Artificial Intelligence

Artificial Intelligence (AI) has increasingly influenced modern society, recently in particular through significant advancements in Large Language Models (LLMs). However, high computational and storage demands of LLMs still limit their deployment in resource-constrained environments. Knowledge distillation addresses this challenge by training a smaller language model (student) from a larger one (teacher). Previous research has introduced several distillation methods for both generating training data and training the student model.

Semiconductor manufacturing is a highly complex process requiring precise control and monitoring to maintain product quality and yield. This research presents a comprehensive comparative analysis of three machine learning algorithms—Random Forest, Support Vector Machine (SVM), and XGBoost—for anomaly detection in semiconductor fabrication. Through extensive experimentation using a real-world wafer dataset, we demonstrate that XGBoost outperforms other models, achieving 97.1\% accuracy, 96.4\% precision, and 95.0\% recall.

The UQTR dataset consists of 7838 real and synthetic images of the Université du Québec à Trois-Rivières (UQTR) campus road under normal and snow conditions. The image resolution is 1280×720. It includes lane labels in .txt files, where each row stores the set of points of a lane. The points are stored as x1 y1 x2 y2, as in the tutorial by Ruijin Liu, Zejian Yuan, Tie Liu, Zhiliang Xiong: Train and Test Your Custom Data.

Walnut and Heart CT Data corresponding to Noisier2Inverse consist of high-resolution computed tomography (CT) scans used for evaluating deep learning-based image reconstruction under severe noise conditions. The dataset includes walnut CT scans from controlled experimental settings and clinical cardiac CT images. The Walnut data stems from this source: https://paperswithcode.com/dataset/cbct-walnut, and the Heart CT data is processed in python before, and is provided in .pt format.