Institution profile

Osaka University

Academic institutionasia · jp
Official website
Research library486linked papers
Opportunities0open roles
Selected work

Representative Papers

MIDAS: Mixing Ambiguous Data with Soft Labels for Dynamic Facial Expression Recognition

Jan 03, 2024IEEE Workshop/Winter Conference on Applications of Computer Vision

In dynamic facial expression recognition (DFER) under unconstrained real-world conditions, motion blur and semantic ambiguity of expressions severely hinder accurate classification. To address this, we propose the first video-level soft-label mixup augmentation method, which jointly performs convex interpolation across video frames and their corresponding multi-emotion probability soft labels—explicitly modeling both expression continuity and semantic uncertainty. Our approach comprises three components: (1) soft label construction via emotion distribution estimation, (2) soft-label-guided frame-level mixup augmentation, and (3) an end-to-end trainable framework. Evaluated on the DFEW benchmark, our method achieves significant improvements over existing state-of-the-art methods, demonstrating that soft-label mixing enhances model robustness to ambiguous, dynamically evolving expressions in the wild. This work establishes a novel paradigm for uncertainty-aware learning in DFER, advancing the integration of probabilistic semantics into video-based representation learning.

3 citations1 influentialRead paper

SpotFormer: Multi-Scale Spatio-Temporal Transformer for Facial Expression Spotting

Jul 30, 2024arXiv.org

Facial expression spotting—particularly micro-expression spotting—faces two major challenges: interference from non-expressive facial movements and difficulty in modeling subtle, transient dynamics. To address these, we propose a multi-scale spatiotemporal modeling framework tailored for video-level temporal localization. First, we introduce Sliding-Window Multi-Resolution Optical Flow (SW-MRO), a novel motion-sensitive feature extraction method that enhances discriminability of fine-grained facial dynamics. Second, we design a multi-scale spatiotemporal Transformer integrating Facial Local Graph Pooling (FLGP) with convolutional layers to jointly capture local and global spatiotemporal dependencies. Third, we pioneer the incorporation of supervised contrastive learning into the spotting task to strengthen frame-level probabilistic discrimination. Our method achieves state-of-the-art performance on SAMM-LV and CAS(ME)², significantly improving micro-expression spotting F1-score while demonstrating strong robustness against head motion and non-expressive facial actions.

2 citationsRead paper

A Global Minimum Tax for Large Firms Only: Implications for Tax Competition

Apr 22, 2024Social Science Research Network

This paper examines how the “partial coverage” design of the Global Minimum Tax (GMT)—applying only to large multinational enterprises—affects multi-jurisdictional tax competition. Method: We develop a dynamic, game-theoretic model calibrated with numerical simulations, incorporating heterogeneous firms and multiple tax jurisdictions. Contribution/Results: We identify, for the first time, that the GMT induces jurisdictional stratification into a “tiered tax rate” structure: tax havens adopt discriminatory, size-contingent rates—establishing a new equilibrium of preferential treatment—while non-haven jurisdictions experience base reversion, increasing both tax revenue and social welfare. Under a 15% GMT, global net welfare rises significantly, and tax-system differentiation emerges as a novel paradigm in international tax competition. Our findings provide theoretical breakthroughs and policy-relevant insights into the structural implications of the GMT’s non-universal design.

2 citationsRead paper

On the Skew Stickiness Ratio

Feb 05, 2026

This study addresses the long-standing lack of a rigorous mathematical formulation and asymptotic analysis for the skew stickiness ratio, which characterizes the joint dynamics of asset prices and volatility. For the first time, the authors derive an explicit analytical expression for this ratio within a stochastic calculus framework by leveraging the Itô–Wentzell and Clark–Ocone formulas. They further conduct a systematic investigation of its short- and long-time asymptotic behavior under Bergomi-type stochastic volatility models. This work not only fills a critical theoretical gap but also provides a solid mathematical foundation for modeling and calibrating the volatility surface.

1 citationsRead paper

Physics-Aware Decoding for Communication Channels Governed by Partial Differential Equations

Jan 27, 2025

To address the low decoding efficiency and poor recovery accuracy of digital signals in physical channels governed by partial differential equations (PDEs)—such as heat conduction and the nonlinear Schrödinger equation (NLSE) channel—this paper proposes a physics-aware decoding framework. The core method introduces a novel gradient-flow decoding mechanism: leveraging backpropagated gradients from a differentiable PDE solver to directly guide iterative error correction, jointly optimizing channel modeling and symbol recovery under strict PDE constraints. Unlike conventional black-box neural networks or purely numerical solvers, our approach establishes a new signal processing paradigm grounded in physics-driven modeling and gradient-based optimization. Experiments on heat equation and NLSE channels demonstrate significantly reduced bit error rates, along with superior robustness and generalization compared to baseline methods. This work provides a differentiable, interpretable, and broadly applicable decoding pathway for physical-layer communications.

1 citationsRead paper
Recent publications

Latest Papers