), CIFAR100 (70.27%) and DVS Gesture (95.83%).

cs.NE↗
arXiv2026

Graph Neural Network Force Fields (GPTFF-mol) for Organic Molecules from Optimization Trajectories (OpenGEM26)

Density functional theory (DFT) serves as a reliable tool for atomistic molecular simulations, while machine learning potentials have become powerful complements to balance accuracy and efficiency. In this work, we release OpenGEM26 (Open Generated Ensemble of Molecules, 2026), a large-scale dataset comprising 200,000 unique molecules and 4.4 million conformations composed of H, C, N, O, S and Cl with up to ten heavy atoms. All calculations are carried out at the ωB97X-D/Def2-SVP and Def2-TZVP levels with dispersion corrections, and complete structural optimization trajectories and abundant non-equilibrium structures are recorded. Statistical analyses confirm that this dataset covers a broader conformational space than QM9 in terms of energy, bond lengths and bond angles. A graph neural network-based potential GPTFF-mol is trained using the new dataset, achieving an energy mean absolute error of 16 meV/molecule, which is equivalent to 0.82meV/atom, and superior force prediction performance compared with ANI-2x. Validated by butane rotation and keto-enol tautomerization tests, the model accurately describes molecular dynamical behaviors and reaction barriers at distorted geometries. This work provides a high-quality resource and robust ML potential for efficient simulations of sulfur- and chlorine-containing organic molecules.

physics.chem-ph↗
arXiv2026

From Noise to Intent: Anchoring Generative VLA Policies with Residual Bridges

Bridging high-level semantic understanding with low-level physical control remains a persistent challenge in embodied intelligence, stemming from the fundamental spatiotemporal scale mismatch between cognition and action. Existing generative VLA policies typically adopt a "Generation-from-Noise" paradigm, which disregards this disparity, leading to representation inefficiency and weak condition alignment during optimization. In this work, we propose ResVLA, an architecture that shifts the paradigm to "Refinement-from-Intent." Recognizing that robotic motion naturally decomposes into global intent and local dynamics, ResVLA utilizes spectral analysis to decouple control into a deterministic low-frequency anchor and a stochastic high-frequency residual. By anchoring the generative process on the predicted intent, our model focuses strictly on refining local dynamics via a residual diffusion bridge. Extensive simulation experiments show that ResVLA achieves competitive performance, strong robustness to language and robot embodiment perturbations, and faster convergence than standard generative baselines. ResVLA also demonstrates strong performance in real-world robot experiments.

cs.RO↗
arXiv2026

Continual LLM Upcycling: A Predictor-Gated Bank-Wise Sparsity Training Recipe for Dense-to-Sparse LLMs

We study dense-to-sparse continual training as a way to construct channel-sparse large language models from dense checkpoints. Starting from a Qwen2.5-8B dense backbone, we continue training at 32K context and introduce a predictor-gated sparse SwiGLU FFN in the 32K stage. For each token and layer, we use a low-rank predictor to produce FFN-channel routing logits. We then apply a bank-wise top-k rule to retain 16 channels in every 64-channel bank, yielding 4x sparsity in the FFN intermediate activation. Unlike post-hoc sparse inference methods, the routing module is placed on the main language modeling path and optimized during continual training, enabling the dense model to be upcycled into a hardware-oriented sparse model. We report the architecture, training recipe, benchmark performance, and training lessons. We also identify a layer-local long-context failure mode on RULER-CWE and propose a single-layer repair algorithm that substantially improves the affected length range.

cs.CL↗
arXiv2026

Faster-GCG: Efficient Discrete Optimization Jailbreak Attacks against Aligned Large Language Models

