Learning Task-Agnostic Representations through Multi-Teacher Distillation

TL;DR

This paper proposes a task-agnostic multi-teacher distillation framework that leverages diverse models to produce versatile representations, improving performance across multiple downstream tasks without relying on task-specific labels.

Contribution

Introduces a novel task-agnostic distillation method using a majority vote objective, enabling diverse teacher models to generate general-purpose embeddings.

Findings

Effective across text, vision, and molecular modeling.

Improves downstream task performance such as classification and clustering.

Releases state-of-the-art embedding models.

Abstract

Casting complex inputs into tractable representations is a critical step across various fields. Diverse embedding models emerge from differences in architectures, loss functions, input modalities and datasets, each capturing unique aspects of the input. Multi-teacher distillation leverages this diversity to enrich representations but often remains tailored to specific tasks. In this paper, we introduce a task-agnostic framework based on a ``majority vote" objective function. We demonstrate that this function is bounded by the mutual information between student and teachers' embeddings, leading to a task-agnostic distillation loss that eliminates dependence on task-specific labels or prior knowledge. Our evaluations across text, vision models, and molecular modeling show that our method effectively leverages teacher diversity, resulting in representations enabling better performance for a wide range of downstream tasks such as classification, clustering, or regression. Additionally, we train and release state-of-the-art embedding models, enhancing downstream performance in various modalities.