Keisuke Maeda

This study aims to accurately predict students’ academic performance trajectories for university entrance examinations by proposing a machine learning framework that explicitly accounts for missing data and uncertainty. Mock examination data are characterized by substantial missing values due to heterogeneous participation in exams, as well as inherent randomness caused by variations in test content and examinee conditions. Conventional single-value imputation methods cannot adequately reconstruct the missing values arising from such heterogeneous participation without introducing strong bias, and existing educational prediction models based on deterministic formulations do not account for the inherent randomness and uncertainty in examination scores, thereby limiting the reliability of their forecasts. To address these challenges, we employ GP-VAE and SAITS, state-of-the-art methods for time-series imputation, to reconstruct incomplete mock examination data. Furthermore, we develop a Bayesian Neural Network (BayesNN) to predict future academic performance while explicitly modeling uncertainty. By integrating temporally aware imputation with probabilistic prediction, the proposed framework aims to provide more accurate and reliable performance forecasts than existing approaches. We evaluate the effectiveness of the proposed method through comparative experiments involving various combinations of imputation techniques and prediction models. Experimental results demonstrate that the proposed framework achieves competitive predictive accuracy: the combination of deep imputation methods and BayesNN yields the lowest average estimation error of 15.98 points, compared with 16.75 points for the conventional combination of mean imputation and linear regression. The contribution of this study does not lie in proposing a new deep learning model itself, but rather in systematically comparing combinations of time-series imputation methods and uncertainty-aware prediction models using real-world mock examination sequence data with missing values, thereby providing effective design guidelines for educational data analysis.

The deployment of data-driven computer vision models for structural health monitoring (SHM) is heavily constrained by the data silo dilemma due to stringent privacy and security regulations. While federated learning (FL) offers a privacy-preserving collaborative alternative, its application to nationwide infrastructure networks is severely hindered by the challenge of ``double heterogeneity'': macro-level physical divergence across disparate structural types and micro-level statistical imbalances within local datasets. To overcome this challenge, this paper proposes a novel hierarchical federated learning framework. The framework orchestrates a synergistic two-tier optimization strategy. At the macro-level, a dynamic gradient-based clustering mechanism autonomously aggregates distributed clients into specialized expert groups based on their structural degradation trajectories, circumventing the need for prior geographical metadata. Concurrently, at the micro-level, an intra-cluster Dynamic Region-Adaptive Proximal Regularization (DRAPR) module computes a real-time statistical Non-IID Intensity Score for each client. By adaptively modulating a proximal penalty based on local label skewness and gradient divergence, DRAPR effectively calibrates local updates, mitigates client drift, and prevents the catastrophic forgetting of minority damage classes. Comprehensive evaluations on a large-scale, real-world structural inspection dataset demonstrate that the hierarchical integration of macro-clustering and micro-regularization successfully neutralizes dual-level heterogeneity, yielding highly robust and specialized diagnostic models for complex infrastructure inspection.

Composed Video Retrieval (CoVR) enables flexible video search by retrieving a target video that reflects a specified modification to a query video. The triplet datasets—consisting of query videos, query text, and target videos—required for model training have been collected manually. Recent studies have explored automatic construction of training triplets for CoVR; however, most existing approaches rely heavily on caption similarity. This limitation is particularly problematic in soccer videos, where identical or highly similar captions can correspond to visually distinct situations, making it difficult to construct triplets with appropriate relationships. To address this issue, this paper proposes a multimodal triplet construction framework specialized for soccer videos. The key idea is to explicitly incorporate visual similarity alongside textual similarity. Specifically, candidate target videos are selected by combining visual similarity with commentary caption filtering, enabling the identification of videos that are visually similar yet semantically different. The semantic difference between videos is then generated as query text using a large language model (LLM) without manual annotation. Furthermore, a multimodal large language model (MLLM) is introduced to estimate whether the generated modification is visually and semantically consistent with the video pair. Rather than replacing human verification, this step provides an automated screening signal to identify potentially unreliable triplets. The experiments show that the proposed framework automatically constructs triplets with reasonable validity under limited human validation. These results demonstrate the potential of scalable triplet construction for CoVR in soccer videos.

