English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Spiking Neural Networks (SNNs) have recently emerged as an alternative todeep learning owing to sparse, asynchronous and binary event (or spike) drivenprocessing, that can yield huge energy efficiency benefits on neuromorphichardware. However, training high-accuracy and low-latency SNNs from scratchsuffers from non-differentiable nature of a spiking neuron. To address thistraining issue in SNNs, we revisit batch normalization and propose a temporalBatch Normalization Through Time (BNTT) technique. Most prior SNN works tillnow have disregarded batch normalization deeming it ineffective for trainingtemporal SNNs. Different from previous works, our proposed BNTT decouples theparameters in a BNTT layer along the time axis to capture the temporal dynamicsof spikes. The temporally evolving learnable parameters in BNTT allow a neuronto control its spike rate through different time-steps, enabling low-latencyand low-energy training from scratch. We conduct experiments on CIFAR-10,CIFAR-100, Tiny-ImageNet and event-driven DVS-CIFAR10 datasets. BNTT allows usto train deep SNN architectures from scratch, for the first time, on complexdatasets with just few 25-30 time-steps. We also propose an early exitalgorithm using the distribution of parameters in BNTT to reduce the latency atinference, that further improves the energy-efficiency.

Revisiting Batch Normalization for Training Low-latency Deep Spiking Neural Networks from Scratch