What does this research mean for the field?

NNS achieves 1.91x-2.16x effective throughput improvement in neural network outputs on commodity mobile GPUs. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

To present the NNS method, which enhances neural network performance using principles from Shannon's theorem.

March 17, 20260 citationsOpen Access

View Full Paper

NNS: Neural Network Dimensioned by Shannon Theorem

APAndrés Sebastián Pirolo

Key Points

To present the NNS method, which enhances neural network performance using principles from Shannon's theorem.
Introduced NNS to generate two independent outputs from one GPU dispatch.
Utilized a bitwise transpose of weight blocks to optimize storage and performance.
Validated the method using standard OpenGL ES 3.1 on mobile GPUs.
Achieved effective throughput improvement between 1.91x to 2.16x across quantization modes.
Documented overhead varied from -7.3% to +4.8%.
Memory overhead was measured at 31%, with a single output buffer reusable.

Abstract

Abstract: We introduce NNS (Neural Network dimensioned by Shannon Theorem), a method producing two independent neural outputs from a single GPU dispatch and a single weight buffer load. The second channel is derived from the bitwise transpose of the packed weight block, computed on-the-fly in shared memory at zero storage cost. Validated on commodity mobile GPU via standard OpenGL ES 3.1 without root access or proprietary SDK, NNS achieves 1.91x-2.16x effective throughput improvement across four quantization modes, with overhead ranging from -7.3% (dual faster than single) to +4.8%. Memory overhead is 31%, confined to a single reutilizable output buffer.

KI fragen

Bookmark

View Full Paper

Cite This Study

Andrés Sebastián Pirolo (Sun,) studied this question.

synapsesocial.com/papers/69b8f13ddeb47d591b8c62bf https://doi.org/https://doi.org/10.5281/zenodo.19027417

KI fragen

Bookmark

View Full Paper