Aligned Large Language Models (LLMs) have attracted significant attention for their safety, particularly in the context of jailbreak attacks that attempt to bypass guardrails via adversarial prompts. Among existing approaches, the Greedy Coordinate Gradient (GCG) attack pioneered automated jailbreaks through discrete token optimization; however, its low sample efficiency limits practical applicability. In particular, GCG requires approximately 256K evaluations per harmful behavior to achieve a satisfactory jailbreak success rate, due to the inherent difficulty of the underlying discrete optimization problem. In this work, we identify three key factors that limit the sample efficiency of GCG: inaccurate gradient-based estimation, inefficient uniform sampling, and repeated evaluation of previously explored suffixes. To address these issues, we propose Faster-GCG, a streamlined variant of GCG that incorporates distance-based regularization for improved estimation, temperature-controlled sampling for more effective exploration, and a visited-suffix marking mechanism to avoid redundant evaluations. Faster-GCG reduced the required evaluations to 32K, achieving up to an $8\times$ improvement in sampling efficiency and a $7\times$ reduction in wall-clock time compared to GCG. Under this reduced budget, Faster-GCG attained an average jailbreak success rate of 78.1\% across five aligned LLMs, and achieved 88.7\% against Qwen3.5-4B, outperforming state-of-the-art white-box jailbreak methods.

cs.LG↗
arXiv2026

Rethinking SNN Online Training and Deployment: Gradient-Coherent Learning via Hybrid-Driven LIF Model

Spiking Neural Networks (SNNs) are considered to have enormous potential in the future development of Artificial Intelligence due to their brain-inspired and energy-efficient properties. Compared to vanilla Spatial-Temporal Back-propagation (STBP) training methods, online training can effectively avoid the risk of GPU memory explosion. However, current online learning frameworks cannot tackle the gradient discrepancy problem between the forward and backward process, merely aiming to optimize the GPU memory, resulting in no performance advantages compared to the STBP-based models in the inference stage. To address the aforementioned challenges, we propose Hybrid-Driven Leaky Integrate-and-Fire (HD-LIF) model family for efficient online learning, which respectively adopt different spiking calculation mechanism in the upper-region and lower-region of the firing threshold. We theoretically point out that our learning framework can effectively separate temporal gradients and address the misalignment problem of surrogate gradients, as well as achieving full-stage optimization towards learning precision, memory complexity and power consumption. Experimental results have demonstrated that our scheme is enable to achieve state-of-the-art performance for multiple evaluation metrics, breaking through the traditional paradigm of SNN online training and deployment. Code is available at \href{https://github.com/hzc1208/HD_LIF}{here}.

cs.NE↗
arXiv2026

LLM-Bootstrapped Targeted Finding Guidance for Factual MLLM-based Medical Report Generation

The automatic generation of medical reports utilizing Multimodal Large Language Models (MLLMs) frequently encounters challenges related to factual instability, which may manifest as the omission of findings or the incorporation of inaccurate information, thereby constraining their applicability in clinical settings. Current methodologies typically produce reports based directly on image features, which inherently lack a definitive factual basis. In response to this limitation, we introduce Fact-Flow, an innovative framework that separates the process of visual fact identification from the generation of reports. This is achieved by initially predicting clinical findings from the image, which subsequently directs the MLLM to produce a report that is factually precise. A pivotal advancement of our approach is a pipeline that leverages a Large Language Model (LLM) to autonomously create a dataset of labeled medical findings, effectively eliminating the need for expensive manual annotation. Extensive experimental evaluations conducted on two disease-focused medical datasets validate the efficacy of our method, demonstrating a significant enhancement in factual accuracy compared to state-of-the-art models, while concurrently preserving high standards of text quality.

cs.CL↗
arXiv2025

Casting a SPELL: Sentence Pairing Exploration for LLM Limitation-breaking

Large language models (LLMs) have revolutionized software development through AI-assisted coding tools, enabling developers with limited programming expertise to create sophisticated applications. However, this accessibility extends to malicious actors who may exploit these powerful tools to generate harmful software. Existing jailbreaking research primarily focuses on general attack scenarios against LLMs, with limited exploration of malicious code generation as a jailbreak target. To address this gap, we propose SPELL, a comprehensive testing framework specifically designed to evaluate the weakness of security alignment in malicious code generation. Our framework employs a time-division selection strategy that systematically constructs jailbreaking prompts by intelligently combining sentences from a prior knowledge dataset, balancing exploration of novel attack patterns with exploitation of successful techniques. Extensive evaluation across three advanced code models (GPT-4.1, Claude-3.5, and Qwen2.5-Coder) demonstrates SPELL's effectiveness, achieving attack success rates of 83.75%, 19.38%, and 68.12% respectively across eight malicious code categories. The generated prompts successfully produce malicious code in real-world AI development tools such as Cursor, with outputs confirmed as malicious by state-of-the-art detection systems at rates exceeding 73%. These findings reveal significant security gaps in current LLM implementations and provide valuable insights for improving AI safety alignment in code generation applications.

