SAF proposes a new paradigm for training SNNs, reducing operations and memory while maintaining accuracy.
