arXiv subjects

Zhixin Pan

Publications and source records attributed to Zhixin Pan.

14 recordsLinked to original sources

arXiv2026

HISA: Efficient Hierarchical Indexing for Fine-Grained Sparse Attention

Token-level sparse attention mechanisms, exemplified by DeepSeek Sparse Attention (DSA), achieve fine-grained key selection by scoring every historical key for each query through a lightweight indexer, then computing attention only on the selected subset. While the downstream sparse attention itself scales favorably, the indexer must still scan the entire prefix for every query, introducing an per-layer bottleneck that grows prohibitively with context length. We propose HISA (Hierarchical Indexed Sparse Attention), a plug-and-play replacement for the indexer that rewrites the search path from a flat token scan into a two-stage hierarchical procedure: (1) a block-level coarse filtering stage that scores pooled block representations to discard irrelevant regions, followed by (2) a token-level refinement stage that applies the original indexer exclusively within the retained candidate blocks. HISA preserves the identical token-level top-sparse pattern consumed by the downstream Sparse MLA operator and requires no additional training. On kernel-level benchmarks, HISA achieves up to speedup at 64K context. On Needle-in-a-Haystack and LongBench, we directly replace the indexer in DeepSeek-V3.2 and GLM-5 with our HISA indexer, without any finetuning. HISA closely matches the original DSA in quality, while substantially outperforming block-sparse baselines.

cs.LG↗

arXiv2026

GQLA: Group-Query Latent Attention for Hardware-Adaptive Decoding

Decoding efficiency depends on KV-cache traffic, computation, and parallelism. Multi-head Latent Attention (MLA) offers absorbed and head-expanded forms, but its head-specific projections limit grouped KV sharing. We introduce Group-Query Latent Attention (GQLA), whose group-shared up-projections preserve a joint latent and enable a shardable expanded cache. The same weights support equivalent MQA-absorb and GQA execution using existing kernels. Roofline analysis relates path preference to GPU compute-to-bandwidth ratio and tensor parallelism (TP): absorption trades traffic for computation and replicates the latent across ranks; GQA shards KV groups. Pretraining with eight query heads per group yields terminal training losses close to MLA's across the tested model scales at matched tokens. TransGQLA converts GQA and MLA checkpoints using calibration data. Without gradient updates, converted DeepSeek-V3.1-Base achieves 77.79% eight-task accuracy versus the source's 77.84%, while WikiText-2 perplexity rises from 3.21 to 6.05. On eight H20-3e GPUs, its GQA path improves output throughput over MQA-absorb by 22.42--34.97% at matched resident batches across 8K--64K inputs. Its smaller per-rank cache permits larger batches, giving 50.07% in a separate capacity-tuned comparison. These results support multiple execution paths as an architectural means of adapting decoding to hardware and parallelism.

cs.LG↗

arXiv2026

When Is RTL Better? Auditing Evaluation Stability with Tool-Grounded Requirement Profiles

RTL-generation studies increasingly report functional correctness and post-synthesis quality, yet the evaluator itself is rarely treated as an experimental variable. A candidate can appear better when a security rule, protocol obligation, target context, feasibility gate, uncertainty policy, or provenance constraint is omitted. We introduce tool-grounded requirement profiles: machine-readable contracts binding each requirement's meaning to its executable evidence and comparison rule, together with cross-requirement gates, uncertainty handling, and provenance. We audit evaluator stability by freezing candidate pairs and perturbing only their evaluation profile. In a preregistered controlled pilot with six designs, twelve pairs, three requirement families, six evaluator conditions, and 72 decisions, a generic cell-count evaluator reports unsupported improvement for 6/12 pairs (50%; Wilson 95% interval 25.4 to 74.6%; design-cluster bootstrap interval 25 to 75%). The matched evaluator detects all twelve injected regressions, whereas syntax, generic synthesis, context mismatch, and gate ablation each lose the required distinction for 12/12 pairs. In a separately frozen external validation of all 290 public RTLLM v1.1 text-only candidates, 81/215 evaluable comparisons (37.7%; task-cluster 95% interval 24.8 to 51.4%) yield unsupported scalar improvements, including 72 functional-gate reversals. Separately, 26 OpenROAD-flow-scripts runs show why timing context and activity qualification must be part of the contract. The study does not rank generators; it establishes evaluator configuration as a measurable, auditable source of conclusion instability.

