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

We introduce a quantization-aware training algorithm that guarantees avoidingnumerical overflow when reducing the precision of accumulators duringinference. We leverage weight normalization as a means of constrainingparameters during training using accumulator bit width bounds that we derive.We evaluate our algorithm across multiple quantized models that we train fordifferent tasks, showing that our approach can reduce the precision ofaccumulators while maintaining model accuracy with respect to a floating-pointbaseline. We then show that this reduction translates to increased designefficiency for custom FPGA-based accelerators. Finally, we show that ouralgorithm not only constrains weights to fit into an accumulator ofuser-defined bit width, but also increases the sparsity and compressibility ofthe resulting weights. Across all of our benchmark models trained with 8-bitweights and activations, we observe that constraining the hidden layers ofquantized neural networks to fit into 16-bit accumulators yields an average98.2% sparsity with an estimated compression rate of 46.5x all whilemaintaining 99.2% of the floating-point performance.

Quantized Neural Networks for Low-Precision Accumulation with Guaranteed Overflow Avoidance