2024
Authors
Radeva, P; Furnari, A; Bouatouch, K; de Sousa, AA;
Publication
VISIGRAPP (4): VISAPP
Abstract
2024
Authors
Baghcheband, H; Soares, C; Reis, LP;
Publication
FOUNDATIONS OF INTELLIGENT SYSTEMS, ISMIS 2024
Abstract
Data valuation, the process of assigning value to data based on its utility and usefulness, is a critical and largely unexplored aspect of data markets. Within the Machine Learning Data Market (MLDM), a platform that enables data exchange among multiple agents, the challenge of quantifying the value of data becomes particularly prominent. Agents within MLDM are motivated to exchange data based on its potential impact on their individual performance. Shapley Value-based methods have gained traction in addressing this challenge, prompting our study to investigate their effectiveness within the MLDM context. Specifically, we propose the Gain Data Shapley Value (GDSV) method tailored for MLDM and compare it to the original data valuation method used in MLDM. Our analysis focuses on two common learning algorithms, Decision Tree (DT) and K-nearest neighbors (KNN), within a simulated society of five agents, tested on 45 classification datasets. results show that the GDSV leads to incremental improvements in predictive performance across both DT and KNN algorithms compared to performance-based valuation or the baseline. These findings underscore the potential of Shapley Value-based methods in identifying high-value data within MLDM while indicating areas for further improvement.
2024
Authors
Costa, L; Almeida, A; Reis, L;
Publication
5TH INTERNATIONAL CONFERENCE ON INDUSTRY 4.0 AND SMART MANUFACTURING, ISM 2023
Abstract
In today's volatile, uncertain, and complex business environments, manufacturing companies must not only adapt to market demands but also minimize the time between problem occurrence and resolution. The implementation of lean manufacturing systems has been crucial in this regard. However, traditional approaches have shown notable inefficiencies that can be effectively addressed through digitalization. By embracing digital solutions, manufacturing companies can ensure efficient continuous improvement, driving performance to higher levels. This study aims to find a digital solution for a specific company that faces daily challenges associated with low visibility into production. An investigation revealed that the Lean tools used by the company were outdated, directly affecting the generated information and consequently, decision-making. The integration of a Manufacturing Execution System into the factory's dynamics was the solution found. In this context, a step-by-step methodology is proposed to guide the implementation. As a result, a prototype of the system was designed. The validation of the system by end-users demonstrates the success of the proposed methodology.
2024
Authors
Abouelmaty, AM; Colaço, A; Fares, AA; Ramos, A; Costa, PA;
Publication
COMPUTERS AND GEOTECHNICS
Abstract
This study focuses on the assessment of ground vibrations due to pile driving activities. Given the likelihood of excessive vibration due to the driving process, it is imperative to predict vibration levels during the design phase. The primary goal of this work is to integrate machine learning techniques, specifically Extreme Gradient Boosting (XGBoost) and Artificial Neural Networks (ANNs) for real-time vibration prediction. The training dataset was generated using a validated numerical model and the trained models were validated based on experimental results. This validation process highlights the efficiency and accuracy of Extreme Gradient Boosting in predicting the-free-field response of the ground.
2024
Authors
Radeva, P; Furnari, A; Bouatouch, K; de Sousa, AA;
Publication
VISIGRAPP (3): VISAPP
Abstract
2024
Authors
Tuna, R; Baghoussi, Y; Soares, C; Mendes-Moreira, J;
Publication
ADVANCES IN INTELLIGENT DATA ANALYSIS XXII, PT II, IDA 2024
Abstract
Forecasting methods are affected by data quality issues in two ways: 1. they are hard to predict, and 2. they may affect the model negatively when it is updated with new data. The latter issue is usually addressed by pre-processing the data to remove those issues. An alternative approach has recently been proposed, Corrector LSTM (cLSTM), which is a Read & Write Machine Learning (RW-ML) algorithm that changes the data while learning to improve its predictions. Despite promising results being reported, cLSTM is computationally expensive, as it uses a meta-learner to monitor the hidden states of the LSTM. We propose a new RW-ML algorithm, Kernel Corrector LSTM (KcLSTM), that replaces the meta-learner of cLSTM with a simpler method: Kernel Smoothing. We empirically evaluate the forecasting accuracy and the training time of the new algorithm and compare it with cLSTM and LSTM. Results indicate that it is able to decrease the training time while maintaining a competitive forecasting accuracy.
The access to the final selection minute is only available to applicants.
Please check the confirmation e-mail of your application to obtain the access code.