cs.CR↗
arXiv2025

Scalable Dendritic Modeling Advances Expressive and Robust Deep Spiking Neural Networks

Dendritic computation endows biological neurons with rich nonlinear integration and high representational capacity, yet it is largely missing in existing deep spiking neural networks (SNNs). Although detailed multi-compartment models can capture dendritic computations, their high computational cost and limited flexibility make them impractical for deep learning. To combine the advantages of dendritic computation and deep network architectures for a powerful, flexible and efficient computational model, we propose the dendritic spiking neuron (DendSN). DendSN explicitly models dendritic morphology and nonlinear integration in a streamlined design, leading to substantially higher expressivity than point neurons and wide compatibility with modern deep SNN architectures. Leveraging the efficient formulation and high-performance Triton kernels, dendritic SNNs (DendSNNs) can be efficiently trained and easily scaled to deeper networks. Experiments show that DendSNNs consistently outperform conventional SNNs on classification tasks. Furthermore, inspired by dendritic modulation and synaptic clustering, we introduce the dendritic branch gating (DBG) algorithm for task-incremental learning, which effectively reduces inter-task interference. Additional evaluations show that DendSNNs exhibit superior robustness to noise and adversarial attacks, along with improved generalization in few-shot learning scenarios. Our work firstly demonstrates the possibility of training deep SNNs with multiple nonlinear dendritic branches, and comprehensively analyzes the impact of dendrite computation on representation learning across various machine learning settings, thereby offering a fresh perspective on advancing SNN design.

cs.NE↗
arXiv2025

Bayesian Statistical Inversion for High-Dimensional Computer Model Output and Spatially Distributed Counts

Data collected by the Interstellar Boundary Explorer (IBEX) satellite, recording heliospheric energetic neutral atoms (ENAs), exhibit a phenomenon that has caused space scientists to revise hypotheses about the physical processes, and computer simulations under those models, in play at the boundary of our solar system. Evaluating the fit of these computer models involves tuning their parameters to observational data from IBEX. This would be a classic (Bayesian) inverse problem if not for three challenges: (1) the computer simulations are slow, limiting the size of campaigns of runs; so (2) surrogate modeling is essential, but outputs are high-resolution images, thwarting conventional methods; and (3) IBEX observations are counts, whereas most inverse problem techniques assume Gaussian field data. To fill that gap we propose a novel approach to Bayesian inverse problems coupling a Poisson response with a sparse Gaussian process surrogate using the Vecchia approximation. We demonstrate the capabilities of our proposed framework, which compare favorably to alternatives, through multiple simulated examples in terms of recovering "true" computer model parameters and accurate out-of-sample prediction. We then apply this new technology to IBEX satellite data and associated computer models developed at Los Alamos National Laboratory.

stat.AP↗
arXiv2025

AsyncHZP: Hierarchical ZeRO Parallelism with Asynchronous Scheduling for Scalable LLM Training

The training efficiency and scalability of language models on massive clusters currently remain a critical bottleneck. Mainstream approaches like ND parallelism are often cumbersome and complex, while flexible alternatives such as the Zero Redundancy Optimizer (ZeRO) are frequently hampered by communication overhead. In this paper, we propose Asynchronous Hierarchical Zero Parallelism (AsyncHZP), a novel asynchronous variant of ZeRO designed to achieve superior performance while maintaining simplicity and memory efficiency. Unlike traditional ZeRO, which employs over-fine-grained sharding that can lead to inefficient communication, AsyncHZP adaptively reshards parameters, gradients, and optimizer states across different replica groups. This strategy optimizes device memory utilization and significantly reduces communication overhead. In addition, we also design a multi-stream asynchronous scheduling method that executes parameter all-gather and gradient reduce-scatter operations in dedicated background threads, effectively overlapping communication with computation while incurring negligible memory fragmentation. Empirical evaluations on both Dense and Mixture-of-Experts (MoE) models confirm that AsyncHZP maintains robust stability at scale. It consistently outperforms classic ND parallelism, achieving state-of-the-art performance without complex strategic tuning, thereby simplifying the path to efficient large-scale training.

