This paper introduces an explainable deep-learning framework for the accurate classification of microplastic and natural microfibers using polarization-resolved digital holographic microscopy. By extracting 72 polarization-based features and employing a deep neural network with SHAP analysis, the method achieves high accuracy and identifies key optical properties for distinguishing fiber types.

This study introduces an explainable deep radiogenomic framework for non-invasively predicting MGMT promoter methylation status in glioblastoma using multi-parametric MRI. By fusing radiomic features with deep learning and providing explainable AI insights, the method offers a robust and interpretable approach to a crucial biomarker for treatment response.

This thesis proposes a novel robust deep learning control framework with guaranteed performance for heavy-duty machinery. It addresses challenges in electrification and AI integration by ensuring safety and reliability across diverse actuation mechanisms and operational conditions.

This paper introduces a novel method for early skin cancer classification that leverages a soft voting ensemble of Convolutional Neural Networks. The approach combines data preprocessing, image segmentation, and an ensemble of MobileNetV2, VGG19, and InceptionV3 to achieve high accuracy and balanced performance for real-world dermatological applications.

This paper introduces the Multi-directional Similarity Network (MSN), a novel deep learning approach designed for accurate and efficient detection of copy-move image forgeries, including those created by both manual manipulation and advanced deep generative networks. It addresses limitations in existing deep detection models by improving feature representation and localization, while also presenting a new benchmark for deep-synthesized copy-move forgeries.

This paper demonstrates that deep neural networks, despite being trained on diverse tasks and initializations, converge to remarkably similar low-dimensional parametric subspaces. This finding offers significant implications for model reusability, multi-task learning, and reducing the computational and environmental costs of large-scale neural models.

This paper explores the use of a lightweight and scalable EfficientNet architecture, combined with global jet features, for the computationally inexpensive yet competitive classification of top-quark jets. It aims to address the high computational demands of current state-of-the-art jet tagging methods like Transformers and GNNs.

This tutorial provides comprehensive lecture notes on regression analysis, covering fundamental concepts from linear models to deep learning. It aims to equip students with a solid understanding of various regression models, including linear, logistic, and Softmax regression, along with essential methodologies like loss function design, parameter estimation, and regularization techniques, bridging classical statistics and modern machine learning practices.

Structured Uncertainty Similarity Score (SUSS) is a new perceptual image similarity metric that models images through structured multivariate Normal distributions for interpretable, human-aligned assessments. It achieves strong perceptual calibration and localized explanations, making it suitable as a robust loss function for computer vision tasks.

This paper examines the frequency bias in DNN-based forgery detectors and proposes a frequency alignment method to improve detection reliability.