Adaptable Embeddings Network (AEN)
This work addresses the need for compute-effective classification models for low-resource environments like edge devices, though it appears incremental as it builds on existing dual-encoder and KDE methods.
The paper tackles the problem of high computational cost in language models for text classification by introducing Adaptable Embeddings Networks (AEN), a dual-encoder architecture using Kernel Density Estimation that achieves comparable or superior results to larger autoregressive models in synthetic experiments.
Modern day Language Models see extensive use in text classification, yet this comes at significant computational cost. Compute-effective classification models are needed for low-resource environments, most notably on edge devices. We introduce Adaptable Embeddings Networks (AEN), a novel dual-encoder architecture using Kernel Density Estimation (KDE). This architecture allows for runtime adaptation of classification criteria without retraining and is non-autoregressive. Through thorough synthetic data experimentation, we demonstrate our model outputs comparable and in certain cases superior results to that of autoregressive models an order of magnitude larger than AEN's size. The architecture's ability to preprocess and cache condition embeddings makes it ideal for edge computing applications and real-time monitoring systems.