cs.AR↗

arXiv2026

Hardware-Assisted Multi-Stage Ransomware Detection with Adaptive Machine Learning

Ransomware is a fast-escalating privacy threat causing significant financial and data losses worldwide. As attacks grow more sophisticated, traditional static signatures and dynamic behavior analysis struggle to defend. In this paper, we propose an novel framework that couples software-level scanning with hardware-level microprocessor activity monitoring to achieve efficient ransomware detection and mitigation. The contributions are threefold: (1) we employ adversarial training to harden static analysis against obfuscation and leverage hardware-assisted monitoring for real-time detection; (2) we integrate neural architecture search (NAS) to automatically tailor models for stronger generalizability; and (3) We validate the proposed approach through extensive experiments, demonstrating substantially improved detection accuracy and reduced latency compared with existing methods. Moreover, it provides trace-level interpretability that explains the underlying decision process.

cs.CR↗

arXiv2026

LatentIndex: Cross-Layer Sharing with Layer-Specific Selection for Sparse Attention

Sparse attention reduces core-attention computation, but its indexers still incur repeated selection work and per-layer key-cache storage. Reusing selected indices across layers reduces this overhead but constrains multiple layers to the same token set. We introduce LatentIndex, which extends the latent-sharing principle of Multi-head Latent Attention across indexer layers. Each layer group constructs a shared latent cache from its first layer's hidden states, while layer-specific scoring enables independent token selection. Absorbing key decoders into queries enables direct scoring of the shared cache without reconstructing historical per-layer keys. We develop training-free calibration and investigate a training-aware instantiation of this principle. To balance quality and computation, a hierarchical selection (HS) variant lets followers independently refine a shared candidate set proposed by the anchor. With four-layer sharing, LatentIndex reduces logical indexer-cache storage by 61.1% on DeepSeek-V3.2. Across DeepSeek-V3.2 and GLM-5, training-free LatentIndex improves head-wise attention-mass recall over IndexCache by up to 3.28 percentage points while maintaining RULER and LongBench performance close to native DSA. HS further achieves 2.30-2.72 times decode indexer speedups over DSA across 8K-128K contexts, retaining most of LatentIndex's recall. LatentIndex offers a new perspective on cross-layer indexing: sharing continuous representations rather than discrete selections enables efficient reuse while preserving layer-specific token selection.

cs.AI↗

arXiv2026

CARB: A Characterization-Guided Framework for CNN Inference Cost Prediction and Deployment Screening

Accurate pre-deployment estimation of CNN inference cost--energy, latency, and peak memory--is increasingly critical as models are deployed on resource-constrained GPU platforms. Existing approaches rely on FLOPs, latency measurements, or single-device profiling as energy proxies, overlooking the non-linear interactions between architectural design and hardware load. We present a workload characterization study of 13 419 CNN configurations on two GPU platforms (RTX 5090 and RTX 3080) under GPU telemetry, revealing that energy, latency, and memory exhibit fundamentally distinct scaling behaviors: energy and latency diverge by 3x under high computational demand, and cross-GPU transferability differs by target--energy and latency require platform-specific models while memory transfers well across the two tested platforms. Building on these characterization findings, we develop CARB, a cascade-blended ensemble that jointly predicts all three targets with R2 ~0.99, and a two-stage deployment screening workflow that eliminates over 90% of candidates in seconds, reducing large design spaces to a Pareto-prioritized shortlist validated against real hardware.

