 |
International Society of Science and Applied Technologies |
| Optimal Checkpoint for Hardwareized Artificial Neural Networks | ||||
| Author | Kenichiro Naruse
|
|||
| Co-Author(s) | Toshio Nakagawa
|
|||
| Abstract | We propose artificial neural network(ANN) evaluation and checkpointing model with considering computing units reliability. ANN is an important essence of artificial intelligence. ANN builds up from several computing units like a neuron. If the computing unit in ANN occurs some error due to noises, hardware failure and so on, it will output some wrong result. To prevent such a matter, we need to make chekcpoint and recovery. Therefore, when we cannot make the error, we ignore some computing units, because all computing units need to work to output the result. We focus that the ANN doesn’t need to work every computing unit depending on ANN weight and threshold. In this paper, we consider ANN computing units checkpoint model with all computing units working. Moreover we consider that resist some computing units faults depending on ANN weight and threshold.
|
|||
| Keywords | Neural Networks, Reliability, Constant Task, Series System | |||
| Article #: RQD2025-435 | ||||
Proceedings of 30th ISSAT International Conference on Reliability & Quality in Design |