cs.DC↗
arXiv2025

Approximate Revenue Maximization for Diffusion Auctions

Reserve prices are widely used in practice. The problem of designing revenue-optimal auctions based on reserve price has drawn much attention in the auction design community. Although they have been extensively studied, most developments rely on the significant assumption that the target audience of the sale is directly reachable by the auctioneer, while a large portion of bidders in the economic network unaware of the sale are omitted. This work follows the diffusion auction design, which aims to extend the target audience of optimal auction theory to all entities in economic networks. We investigate the design of simple and provably near-optimal network auctions via reserve price. Using Bayesian approximation analysis, we provide a simple and explicit form of the reserve price function tailored to the most representative network auction. We aim to balance setting a sufficiently high reserve price to induce high revenue in a successful sale, and attracting more buyers from the network to increase the probability of a successful sale. This reserve price function preserves incentive compatibility for network auctions, allowing the seller to extract additional revenue beyond that achieved by the Myerson optimal auction. Specifically, if the seller has $ρ$ direct neighbours in a network of size $n$, this reserve price guarantees a $1-{1 \over ρ}$ approximation to the theoretical upper bound, i.e., the maximum possible revenue from any network of size $n$. This result holds for any size and any structure of the networked market.

econ.TH↗
arXiv2025

PBCAT: Patch-based composite adversarial training against physically realizable attacks on object detection

Object detection plays a crucial role in many security-sensitive applications. However, several recent studies have shown that object detectors can be easily fooled by physically realizable attacks, \eg, adversarial patches and recent adversarial textures, which pose realistic and urgent threats. Adversarial Training (AT) has been recognized as the most effective defense against adversarial attacks. While AT has been extensively studied in the $l_\infty$ attack settings on classification models, AT against physically realizable attacks on object detectors has received limited exploration. Early attempts are only performed to defend against adversarial patches, leaving AT against a wider range of physically realizable attacks under-explored. In this work, we consider defending against various physically realizable attacks with a unified AT method. We propose PBCAT, a novel Patch-Based Composite Adversarial Training strategy. PBCAT optimizes the model by incorporating the combination of small-area gradient-guided adversarial patches and imperceptible global adversarial perturbations covering the entire image. With these designs, PBCAT has the potential to defend against not only adversarial patches but also unseen physically realizable attacks such as adversarial textures. Extensive experiments in multiple settings demonstrated that PBCAT significantly improved robustness against various physically realizable attacks over state-of-the-art defense methods. Notably, it improved the detection accuracy by 29.7\% over previous defense methods under one recent adversarial texture attack.

cs.CV↗
arXiv2025

Seed-Coder: Let the Code Model Curate Data for Itself

Code data in large language model (LLM) pretraining is recognized crucial not only for code-related tasks but also for enhancing general intelligence of LLMs. Current open-source LLMs often heavily rely on human effort to produce their code pretraining data, such as employing hand-crafted filtering rules tailored to individual programming languages, or using human-annotated data to train quality filters. However, these approaches are inherently limited in scalability, prone to subjective biases, and costly to extend and maintain across diverse programming languages. To address these challenges, we introduce Seed-Coder, a series of open-source LLMs comprising base, instruct and reasoning models of 8B size, minimizing human involvement in data construction. Our code pretraining data is produced by a model-centric data pipeline, which predominantly leverages LLMs for scoring and filtering code data. The instruct model is further trained via supervised fine-tuning and preference optimization, and the reasoning model leverages Long-Chain-of-Thought (LongCoT) reinforcement learning to improve multi-step code reasoning. Seed-Coder achieves state-of-the-art results among open-source models of similar size and even surpasses some much larger models, demonstrating superior performance in code generation, code completion, code editing, code reasoning, and software engineering tasks.

