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

Analog computing hardwares, such as Processing-in-memory (PIM) accelerators,have gradually received more attention for accelerating the neural networkcomputations. However, PIM accelerators often suffer from intrinsic noise inthe physical components, making it challenging for neural network models toachieve the same performance as on the digital hardware. Previous works inmitigating intrinsic noise assumed the knowledge of the noise model, andretraining the neural networks accordingly was required. In this paper, wepropose a noise-agnostic method to achieve robust neural network performanceagainst any noise setting. Our key observation is that the degradation ofperformance is due to the distribution shifts in network activations, which arecaused by the noise. To properly track the shifts and calibrate the biaseddistributions, we propose a "noise-aware" batch normalization layer, which isable to align the distributions of the activations under variational noiseinherent in the analog environments. Our method is simple, easy to implement,general to various noise settings, and does not need to retrain the models. Weconduct experiments on several tasks in computer vision, includingclassification, object detection and semantic segmentation. The resultsdemonstrate the effectiveness of our method, achieving robust performance undera wide range of noise settings, more reliable than existing methods. We believethat our simple yet general method can facilitate the adoption of analogcomputing devices for neural networks.

Robust Processing-In-Memory Neural Networks via Noise-Aware Normalization