Learning Towards Emergence: Paving the Way to Induce Emergence by Inhibiting Monosemantic Neurons on Pre-trained Models

2025-03-30

AI Summary

This work addresses the challenge of suppressing single-meaning neurons—neurons exhibiting highly specific, interpretable activation patterns—during large language model (LLM) pretraining, a critical yet underexplored aspect of controlling emergent capabilities. We propose the first proactive pretraining-stage intervention method, comprising a tunable-threshold neuron retrieval mechanism, a false-killing-rate metric for precise evaluation, and a regularization-based suppression loss. Evaluated across the Pythia family (70M–6.9B parameters), our approach enables scalable, cross-size pretraining intervention. We empirically establish, for the first time, that neuron monosemanticity decreases with model scale. Our method achieves a 3.2× improvement in suppression efficiency without compromising training stability, yielding consistent gains on downstream tasks. By bridging the long-standing gap between fine-tuning–level and pretraining-level interventions, this work introduces a novel paradigm for controllable induction of emergent behaviors in LLMs.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Computer Vision: Large Vision Models

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Emergence, the phenomenon of a rapid performance increase once the model scale reaches a threshold, has achieved widespread attention recently. The literature has observed that monosemantic neurons in neural networks gradually diminish as the model scale increases. Subsequently, Learning From Emergence is proposed to actively inhibit monosemantic neurons in relatively small neural networks (e.g., BERT and Swin-Transformer) for promoting model performance with fine-tuning. However, to ultimately achieve emergence, it is demanding to support the monosemantic neuron inhibition in the pretraining phase of large-scale models. Thus, this work further pushes the boundary of this research direction to be Learning Towards Emergence (L2E) and enables the training and validating of the impact of inhibiting monosemantic neurons on larger pre-trained neural networks (e.g., Pythia-70M, 410M, and 2.8B). More specifically, to bridge the gap in current research, we first conduct experiments on models of various scales (up to 6.9B) to validate the monosemantic ideas. Then, we present a novel method L2E to address the inefficient monosemantic neuron retrieval and ineffective monosemantic neuron inhibition when existing methods are applied in the pretraining phase of large-scale models. It employs an adjustable thresholding technique for efficient neuron retrieval, incorporates a False Killing Rate metric to assess inhibition effects, and proposes a regularization-style inhibition approach, which addresses the limitations of previous approaches in both efficiency and effectiveness. Experimental results demonstrate the effectiveness of L2E's monosemantic neuron inhibition and its efficiency in implementation with large-scale models.

Problem

• Inhibiting monosemantic neurons in pre-trained large-scale models
• Improving efficiency of monosemantic neuron retrieval and inhibition
• Validating monosemantic neuron impact on models up to 6.9B

Innovation

• Inhibiting monosemantic neurons in pretraining
• Adjustable thresholding for efficient neuron retrieval
• Regularization-style inhibition for large models