Geometrically Aligned Transfer Encoder for Inductive Transfer in Regression Tasks

TL;DR

This paper introduces GATE, a novel transfer learning method for regression tasks that leverages differential geometry to align latent spaces on Riemannian manifolds, improving transfer performance and stability.

Contribution

The paper proposes GATE, a differential geometry-based transfer encoder that aligns latent spaces across tasks for regression, extending transfer learning beyond classification.

Findings

GATE outperforms conventional transfer methods on molecular graph datasets.

GATE provides stable transfer and extrapolation behavior.

The method effectively regularizes models in latent and extrapolation regions.

Abstract

Transfer learning is a crucial technique for handling a small amount of data that is potentially related to other abundant data. However, most of the existing methods are focused on classification tasks using images and language datasets. Therefore, in order to expand the transfer learning scheme to regression tasks, we propose a novel transfer technique based on differential geometry, namely the Geometrically Aligned Transfer Encoder (GATE). In this method, we interpret the latent vectors from the model to exist on a Riemannian curved manifold. We find a proper diffeomorphism between pairs of tasks to ensure that every arbitrary point maps to a locally flat coordinate in the overlapping region, allowing the transfer of knowledge from the source to the target data. This also serves as an effective regularizer for the model to behave in extrapolation regions. In this article, we demonstrate that GATE outperforms conventional methods and exhibits stable behavior in both the latent space and extrapolation regions for various molecular graph datasets.