cs.CL↗
arXiv2025

Optical Wireless Communications: Enabling the Next Generation Network of Networks

Optical wireless communication (OWC) is a promising technology anticipated to play a key role in the next-generation network of networks, especially as a complementary technology to traditional radio frequency communications, for enhancing networking capabilities beyond conventional terrestrial networks. OWC is already a mature technology with diverse usage scenarios, and can enable integrated applications via wireless access and backhaul networks, dynamic drone and satellite networks, underwater networks, inter- and intra-system interconnecting networks, and vehicular communication networks. Furthermore, novel and emerging technological opportunities such as photovoltaic cells, orbital angular momentum-based modulation, optical reconfigurable intelligent surfaces, organic light-emitting and photo diodes, and recent advances in ultraviolet communications can help enhance future OWC capabilities even further. Moreover, OWC networks can also support value-added services such as enhanced positioning and gesture recognition. Hence, OWC provides unique functionalities that can play a crucial role in building convergent and resilient future network of networks alongside radio frequency and optical fiber technologies.

eess.SP↗
arXiv2025

A Vulnerability Code Intent Summary Dataset

In the era of Large Language Models (LLMs), the code summarization technique boosts a lot, along with the emergence of many new significant works. However, the potential of code summarization in the Computer Security Area still remains explored. Can we generate a code summary of a code snippet for its security intention? Thus, this work proposes an innovative large-scale multi-perspective Code Intent Summary Dataset named BADS , aiming to increase the understanding of a given code snippet and reduce the risk in the code developing process. The procedure of establishing a dataset can be divided into four steps: First, we collect samples of codes with known vulnerabilities as well as code generated by AI from multiple sources. Second, we do the data clean and format unification, then do the data combination. Third, we utilize the LLM to automatically Annotate the code snippet. Last, We do the human evaluation to double-check. The dataset contains X code examples which cover Y categories of vulnerability. Our data are from Z open-source projects and CVE entries, and compared to existing work, our dataset not only contains original code but also code function summary and security intent summary, providing context information for research in code security analysis. All information is in CSV format. The contributions of this paper are four-fold: the establishment of a high-quality, multi-perspective Code Intent Summary Dataset; an innovative method in data collection and processing; A new multi-perspective code analysis framework that promotes cross-disciplinary research in the fields of software engineering and cybersecurity; improving the practicality and scalability of the research outcomes by considering the code length limitations in real-world applications. Our dataset and related tools have been publicly released on GitHub.

cs.SE↗
arXiv2025

Velocity and Density Fluctuations in the Quiet Sun Corona

We investigate the properties and relationship between Doppler-velocity fluctuations and intensity fluctuations in the off-limb quiet Sun corona. These are expected to reflect the properties of Alfvenic and compressive waves, respectively. The data come from the Coronal Multichannel Polarimeter (COMP). These data were studied using spectral methods to estimate the power spectra, amplitudes, perpendicular correlation lengths, phases, trajectories, dispersion relations, and propagation speeds of both types of fluctuations. We find that most velocity fluctuations are due to Alfvenic waves, but that intensity fluctuations come from a variety of sources, likely including fast and slow mode waves, as well as aperiodic variations. The relation between the velocity and intensity fluctuations differs depending on the underlying coronal structure. On short closed loops, the velocity and intensity fluctuations have similar power spectra and speeds. In contrast, on longer nearly radial trajectories, the velocity and intensity fluctuations have different power spectra, with the velocity fluctuations propagating at much faster speeds than the intensity fluctuations. Considering the temperature sensitivity of COMP, these longer structures are more likely to be closed fields lines of the quiet Sun rather than cooler open field lines. That is, we find the character of the interactions of Alfvenic waves and density fluctuations depends on the length of the magnetic loop on which they are traveling.

astro-ph.SR↗