This paper presents a novel adversarial defense framework that strategically exploits the non-transferability of adversarial attacks across multi-modal foundation models. While Contrastive Language–Image Pre-training (CLIP) models demonstrate remarkable zero-shot capabilities, they remain vulnerable to adversarial samples. Adversarial fine-tuning is widely adopted as a standard defense, yet the resulting robustness against sophisticated white-box attacks is often insufficient. To address this limitation, we aim to boost the robustness of an adversarially fine-tuned model by utilizing a pre-trained auxiliary model to leverage attack non-transferability. Specifically, we construct a common embedding space and introduce a detection scheme that identifies the attack target based on feature distances. By adaptively switching the prediction output, we effectively mitigate attacks. Experimental results demonstrate that our approach outperforms state-of-the-art adversarial fine-tuning methods in terms of adversarial robustness.

Visual Emotion Recognition (VER) is an important research topic due to its wide range of applications, including opinion mining and advertisement design. Extending this capability to recognize emotions at the individual level further broadens its potential applications. Recently, Multimodal Large Language Models (MLLMs) have attracted increasing attention and demonstrated performance comparable to that of conventional VER methods. However, MLLMs are trained on large and diverse datasets containing general opinions, which causes them to favor majority viewpoints and familiar patterns. This tendency limits their performance in a personalized VER, which is crucial for practical and real-world applications, and indicates a key area for improvement. To address this limitation, the proposed method employs discrete prompt tuning inspired by the process of humans’ prompt engineering to adapt the VER task to each individual. Our method selects the best natural language representation from the generated prompts and uses it to update the prompt for the realization of accurate personalized VER.

Egocentric video gaze estimation ipredicts where an individual looks in first-person videos providing an essential understanding of user attention. However, it must accommodate the variability in individuals’ gaze patterns and support learning in scenarios in which data are distributed across multiple devices. To overcome these challenges, we introduce a Personalized Federated Learning (PFL) framework with Comprehensive Parameters Freezing (FedCPF) that improves gaze estimation accuracy and personalization in distributed settings. Our approach employs a transformer-based backbone to capture temporal dependencies and spatial correlations in egocentric video frames.Within a PFL setting, only those parameters that exhibit the most significantly updated parameters during local training are selected and subsequently frozen, developing tailored client models that preserve high-impact features. By freezing these key parameters, FedCPF reduces the communication overhead and mitigates issues from heterogeneous data distributions while allowing clients to fine-tune models for their unique gaze behaviors. We validate FedCPF through extensive experiments on the EGTEA Gaze+ and Ego4D datasets. The experimental results demonstrate that our comprehensive parameter freezing strategy effectively balances global collaboration and local adaptation, making FedCPF a promising solution for accurate, adaptive gaze estimation in federated learning.

Recent research on text-to-image diffusion models has involved personalization methods to learn novel concepts from a few user-provided images for target image generation. Continual learning extends personalization to multiple concepts through long sequence training, resulting in more practical applications. However, the continual learning approach causes the following trade-off problems: reduced stability for previously learned concepts as the number of concepts increases, and reduced plasticity by maintaining the performance of past concepts. In this study, we propose a continual learning approach for personalization that maintains high-performance image generation across multiple concepts while learning new targets. Our method leverages deep generative replay with latent variables to balance the past and new concepts, demonstrating that replay in the latent space effectively suppresses performance degradation. Furthermore, by integrating a sub-model with each layer of the latent diffusion model through the conditional input of latent variables, the proposed method achieves robust personalization for diverse concepts. Our experiments demonstrate that the proposed method, which is optimized for many concepts, achieves superior performance comparable to state-of-the-art methods specialized for a few concepts. In addition, in the context of continual learning, we demonstrate that the proposed method effectively mitigates performance degradation for previously learned concepts, indicating its potential for realistic image generation applications involving long-tail scenarios.

In this paper, we propose a foreground-aware dataset distillation method that enhances patch selection in a content-adaptive manner. With the rising computational cost of training large-scale deep models, dataset distillation has emerged as a promising approach for constructing compact synthetic datasets that retain the knowledge of their large original counterparts. However, traditional optimization-based methods often suffer from high computational overhead, memory constraints, and the generation of unrealistic, noise-like images with limited architectural generalization. Recent non-optimization methods alleviate some of these issues by constructing distilled data from real image patches, but the used rigid patch selection strategies can still discard critical information about the main objects. To solve this problem, we first leverage Grounded SAM2 to identify foreground objects and compute per-image foreground occupancy, from which we derive a category-wise patch decision threshold. Guided by these thresholds, we design a dynamic patch selection strategy that, for each image, either selects the most informative patch from multiple candidates or directly resizes the full image when the foreground dominates. This dual-path mechanism preserves more key information about the main objects while reducing redundant background content. Extensive experiments on multiple benchmarks show that the proposed method consistently improves distillation performance over existing approaches, producing more informative and representative distilled datasets and enhancing robustness across different architectures and image compositions.