| Meno: | Hana |
|---|---|
| Priezvisko: | Hladíková |
| Názov: | Analysis of sparse echo-state network |
| Vedúci: | prof. Ing. Igor Farkaą, Dr. |
| Rok: | 2026 |
| Kµúčové slová: | Echo state networks, Reservoir computing, Sparse neural networks, SpaRCe, MNIST |
| Abstrakt: | This thesis focuses on the analysis, reimplementation, and experimental evaluation of the SpaRCe model, a sparse extension of the Echo State Network framework based on adaptive threshold-driven representations. The work introduces the theoretical foundations of reservoir computing, sparsity mechanisms, and non-dissipative reservoir architectures designed for long-term temporal processing. A complete reimplementation of the SpaRCe model was developed in the PyTorch framework and validated on the MNIST and pMNIST benchmark datasets, achieving performance comparable to the original implementation. Additional experiments investigated the influence of sparsification on coding level and sparse reservoir representations. The thesis also presents a reproducible experimental framework for evaluation on benchmark datasets commonly used in non-dissipative reservoir computing research. The framework includes automated preprocessing, multi-seed evaluation, result aggregation, and SLURM-based execution for large-scale experimentation. Preliminary experiments on selected benchmark datasets demonstrated behavior consistent with previously reported reservoir computing results and confirmed the functionality of the implemented framework. The developed implementation provides a flexible and extensible foundation for future research on sparse and non-dissipative reservoir computing architectures. |
Súbory diplomovej práce:
| sparce_thesis.pdf |
| DIPLO.zip |
Súbory prezentácie na obhajobe: