SPIN: Sparsifying and Integrating Internal Neurons in Large Language Models for Text Classification
This work addresses a bottleneck in text classification for users of LLMs by enhancing performance through better use of internal model data, though it is incremental as it builds on existing paradigms.
The authors tackled the problem of underutilizing internal neuron information in LLMs for text classification by proposing SPIN, a framework that sparsifies and integrates neurons from intermediate layers, resulting in significant improvements in accuracy, efficiency, and interpretability.
Among the many tasks that Large Language Models (LLMs) have revolutionized is text classification. Current text classification paradigms, however, rely solely on the output of the final layer in the LLM, with the rich information contained in internal neurons largely untapped. In this study, we present SPIN: a model-agnostic framework that sparsifies and integrates internal neurons of intermediate layers of LLMs for text classification. Specifically, SPIN sparsifies internal neurons by linear probing-based salient neuron selection layer by layer, avoiding noise from unrelated neurons and ensuring efficiency. The cross-layer salient neurons are then integrated to serve as multi-layered features for the classification head. Extensive experimental results show our proposed SPIN significantly improves text classification accuracy, efficiency, and interpretability.