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) offer a promising alternative to traditionaldeep learning frameworks, since they provide higher computational efficiencydue to event-driven information processing. SNNs distribute the analog valuesof pixel intensities into binary spikes over time. However, the most widelyused input coding schemes, such as Poisson based rate-coding, do not leveragethe additional temporal learning capability of SNNs effectively. Moreover,these SNNs suffer from high inference latency which is a major bottleneck totheir deployment. To overcome this, we propose a scalable time-based encodingscheme that utilizes the Discrete Cosine Transform (DCT) to reduce the numberof timesteps required for inference. DCT decomposes an image into a weightedsum of sinusoidal basis images. At each time step, the Hadamard product of theDCT coefficients and a single frequency base, taken in order, is given to anaccumulator that generates spikes upon crossing a threshold. We use theproposed scheme to learn DCT-SNN, a low-latency deep SNN withleaky-integrate-and-fire neurons, trained using surrogate gradient descentbased backpropagation. We achieve top-1 accuracy of 89.94%, 68.3% and 52.43% onCIFAR-10, CIFAR-100 and TinyImageNet, respectively using VGG architectures.Notably, DCT-SNN performs inference with 2-14X reduced latency compared toother state-of-the-art SNNs, while achieving comparable accuracy to theirstandard deep learning counterparts. The dimension of the transform allows usto control the number of timesteps required for inference. Additionally, we cantrade-off accuracy with latency in a principled manner by dropping the highestfrequency components during inference.

DCT-SNN: Using DCT to Distribute Spatial Information over Time for Learning Low-Latency Spiking Neural Networks