LENS: Learning Ensemble Confidence from Neural States for Multi-LLM Answer Integration

TL;DR

LENS introduces a novel ensemble method that learns to estimate model confidence from internal neural states, improving multi-LLM answer integration without modifying models or adding significant computation.

Contribution

It proposes a lightweight confidence predictor using internal representations to enhance ensemble performance of multiple LLMs.

Findings

LENS outperforms traditional ensemble methods on question-answering tasks.

Internal neural states provide valuable signals for confidence estimation.

The method requires negligible additional computation.

Abstract

Large Language Models (LLMs) have demonstrated impressive performance across various tasks, with different models excelling in distinct domains and specific abilities. Effectively combining the predictions of multiple LLMs is crucial for enhancing system robustness and performance. However, existing ensemble methods often rely on simple techniques like voting or logits ensembling, which overlook the varying confidence and reliability of models in different contexts. In this work, we propose LENS (Learning ENsemble confidence from Neural States), a novel approach that learns to estimate model confidence by analyzing internal representations. For each LLM, we train a lightweight linear confidence predictor that leverages layer-wise hidden states and normalized probabilities as inputs. This allows for more nuanced weighting of model predictions based on their context-dependent reliability. Our method does not require modifying the model parameters and requires negligible additional computation. Experimental results on multiple-choice and boolean question-answering tasks demonstrate that LENS outperforms traditional ensemble methods by a substantial margin. Our findings suggest that internal representations provide valuable signals for determining model confidence and can be effectively leveraged for ensemble learning.