Institution profile

Meiji University

Academic institutionasia · jp
Official website
Research library49linked papers
Opportunities0open roles
Selected work

Representative Papers

Using Stochastic Gradient Descent to Smooth Nonconvex Functions: Analysis of Implicit Graduated Optimization with Optimal Noise Scheduling

Nov 15, 2023arXiv.org

This work addresses the poorly understood implicit smoothing mechanism of stochastic gradient descent (SGD) in non-convex optimization. We provide the first theoretical characterization of how the SGD noise—determined jointly by learning rate, batch size, and gradient variance—induces a quantifiable smoothing effect on the objective function, and establish intrinsic links among smoothness, sharpness, and generalization performance. We propose a progressive optimization algorithm featuring joint scheduling of learning rate and batch size to dynamically modulate noise intensity for optimal implicit regularization. Empirical validation on ResNet-based image classification demonstrates significantly improved convergence stability and test accuracy; moreover, smoothness strongly correlates with generalization accuracy. Our core contributions are threefold: (i) an explicit, interpretable smoothing interpretation of SGD noise; (ii) a sharpness-driven theoretical framework for generalization; and (iii) the first noise-adaptive optimization paradigm with rigorous theoretical justification.

4 citationsRead paper

Improved Convergence Rates of Muon Optimizer for Nonconvex Optimization

Jan 27, 2026

This work addresses the limitations of existing convergence analyses for Muon optimizers in non-convex optimization, which often rely on strong assumptions and yield coarse bounds. We propose a streamlined analytical framework that dispenses with restrictive conditions on the update rule and directly characterizes convergence behavior under general non-convex settings. Our approach significantly tightens the upper bound on the convergence rate and applies to a broader class of non-convex optimization problems. Furthermore, it delivers a unified and refined theoretical guarantee for first-order methods based on orthogonalization, enhancing both the generality and precision of convergence analysis in this domain.

1 citations1 influentialRead paper

Spatio-temporal smoothing, interpolation and prediction of income distributions based on grouped data

Jul 18, 2022

Japanese municipal-level household income data—derived from the Household Labour Survey (HLS)—suffer from severe limitations: coarse income grouping, incomplete spatial coverage, and low temporal frequency (only quinquennial). These constraints impede evidence-based local policymaking. To address this, we propose the Spatio-Temporal Finite Mixture Model (ST-FMM), the first framework to jointly capture regional heterogeneity and dynamic evolution via a “shared latent income distribution + spatio-temporally varying mixture proportions” mechanism. ST-FMM integrates grouped-data likelihood modeling, EM-based parameter estimation, and a Bayesian smoothing–prediction framework to impute missing municipalities, smooth distributional estimates, and forecast future time points. The model generates complete, high-resolution municipal-level maps of income distributions and poverty metrics. Empirical results demonstrate substantial improvements in spatial coverage, distributional fidelity, and temporal responsiveness—enabling faster, more precise, and geographically targeted policy interventions.

1 citationsRead paper

A Regulatory Placebo? The Systemic Failure of Mandatory GenAI Labeling

Aug 17, 2026

This study addresses the concern that mandatory labeling for generative AI has devolved into a "regulatory placebo" that impedes technological advancement. By integrating technical feasibility analysis, regulatory theory critique, and comparative paradigm research, this work systematically deconstructs three fundamental theoretical deficiencies in existing frameworks. The findings confirm that mandatory labeling entails significant implementation challenges and technical risks, prompting the proposal of a novel paradigm shift from "identity labeling" to "content governance." This research transcends traditional regulatory cognitive limitations by establishing content governance as the theoretically superior direction for GenAI oversight. Ultimately, it provides essential scholarly support and practical guidance for constructing a substantive governance system tailored to the generative AI era.

0 citationsRead paper
Recent publications

Latest Papers

A Regulatory Placebo? The Systemic Failure of Mandatory GenAI Labeling

Aug 17, 2026

This study addresses the concern that mandatory labeling for generative AI has devolved into a "regulatory placebo" that impedes technological advancement. By integrating technical feasibility analysis, regulatory theory critique, and comparative paradigm research, this work systematically deconstructs three fundamental theoretical deficiencies in existing frameworks. The findings confirm that mandatory labeling entails significant implementation challenges and technical risks, prompting the proposal of a novel paradigm shift from "identity labeling" to "content governance." This research transcends traditional regulatory cognitive limitations by establishing content governance as the theoretically superior direction for GenAI oversight. Ultimately, it provides essential scholarly support and practical guidance for constructing a substantive governance system tailored to the generative AI era.

0 citationsRead paper

Exploratory Integration of EEG Spectral Features and Gaze Variability for Mild Cognitive Impairment Discrimination

Jul 31, 2026

Early identification of mild cognitive impairment (MCI) is crucial in aging societies, yet unimodal approaches often exhibit limited discriminative power. This study systematically investigates, for the first time, the complementary value of electroencephalographic (EEG) spectral features and eye movement variability in MCI classification. EEG signals were recorded using the 10–20 system, and spectral power features were extracted and refined via LASSO regularization for feature selection, then integrated with eye movement variability metrics. Experimental results demonstrate that raw high-dimensional EEG features yield an AUC of 0.52, which improves to 0.64 after LASSO-based selection; notably, multimodal fusion with eye movement variability further elevates the AUC to 0.78, substantiating the efficacy and added benefit of integrating complementary neurophysiological and oculomotor biomarkers for MCI detection.

0 citationsRead paper

Vanilla SGD with Momentum Survives Heavy-Tailed Noise: Convergence Analysis without Gradient Clipping or Normalization

Jul 09, 2026

This work investigates the convergence properties of standard stochastic gradient descent (SGD) and its momentum variant under heavy-tailed gradient noise, without resorting to stabilization techniques such as gradient clipping or normalization. For strongly convex, convex, and non-convex objective functions, it establishes the first comprehensive theoretical convergence guarantees for vanilla SGD with momentum in the presence of heavy-tailed noise. Leveraging probabilistic inequalities and tools from stochastic optimization analysis, the study demonstrates that while the method does converge, its convergence rate is significantly slower compared to variants employing gradient clipping or normalization. These theoretical findings are corroborated by experiments on synthetic functions, revealing the inherent limitations of momentum SGD in heavy-tailed stochastic environments.

0 citationsRead paper

Profit-Based Counterfactual Explanations for Product Improvement: A Case Study of Manga Sales in Japan

Jul 01, 2026

This study addresses the limitations of existing counterfactual explanation methods in regression tasks, which rely on exogenous target values and distance functions lacking economic interpretability, thereby hindering profit-driven decision-making. To overcome this, the paper proposes a Profit-Based Counterfactual Explanation (PBCE) framework that directly optimizes profit as the objective and models feature modification costs as a distance term with clear economic meaning, eliminating the need for manually specified target values. By integrating predictive models with cost constraints, PBCE employs optimization algorithms to generate actionable product improvement recommendations. Empirical evaluation on a Japanese manga sales dataset demonstrates that counterfactuals produced by PBCE significantly enhance profitability while maintaining practical feasibility, outperforming conventional target-oriented approaches.

0 citationsRead paper