Domain Expansion: Parameter-Efficient Modules as Building Blocks for Composite Domains

TL;DR

The work tackles distribution generalization across composite domains by composing Parameter-Efficient Modules (PEMs) trained via Parameter-Efficient Fine-Tuning (PEFT). It trains eight trait PEMs for MBTI dichotomies using LoRA and IA3, then forms 16 personality PEMs through weight-space fusion, first with an unweighted sum and then with a constrained weighted sum $\Theta_P = \sum_{i=1}^{4} \lambda_i \bm{\theta}_{T_i}$ where $\lambda_i \in (0,1)$ and $\sum_i \lambda_i = 1$. Evaluation uses a MBTI-style online questionnaire and synthetic trait data generated by ChatGPT-4, showing that trait PEMs achieve clear trait alignment and that most personality PEMs align with their target MBTI types without additional fine-tuning. A Streamlit-based demo API illustrates practical deployment, highlighting the approach’s efficiency and potential applicability to other composite-domain problems.

Abstract

Parameter-Efficient Fine-Tuning (PEFT) is an efficient alternative to full scale fine-tuning, gaining popularity recently. With pre-trained model sizes growing exponentially, PEFT can be effectively utilized to fine-tune compact modules, Parameter-Efficient Modules (PEMs), trained to be domain experts over diverse domains. In this project, we explore composing such individually fine-tuned PEMs for distribution generalization over the composite domain. To compose PEMs, simple composing functions are used that operate purely on the weight space of the individually fine-tuned PEMs, without requiring any additional fine-tuning. The proposed method is applied to the task of representing the 16 Myers-Briggs Type Indicator (MBTI) composite personalities via 4 building block dichotomies, comprising of 8 individual traits which can be merged (composed) to yield a unique personality. We evaluate the individual trait PEMs and the composed personality PEMs via an online MBTI personality quiz questionnaire, validating the efficacy of PEFT to fine-tune PEMs and merging PEMs without further fine-tuning for domain composition.

Figures (3)

• Figure 1: MBTI Dichotomies, Traits and Personalities
• Figure 2: Method Overview
• Figure 3: PEFT Training