1858. Advancing Value-Based Preference Learning through Neural Network Architectures and Transfer Learning
Invited abstract in session MA-8: Advances in preference learning methods, stream Multiple Criteria Decision Aiding.
Monday, 8:30-10:00Room: Clarendon SR 2.08
Authors (first author is the speaker)
| 1. | Michał Fredrych
|
| Institute of Computing Science, Poznan University of Technology | |
| 2. | Milosz Kadzinski
|
| Institute of Computing Science, Poznan University of Technology |
Abstract
We introduce novel methods in value-based preference learning, extending traditional decision-aiding models within deep learning frameworks. This enhances robustness, interpretability, and adaptability while addressing inconsistencies inherent in preference data. Additionally, we explore transfer learning to improve model performance across diverse decision contexts.
Our primary contribution is a neural network-based reinterpretation of the UTADIS method, incorporating granular function layers to model both monotonic and non-monotonic utility functions. This architecture improves flexibility and robustness while preserving explainability in multi-criteria decision-making.
The study systematically applies transfer learning within the neural network framework, using techniques such as parameter fine-tuning and characteristic point reconstruction. For comparison, similar transfer techniques are applied to mathematical programming models, enabling an evaluation of their effectiveness. While the focus remains on the neural model, the benchmark analysis provides insights into the trade-offs between structured optimization and deep learning.
Experimental results indicate that the proposed neural model performs comparably to traditional UTADIS while offering advantages in handling complex decision structures. Its ability to integrate transfer learning highlights its potential for scalable and interpretable decision support, advancing explainable AI in decision-aiding.
Keywords
- Multi-Objective Decision Making
- Machine Learning
- Continuous Optimization
Status: accepted
Back to the list of papers