Institution profile

Chulalongkorn University

Academic institutionasia · th
Official website
Research library109linked papers
Opportunities0open roles
Selected work

Representative Papers

Seeing Isn't Always Believing: Analysis of Grad-CAM Faithfulness and Localization Reliability in Lung Cancer CT Classification

Jan 19, 2026

This study evaluates the faithfulness and localization reliability of Grad-CAM for interpreting lung cancer classification in chest CT scans. It presents the first systematic comparison between convolutional architectures (ResNet, DenseNet, EfficientNet) and Vision Transformers (ViT) in terms of explanation consistency, localization accuracy, and robustness to perturbations, thereby elucidating how attention mechanisms influence interpretability. The findings reveal that Grad-CAM yields robust explanations in convolutional models but suffers from distorted visualizations in ViT due to its non-local attention, with significant inter-model discrepancies in localization performance—raising concerns about its clinical generalizability. To address these issues, this work proposes a model-aware interpretability evaluation framework, offering a new perspective toward the trustworthy deployment of medical AI systems.

1 citationsRead paper

Role of scrambling and noise in temporal information processing with quantum systems

May 15, 2025

This work investigates the scalability and memory retention of quantum scrambling systems for temporal information processing. We develop a quantum reservoir computing (QRC) framework based on higher-order unitary designs, rigorously analyzing its behavior under both noiseless and noisy environments. Our analysis integrates local noise modeling, concentration inequalities, and random matrix theory. We establish, for the first time, that in the noiseless case, measurement-based readout concentrates exponentially with reservoir size and remains stable across iterations, whereas memory of the initial state and early inputs decays doubly exponentially in both reservoir size and iteration count. Under noise, iterative dynamics induce additional exponential memory decay. We introduce novel proof techniques that demonstrate the feasibility of small, reusable reservoirs while identifying fundamental bottlenecks to large-scale generalization. These results provide critical theoretical foundations for the limits of quantum temporal learning and inform hardware-aware design principles.

1 citationsRead paper
Recent publications

Latest Papers