Hashing and Filtering Techniques for Weightless Neural Networks
Date:
Title – “Hashing and Filtering Techniques for Weightless Neural Networks”. Abstract – This talk covers advanced hashing and filtering techniques to improve the efficiency of Weightless Neural Networks (WNNs) for edge inference. We will delve into the BTHOWeN and ULEEN architectures, which leverage counting Bloom filters and H3 hash functions to reduce memory use, enhance accuracy, and cut energy consumption. We will also present SoonFilter, which boosts WNN performance through gradient-driven architecture adjustments. These innovations result in significant gains in latency and power efficiency, positioning WNNs as a strong choice for edge computing.