HANSARD: A Reference Architecture for Forensic Readiness, Runtime Witnessing, and Graded Attribution in Autonomous Multi-Agent AI Systems
为解决自主多代理AI系统中的责任归属问题，提出HANSARD架构，通过准备性配置、关键点监控、因果图构建等方法实现可追溯性和责任分配。
0 citationsRead paper
Selected work
Representative Papers
AmalthAI: An Open-Source Computer Vision Platform for Cultural Heritage
This work addresses the limited machine learning expertise among cultural heritage specialists, which hinders their ability to independently conduct vision-based archaeological analysis. To bridge this gap, we present the first end-to-end, self-hosted computer vision platform tailored to this domain, enabling non-technical users to manage data, train models, and perform inference for classification, segmentation, and object detection tasks—all while keeping sensitive data within institutional boundaries. The system integrates Grad-CAM for prediction visualization and leverages a vision-language model to generate explanatory textual descriptions, enhancing interpretability. Built on Kubeflow and Katib, the backend supports scalable training and automated hyperparameter optimization. Experiments on a newly curated dataset of pottery textile impressions demonstrate the platform’s effectiveness, empowering domain experts to autonomously test hypotheses and produce archaeologically meaningful insights.
0 citationsRead paper
Energy Efficient AI-Enabled Wireless Sensor Networks for Mission Critical Environments: A Systematic Review across Smart Grid, AI, and Urban Infrastructure Applications
This study addresses the challenge of jointly optimizing energy efficiency, reliability, low latency, and security in mission-critical wireless sensor networks (WSNs), a problem often approached in isolation by existing research. Through a systematic review of 50 high-quality studies published between 2023 and 2026, the work employs qualitative thematic coding and comparative analysis to examine the application of reinforcement learning, fuzzy logic, metaheuristics, and AI-driven security techniques in routing, clustering, and edge computing. The paper proposes a novel, lightweight, interpretable, and field-validated AI-driven paradigm for WSNs that emphasizes multi-objective co-design. Findings demonstrate that AI significantly enhances both energy efficiency and overall system performance, offering robust theoretical foundations and practical guidance for the architecture of future mission-critical systems.
0 citationsRead paper
Visual Place Recognition Using Rate-Encoded Spiking Neural Networks with Discrete STDP Learning
This work addresses the challenge that existing spike-timing-dependent plasticity (STDP)-based spiking neural networks (SNNs) struggle to achieve high recall under 100% precision in visual place recognition. The authors propose a tensor-native, discretized STDP-SNN pipeline incorporating closed-form deterministic tensor neuron assignment, a post-query state reset mechanism, and a velocity-compensated sliding-window frame aggregation strategy. By integrating rate coding, unsupervised STDP learning, and an efficient inference architecture, the method achieves 100.00% recall at 100% precision (R@100P) on the Nordland dataset under constant-velocity conditions, with only 0.20 milliseconds of latency, substantially improving recall performance under stringent accuracy requirements.
0 citationsRead paper
MIVE: A Minimalist Integer Vector Engine for Softmax LayerNorm and RMSNorm Acceleration
This work addresses the hardware inefficiency in large language model inference caused by nonlinear normalization operations—such as LayerNorm, RMSNorm, and Softmax—which typically rely on dedicated hardware modules, leading to resource redundancy and excessive silicon area consumption. To overcome this limitation, the authors propose MIVE (Minimalist Integer Vector Engine), a unified programmable architecture that integrates all three normalization functions into a single design. By leveraging a shared data path, integer arithmetic, and reusable computation patterns, MIVE enables extensive hardware resource sharing across these operations. ASIC implementation results demonstrate that MIVE achieves significantly improved area efficiency and energy efficiency while supporting multifunctional normalization, outperforming existing specialized accelerators.
0 citationsRead paper
Recent publications
Latest Papers
HANSARD: A Reference Architecture for Forensic Readiness, Runtime Witnessing, and Graded Attribution in Autonomous Multi-Agent AI Systems
为解决自主多代理AI系统中的责任归属问题，提出HANSARD架构，通过准备性配置、关键点监控、因果图构建等方法实现可追溯性和责任分配。
0 citationsRead paper
AmalthAI: An Open-Source Computer Vision Platform for Cultural Heritage
This work addresses the limited machine learning expertise among cultural heritage specialists, which hinders their ability to independently conduct vision-based archaeological analysis. To bridge this gap, we present the first end-to-end, self-hosted computer vision platform tailored to this domain, enabling non-technical users to manage data, train models, and perform inference for classification, segmentation, and object detection tasks—all while keeping sensitive data within institutional boundaries. The system integrates Grad-CAM for prediction visualization and leverages a vision-language model to generate explanatory textual descriptions, enhancing interpretability. Built on Kubeflow and Katib, the backend supports scalable training and automated hyperparameter optimization. Experiments on a newly curated dataset of pottery textile impressions demonstrate the platform’s effectiveness, empowering domain experts to autonomously test hypotheses and produce archaeologically meaningful insights.
0 citationsRead paper
Energy Efficient AI-Enabled Wireless Sensor Networks for Mission Critical Environments: A Systematic Review across Smart Grid, AI, and Urban Infrastructure Applications
This study addresses the challenge of jointly optimizing energy efficiency, reliability, low latency, and security in mission-critical wireless sensor networks (WSNs), a problem often approached in isolation by existing research. Through a systematic review of 50 high-quality studies published between 2023 and 2026, the work employs qualitative thematic coding and comparative analysis to examine the application of reinforcement learning, fuzzy logic, metaheuristics, and AI-driven security techniques in routing, clustering, and edge computing. The paper proposes a novel, lightweight, interpretable, and field-validated AI-driven paradigm for WSNs that emphasizes multi-objective co-design. Findings demonstrate that AI significantly enhances both energy efficiency and overall system performance, offering robust theoretical foundations and practical guidance for the architecture of future mission-critical systems.
0 citationsRead paper
Visual Place Recognition Using Rate-Encoded Spiking Neural Networks with Discrete STDP Learning
This work addresses the challenge that existing spike-timing-dependent plasticity (STDP)-based spiking neural networks (SNNs) struggle to achieve high recall under 100% precision in visual place recognition. The authors propose a tensor-native, discretized STDP-SNN pipeline incorporating closed-form deterministic tensor neuron assignment, a post-query state reset mechanism, and a velocity-compensated sliding-window frame aggregation strategy. By integrating rate coding, unsupervised STDP learning, and an efficient inference architecture, the method achieves 100.00% recall at 100% precision (R@100P) on the Nordland dataset under constant-velocity conditions, with only 0.20 milliseconds of latency, substantially improving recall performance under stringent accuracy requirements.
0 citationsRead paper
MIVE: A Minimalist Integer Vector Engine for Softmax LayerNorm and RMSNorm Acceleration
This work addresses the hardware inefficiency in large language model inference caused by nonlinear normalization operations—such as LayerNorm, RMSNorm, and Softmax—which typically rely on dedicated hardware modules, leading to resource redundancy and excessive silicon area consumption. To overcome this limitation, the authors propose MIVE (Minimalist Integer Vector Engine), a unified programmable architecture that integrates all three normalization functions into a single design. By leveraging a shared data path, integer arithmetic, and reusable computation patterns, MIVE enables extensive hardware resource sharing across these operations. ASIC implementation results demonstrate that MIVE achieves significantly improved area efficiency and energy efficiency while supporting multifunctional normalization, outperforming existing specialized accelerators.
0 citationsRead paper