Comply: Learning Sentences with Complex Weights inspired by Fruit Fly Olfaction

TL;DR

Comply is a biologically inspired neural network that uses complex weights to learn sentence representations efficiently, outperforming previous models like FlyVec and rivaling larger models without extra parameters.

Contribution

Introducing Comply, a novel single-layer neural network utilizing complex weights and positional information to improve sentence embedding performance.

Findings

Comply surpasses FlyVec in sentence representation quality.

Comply matches larger state-of-the-art models in performance.

Comply produces interpretable, sparse contextual sentence representations.

Abstract

Biologically inspired neural networks offer alternative avenues to model data distributions. FlyVec is a recent example that draws inspiration from the fruit fly's olfactory circuit to tackle the task of learning word embeddings. Surprisingly, this model performs competitively even against deep learning approaches specifically designed to encode text, and it does so with the highest degree of computational efficiency. We pose the question of whether this performance can be improved further. For this, we introduce Comply. By incorporating positional information through complex weights, we enable a single-layer neural network to learn sequence representations. Our experiments show that Comply not only supersedes FlyVec but also performs on par with significantly larger state-of-the-art models. We achieve this without additional parameters. Comply yields sparse contextual representations of sentences that can be interpreted explicitly from the neuron weights.