polito.it
Politecnico di Torino (logo)

Are spiking neural networks resilient to internal faults? A comparison with other neural network models and an analysis of possible solutions to increase resilience

Francesco Grandi

Are spiking neural networks resilient to internal faults? A comparison with other neural network models and an analysis of possible solutions to increase resilience.

Rel. Stefano Di Carlo, Alessandro Savino, Alessio Carpegna. Politecnico di Torino, Corso di laurea magistrale in Data Science And Engineering, 2024

[img]
Preview
PDF (Tesi_di_laurea) - Tesi
Licenza: Creative Commons Attribution Non-commercial No Derivatives.

Download (2MB) | Preview
Abstract:

Are spiking neural networks resilient to internal faults? A comparison with other neural network models and an analysis of possible solutions to increase resilience. Spiking neural networks (SNNs) are a new type of computational model whose potential is worth examining due to their similarity to the biological brain, of which AI researchers hope to harness the power and capabilities. These networks achieve comparable performances and accuracy to traditional neural networks in temporal data-related tasks while employing fewer resources. Maintaining the comparison with the biological brain, a question arises spontaneously: Are spiking neural networks able to retain the biological brain capability of resiliency, allowing it to remain functional even when damaged? The question is crucial when implementing artificial neural networks since they play a role in many aspects of our daily lives, including safety-critical areas like self-driving cars, disease detection, and more. Therefore, it is essential to assess their resilience to degradation, which frequently occurs throughout a typical software life cycle. The main intent of this work is to compare the reliability of more traditional Artificial Neural Networks (ANNs) with Spiking Neural Networks. The methodology employed is based on conducting fault-injection campaigns. These processes entail deliberately introducing faults into the target model to evaluate its performance and robustness under faulty conditions. In particular, metrics such as accuracy and class probability on different datasets assess fault-caused damages on ANNs. In addition, a comparison with the most traditional types of Artificial Neural Networks is provided.

Relators: Stefano Di Carlo, Alessandro Savino, Alessio Carpegna
Academic year: 2024/25
Publication type: Electronic
Number of Pages: 75
Subjects:
Corso di laurea: Corso di laurea magistrale in Data Science And Engineering
Classe di laurea: New organization > Master science > LM-32 - COMPUTER SYSTEMS ENGINEERING
Aziende collaboratrici: UNSPECIFIED
URI: http://webthesis.biblio.polito.it/id/eprint/33248
Modify record (reserved for operators) Modify record (reserved for operators)