cs.AR↗

arXiv2025

Towards Low-Latency and Adaptive Ransomware Detection Using Contrastive Learning

Ransomware has become a critical threat to cybersecurity due to its rapid evolution, the necessity for early detection, and growing diversity, posing significant challenges to traditional detection methods. While AI-based approaches had been proposed by prior works to assist ransomware detection, existing methods suffer from three major limitations, ad-hoc feature dependencies, delayed response, and limited adaptability to unseen variants. In this paper, we propose a framework that integrates self-supervised contrastive learning with neural architecture search (NAS) to address these challenges. Specifically, this paper offers three important contributions. (1) We design a contrastive learning framework that incorporates hardware performance counters (HPC) to analyze the runtime behavior of target ransomware. (2) We introduce a customized loss function that encourages early-stage detection of malicious activity, and significantly reduces the detection latency. (3) We deploy a neural architecture search (NAS) framework to automatically construct adaptive model architectures, allowing the detector to flexibly align with unseen ransomware variants. Experimental results show that our proposed method achieves significant improvements in both detection accuracy (up to 16.1%) and response time (up to 6x) compared to existing approaches while maintaining robustness under evasive attacks.

cs.CR↗

arXiv2025

SAND: A Self-supervised and Adaptive NAS-Driven Framework for Hardware Trojan Detection

The globalized semiconductor supply chain has made Hardware Trojans (HT) a significant security threat to embedded systems, necessitating the design of efficient and adaptable detection mechanisms. Despite promising machine learning-based HT detection techniques in the literature, they suffer from ad hoc feature selection and the lack of adaptivity, all of which hinder their effectiveness across diverse HT attacks. In this paper, we propose SAND, a selfsupervised and adaptive NAS-driven framework for efficient HT detection. Specifically, this paper makes three key contributions. (1) We leverage self-supervised learning (SSL) to enable automated feature extraction, eliminating the dependency on manually engineered features. (2) SAND integrates neural architecture search (NAS) to dynamically optimize the downstream classifier, allowing for seamless adaptation to unseen benchmarks with minimal fine-tuning. (3) Experimental results show that SAND achieves a significant improvement in detection accuracy (up to 18.3%) over state-of-the-art methods, exhibits high resilience against evasive Trojans, and demonstrates strong generalization.

cs.CR↗

arXiv2024

Privacy-Preserving Debiasing using Data Augmentation and Machine Unlearning

Data augmentation is widely used to mitigate data bias in the training dataset. However, data augmentation exposes machine learning models to privacy attacks, such as membership inference attacks. In this paper, we propose an effective combination of data augmentation and machine unlearning, which can reduce data bias while providing a provable defense against known attacks. Specifically, we maintain the fairness of the trained model with diffusion-based data augmentation, and then utilize multi-shard unlearning to remove identifying information of original data from the ML model for protection against privacy attacks. Experimental evaluation across diverse datasets demonstrates that our approach can achieve significant improvements in bias reduction as well as robustness against state-of-the-art privacy attacks.

cs.LG↗

arXiv2024

SDIP: Self-Reinforcement Deep Image Prior Framework for Image Processing

Deep image prior (DIP) proposed in recent research has revealed the inherent trait of convolutional neural networks (CNN) for capturing substantial low-level image statistics priors. This framework efficiently addresses the inverse problems in image processing and has induced extensive applications in various domains. However, as the whole algorithm is initialized randomly, the DIP algorithm often lacks stability. Thus, this method still has space for further improvement. In this paper, we propose the self-reinforcement deep image prior (SDIP) as an improved version of the original DIP. We observed that the changes in the DIP networks' input and output are highly correlated during each iteration. SDIP efficiently utilizes this trait in a reinforcement learning manner, where the current iteration's output is utilized by a steering algorithm to update the network input for the next iteration, guiding the algorithm toward improved results. Experimental results across multiple applications demonstrate that our proposed SDIP framework offers improvement compared to the original DIP method and other state-of-the-art methods.

cs.CV↗

arXiv2023

Hardware Acceleration of Explainable Artificial Intelligence

Machine learning (ML) is successful in achieving human-level artificial intelligence in various fields. However, it lacks the ability to explain an outcome due to its black-box nature. While recent efforts on explainable AI (XAI) has received significant attention, most of the existing solutions are not applicable in real-time systems since they map interpretability as an optimization problem, which leads to numerous iterations of time-consuming complex computations. Although there are existing hardware-based acceleration framework for XAI, they are implemented through FPGA and designed for specific tasks, leading to expensive cost and lack of flexibility. In this paper, we propose a simple yet efficient framework to accelerate various XAI algorithms with existing hardware accelerators. Specifically, this paper makes three important contributions. (1) The proposed method is the first attempt in exploring the effectiveness of Tensor Processing Unit (TPU) to accelerate XAI. (2) Our proposed solution explores the close relationship between several existing XAI algorithms with matrix computations, and exploits the synergy between convolution and Fourier transform, which takes full advantage of TPU's inherent ability in accelerating matrix computations. (3) Our proposed approach can lead to real-time outcome interpretation. Extensive experimental evaluation demonstrates that proposed approach deployed on TPU can provide drastic improvement in interpretation time (39x on average) as well as energy efficiency (69x on average) compared to existing acceleration techniques.

cs.LG↗

arXiv2022

Backdoor Attacks on Bayesian Neural Networks using Reverse Distribution

Due to cost and time-to-market constraints, many industries outsource the training process of machine learning models (ML) to third-party cloud service providers, popularly known as ML-asa-Service (MLaaS). MLaaS creates opportunity for an adversary to provide users with backdoored ML models to produce incorrect predictions only in extremely rare (attacker-chosen) scenarios. Bayesian neural networks (BNN) are inherently immune against backdoor attacks since the weights are designed to be marginal distributions to quantify the uncertainty. In this paper, we propose a novel backdoor attack based on effective learning and targeted utilization of reverse distribution. This paper makes three important contributions. (1) To the best of our knowledge, this is the first backdoor attack that can effectively break the robustness of BNNs. (2) We produce reverse distributions to cancel the original distributions when the trigger is activated. (3) We propose an efficient solution for merging probability distributions in BNNs. Experimental results on diverse benchmark datasets demonstrate that our proposed attack can achieve the attack success rate (ASR) of 100%, while the ASR of the state-of-the-art attacks is lower than 60%.

cs.CR↗

arXiv2021

Hardware Acceleration of Explainable Machine Learning using Tensor Processing Units

Machine learning (ML) is successful in achieving human-level performance in various fields. However, it lacks the ability to explain an outcome due to its black-box nature. While existing explainable ML is promising, almost all of these methods focus on formatting interpretability as an optimization problem. Such a mapping leads to numerous iterations of time-consuming complex computations, which limits their applicability in real-time applications. In this paper, we propose a novel framework for accelerating explainable ML using Tensor Processing Units (TPUs). The proposed framework exploits the synergy between matrix convolution and Fourier transform, and takes full advantage of TPU's natural ability in accelerating matrix computations. Specifically, this paper makes three important contributions. (1) To the best of our knowledge, our proposed work is the first attempt in enabling hardware acceleration of explainable ML using TPUs. (2) Our proposed approach is applicable across a wide variety of ML algorithms, and effective utilization of TPU-based acceleration can lead to real-time outcome interpretation. (3) Extensive experimental results demonstrate that our proposed approach can provide an order-of-magnitude speedup in both classification time (25x on average) and interpretation time (13x on average) compared to state-of-the-art techniques.

cs.LG↗

arXiv2021

Fast Approximate Spectral Normalization for Robust Deep Neural Networks

Deep neural networks (DNNs) play an important role in machine learning due to its outstanding performance compared to other alternatives. However, DNNs are not suitable for safety-critical applications since DNNs can be easily fooled by well-crafted adversarial examples. One promising strategy to counter adversarial attacks is to utilize spectral normalization, which ensures that the trained model has low sensitivity towards the disturbance of input samples. Unfortunately, this strategy requires exact computation of spectral norm, which is computation intensive and impractical for large-scale networks. In this paper, we introduce an approximate algorithm for spectral normalization based on Fourier transform and layer separation. The primary contribution of our work is to effectively combine the sparsity of weight matrix and decomposability of convolution layers. Extensive experimental evaluation demonstrates that our framework is able to significantly improve both time efficiency (up to 60\%) and model robustness (61\% on average) compared with the state-of-the-art spectral normalization.

cs.LG↗