Evolving Kolmogorov-Arnold Networks
From ISLAB/CAISR
| Title | Evolving Kolmogorov-Arnold Networks |
|---|---|
| Summary | This project aims to enhance the architecture of Kolmogorov-Arnold Networks (KANs) by optimizing key components such as loss functions, activation functions, initialization methods, and learning processes to improve their performance and interpretability. |
| Keywords | |
| TimeFrame | Fall 2024 |
| References | |
| Prerequisites | |
| Author | |
| Supervisor | Mohammed Ghaith Altarabichi |
| Level | Master |
| Status | Open |
Kolmogorov-Arnold Networks (KANs), recently proposed by researchers at MIT, present a promising alternative to traditional Multi-Layer Perceptrons (MLPs), demonstrating superior performance in terms of both accuracy and interpretability. The goal of this project is to further advance the architecture of KANs by enhancing their computational graph through various research directions. These include optimizing loss functions, refining activation functions, developing more effective initialization schemes, and improving learning processes.
