Computational Problems of Electrical Engineering 2024

Session

Session 1 - Machine Learning

11 Sept 2024, 11:00

Holimo Hotel

Conveners

Session 1 - Machine Learning

• Krzysztof Siwek
• Michał Kruk

50. A comparative analysis of methods for preparing input data for a neural network to assess lightning hazard.

Mrs Emilia Sobieska (Warsaw University of Technology)

11/09/2024, 11:00

Machine Learning

Presentation at the conference

Designing lightning protection systems by Polish standardization is a standardized, albeit ossified process. Estimating lightning risk, which is the basic stage of this procedure according to the PN-EN 62305 standard, is at risk of error and discretion. This article presents a proposal for automating the above process. Starting with the simulation of the electric field distribution around...

36. Textual explanations for image classification using multimodal LLM

Bartosz Sawicki (Warsaw University of Technology), Tomasz Leś (Warsaw University of Technology)

11/09/2024, 11:20

Machine Learning

Presentation at the conference

In recent years, models based on deep neural networks have demonstrated exceptional capabilities in recognizing and classifying objects in images. Nevertheless, the issue remains with users' trust in the results of these models. In our research, we utilized the multimodal GPT-4o model, which not only classifies objects but also generates explanations for its decisions in natural language....

23. Methods of classifying voltage surges using deep neural networks

Paweł Kluge (Warsaw University of Technology), Zuzanna Krawczyk-Borysiak (Warsaw University of Technology)

11/09/2024, 11:40

Machine Learning

Presentation at the conference

Insulating systems of high-voltage electrical devices used in the power industry are exposed to various types of exposure during their operation. Overvoltages in the nature of lightning surges, resulting from the direct or indirect impact of atmospheric discharges - lightning, constitute a certain group of these exposures. For this reason, the need to perform tests and design and acceptance...

12. Efficient Information Extraction from Resumes Using Small Language Models for SMEs Based on Zero-Shot Learning Approach

Prof. Jarosław Kurek (Warsaw University of Life Sciences, Institute of Information Technology, Department of Artificial Intelligence)

11/09/2024, 12:00

Machine Learning

Presentation at the conference

In today's recruitment field, accurately extracting information from resumes is crucial. This paper looks at how three small language models—Llama 2, Llama 3, and Phi-3—can help with this task using a Zero-Shot approach. We checked how well these models perform by comparing their results with a hand-made dataset, focusing on accuracy and the time each model takes to run on computers that small...

7. Grad-CAM for Explainable AI (XAI) in CNN-Based Drilled Hole Classification in Melamine Faced Chipboard

Prof. Jarosław Kurek (Warsaw University of Life Sciences, Institute of Information Technology, Department of Artificial Intelligence)

11/09/2024, 12:20

Machine Learning

Presentation at the conference

In the furniture industry, precision in drilling holes in melamine-faced chipboard is crucial to maintaining product quality and minimizing financial losses. Manual monitoring of drill conditions, while somewhat effective, is inefficient and imprecise. This paper presents a Convolutional Neural Network (CNN) based approach for automated tool condition monitoring (TCM) using Gradient-weighted...

6. Application of LIME for Explainable AI (XAI) in Classifying Drilled Holes in Melamine Faced Chipboard based on CNN using VGG16 pretrained network

Prof. Jarosław Kurek (Warsaw University of Life Sciences, Institute of Information Technology, Department of Artificial Intelligence)

11/09/2024, 12:40

Machine Learning

Presentation at the conference

This study explores the application of Local Interpretable Model-Agnostic Explanations (LIME) for enhancing the explainability of a Convolutional Neural Network (CNN) used in classifying the condition of drilled holes in melamine-faced chipboard. A VGG16 pretrained network serves as the foundation of our CNN model, which is tasked with classifying the holes based on their wear states. The...