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/

Zendy Plus

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

Zendy Tools

Zendy Open

Neural network quantization aims to transform high-precision weights andactivations of a given neural network into low-precision weights/activationsfor reduced memory usage and computation, while preserving the performance ofthe original model. However, extreme quantization (1-bit weight/1-bitactivations) of compactly-designed backbone architectures (e.g., MobileNets)often used for edge-device deployments results in severe performancedegeneration. This paper proposes a novel Quantization-Aware Training (QAT)method that can effectively alleviate performance degeneration even withextreme quantization by focusing on the inter-weight dependencies, between theweights within each layer and across consecutive layers. To minimize thequantization impact of each weight on others, we perform an orthonormaltransformation of the weights at each layer by training an input-dependentcorrelation matrix and importance vector, such that each weight is disentangledfrom the others. Then, we quantize the weights based on their importance tominimize the loss of the information from the original weights/activations. Wefurther perform progressive layer-wise quantization from the bottom layer tothe top, so that quantization at each layer reflects the quantizeddistributions of weights and activations at previous layers. We validate theeffectiveness of our method on various benchmark datasets against strong neuralquantization baselines, demonstrating that it alleviates the performancedegeneration on ImageNet and successfully preserves the full-precision modelperformance on CIFAR-100 with compact backbone networks.

BiTAT: Neural Network Binarization with Task-dependent Aggregated Transformation