Many Minds from One Model: Bayesian-Inspired Transformers for Population Diversity

TL;DR

This paper introduces Population Bayesian Transformers (B-Trans), a method to generate diverse, coherent transformer model instances from a single pre-trained model by injecting stochasticity into normalization layers, inspired by population diversity.

Contribution

The paper proposes a novel Bayesian-inspired approach to produce diverse transformer model instances from one pre-trained model, enhancing response diversity and task performance.

Findings

B-Trans generates diverse yet coherent model instances.

B-Trans improves response diversity in zero-shot tasks.

B-Trans outperforms deterministic baselines in task performance.

Abstract

Despite their scale and success, modern transformers are usually trained as single-minded systems: optimization produces a deterministic set of parameters, representing a single functional hypothesis about the data. Motivated by the analogy to human populations, in which population-level intelligence emerges from diverse individual behaviors, we propose Population Bayesian Transformers (B-Trans), which enable sampling diverse yet coherent transformer large language model instances (hereafter referred to as a 'mind') from a single pre-trained LLM. B-Trans introduces a Bayesian-inspired posterior proxy by injecting stochasticity directly into normalization layers, avoiding the prohibitive cost of training full Bayesian neural networks. Sampling from this proxy yields a population of minds with diverse behaviors while maintaining general competence. During the generation of each response, we sample a single realization from the random distribution and hold it fixed, ensuring temporal consistency and reasoning coherence. Experiments on zero-shot generation and Reinforcement Learning with Verifiable Rewards (RLVR) demonstrate that B-Trans effectively leverages the stochastic model diversity, yielding superior response diversity while achieving better task performance compared to deterministic baselines.