Selected work

Representative Papers

A Comparative Study of CNN Optimization Methods for Edge AI: Exploring the Role of Early Exits

Apr 16, 2026

Deploying deep neural networks on edge devices entails balancing accuracy, latency, and resource constraints. This work presents the first end-to-end hardware evaluation comparing static compression techniques—namely pruning and quantization—with dynamic early-exit mechanisms, all implemented within a unified ONNX inference framework. Experimental results demonstrate that static methods substantially reduce memory footprint, while early-exit strategies achieve input-adaptive computational savings. Crucially, combining both approaches yields simultaneous reductions in both latency and memory consumption with negligible accuracy loss, revealing their complementary nature and significant potential for joint optimization in edge computing scenarios.

0 citationsRead paper

Beyond the Beta Lorenz Curve: A New Parametric Family for Poverty and Inequality Estimation

Apr 01, 2026

This study addresses a critical flaw in the existing Beta Lorenz curve, whose parameter space fails to satisfy the theoretical constraints inherent to Lorenz curves, leading to systematic bias in estimating poverty and inequality from grouped income data. The authors explicitly identify this deficiency for the first time and propose a novel four-parameter family of Lorenz curves that rigorously adheres to all formal properties of genuine Lorenz curves while retaining practical usability. Through parametric modeling, derivation of necessary constraints, and extensive empirical validation across more than 2,000 datasets, the new model demonstrates superior performance in estimating poverty and inequality metrics. Specifically, it significantly reduces the systematic underestimation of poverty levels observed in over 80% of cases compared to the widely used General Quadratic (GQ) Lorenz curve.

0 citationsRead paper

Correlation Bounds and Markov Analysis for Ring-Oscillator TRNGs: A Joint Validation Framework

Mar 13, 2026

This work addresses the absence of a unified framework that integrates theoretically grounded correlation measures with empirical entropy tests for comprehensively evaluating the cryptographic quality of ring oscillator true random number generators (TRNGs). The study proposes the first joint validation framework, revealing a strong positive correlation between the Z-score of Maurer’s universal statistical test and the second-order correlation measure \( C_2 \), and establishing their mathematical connection to higher-order Markov chain transition probabilities. By integrating the Mauduit–Sárközy correlation measure, Maurer’s test, and Markov modeling, the approach is empirically validated on the OpenTRNG platform, demonstrating that practical TRNG implementations can achieve the Schmidt-improved bound. This provides a concise and reliable unified benchmark for TRNG design and evaluation.

0 citationsRead paper

Edge Deployment of Small Language Models, a comprehensive comparison of CPU, GPU and NPU backends

Nov 27, 2025

Selecting optimal hardware for deploying small language models (SLMs) in resource-constrained edge computing environments remains challenging due to the lack of systematic, cross-architecture performance and efficiency comparisons. Method: This work conducts the first unified, empirical evaluation—within a consistent experimental framework—of Intel/ARM CPUs, NVIDIA GPUs, and the RaiderChip NPU across mainstream SLMs. We introduce bandwidth-normalized analysis and adopt holistic metrics—including Energy-Delay Product (EDP)—to jointly quantify inference throughput, latency, and energy efficiency. Results: Dedicated NPUs achieve substantially higher throughput and 1–2 orders-of-magnitude lower EDP than general-purpose CPUs. While low-power ARM CPUs exhibit limited raw performance, they demonstrate competitive inference-per-watt efficiency. Our reproducible benchmarking methodology and empirical findings provide actionable, evidence-based guidance for hardware selection in edge-deployed SLM applications.

0 citationsRead paper

Fast and accurate computation of classical Gaussian quadratures

Sep 20, 2025

This paper addresses the efficient and high-precision computation of classical Gaussian quadrature nodes and weights—including Gauss–Jacobi, Gauss–Laguerre, Gauss–Hermite, and their Radau/Lobatto variants. We propose an adaptive hybrid strategy: a globally convergent, fourth-order Newton-type iteration is employed in well-conditioned parameter regions, while highly accurate asymptotic approximations are activated for ill-conditioned or large-parameter regimes. Additionally, we derive and implement a numerically stable algorithm for computing barycentric weights. The method supports arbitrary-precision arithmetic and, for the first time, systematically covers parameter regimes where conventional algorithms fail—particularly in symmetric cases such as Gauss–Gegenbauer and Gauss–Hermite quadratures. Experimental results demonstrate superior numerical stability, computational efficiency, and full-parameter robustness compared to state-of-the-art implementations.

0 citationsRead paper

Recent publications

Latest Papers

A Comparative Study of CNN Optimization Methods for Edge AI: Exploring the Role of Early Exits

Apr 16, 2026

Deploying deep neural networks on edge devices entails balancing accuracy, latency, and resource constraints. This work presents the first end-to-end hardware evaluation comparing static compression techniques—namely pruning and quantization—with dynamic early-exit mechanisms, all implemented within a unified ONNX inference framework. Experimental results demonstrate that static methods substantially reduce memory footprint, while early-exit strategies achieve input-adaptive computational savings. Crucially, combining both approaches yields simultaneous reductions in both latency and memory consumption with negligible accuracy loss, revealing their complementary nature and significant potential for joint optimization in edge computing scenarios.

0 citationsRead paper

Beyond the Beta Lorenz Curve: A New Parametric Family for Poverty and Inequality Estimation

Apr 01, 2026

This study addresses a critical flaw in the existing Beta Lorenz curve, whose parameter space fails to satisfy the theoretical constraints inherent to Lorenz curves, leading to systematic bias in estimating poverty and inequality from grouped income data. The authors explicitly identify this deficiency for the first time and propose a novel four-parameter family of Lorenz curves that rigorously adheres to all formal properties of genuine Lorenz curves while retaining practical usability. Through parametric modeling, derivation of necessary constraints, and extensive empirical validation across more than 2,000 datasets, the new model demonstrates superior performance in estimating poverty and inequality metrics. Specifically, it significantly reduces the systematic underestimation of poverty levels observed in over 80% of cases compared to the widely used General Quadratic (GQ) Lorenz curve.

0 citationsRead paper

Correlation Bounds and Markov Analysis for Ring-Oscillator TRNGs: A Joint Validation Framework

Mar 13, 2026

This work addresses the absence of a unified framework that integrates theoretically grounded correlation measures with empirical entropy tests for comprehensively evaluating the cryptographic quality of ring oscillator true random number generators (TRNGs). The study proposes the first joint validation framework, revealing a strong positive correlation between the Z-score of Maurer’s universal statistical test and the second-order correlation measure \( C_2 \), and establishing their mathematical connection to higher-order Markov chain transition probabilities. By integrating the Mauduit–Sárközy correlation measure, Maurer’s test, and Markov modeling, the approach is empirically validated on the OpenTRNG platform, demonstrating that practical TRNG implementations can achieve the Schmidt-improved bound. This provides a concise and reliable unified benchmark for TRNG design and evaluation.

0 citationsRead paper

Edge Deployment of Small Language Models, a comprehensive comparison of CPU, GPU and NPU backends

Nov 27, 2025

Selecting optimal hardware for deploying small language models (SLMs) in resource-constrained edge computing environments remains challenging due to the lack of systematic, cross-architecture performance and efficiency comparisons. Method: This work conducts the first unified, empirical evaluation—within a consistent experimental framework—of Intel/ARM CPUs, NVIDIA GPUs, and the RaiderChip NPU across mainstream SLMs. We introduce bandwidth-normalized analysis and adopt holistic metrics—including Energy-Delay Product (EDP)—to jointly quantify inference throughput, latency, and energy efficiency. Results: Dedicated NPUs achieve substantially higher throughput and 1–2 orders-of-magnitude lower EDP than general-purpose CPUs. While low-power ARM CPUs exhibit limited raw performance, they demonstrate competitive inference-per-watt efficiency. Our reproducible benchmarking methodology and empirical findings provide actionable, evidence-based guidance for hardware selection in edge-deployed SLM applications.

0 citationsRead paper

Fast and accurate computation of classical Gaussian quadratures

Sep 20, 2025

This paper addresses the efficient and high-precision computation of classical Gaussian quadrature nodes and weights—including Gauss–Jacobi, Gauss–Laguerre, Gauss–Hermite, and their Radau/Lobatto variants. We propose an adaptive hybrid strategy: a globally convergent, fourth-order Newton-type iteration is employed in well-conditioned parameter regions, while highly accurate asymptotic approximations are activated for ill-conditioned or large-parameter regimes. Additionally, we derive and implement a numerically stable algorithm for computing barycentric weights. The method supports arbitrary-precision arithmetic and, for the first time, systematically covers parameter regimes where conventional algorithms fail—particularly in symmetric cases such as Gauss–Gegenbauer and Gauss–Hermite quadratures. Experimental results demonstrate superior numerical stability, computational efficiency, and full-parameter robustness compared to state-of-the-art implementations.

0 citationsRead paper