Fast and Accurate Transferability Measurement for Heterogeneous Multivariate Data

TL;DR

This paper introduces Transmeter, a novel method for rapidly and accurately measuring transferability between heterogeneous multivariate datasets, enabling better source dataset selection for target tasks.

Contribution

We propose Transmeter, a new approach combining pre-trained models, adversarial networks, and encoder-decoder architectures to efficiently estimate transferability across diverse datasets.

Findings

Transmeter achieves up to 10.3 times faster measurement than competitors.

It provides the most accurate transferability estimates in heterogeneous settings.

Using Transmeter for source selection improves transfer accuracy and reduces runtime.

Abstract

Given a set of heterogeneous source datasets with their classifiers, how can we quickly find the most useful source dataset for a specific target task? We address the problem of measuring transferability between source and target datasets, where the source and the target have different feature spaces and distributions. We propose Transmeter, a fast and accurate method to estimate the transferability of two heterogeneous multivariate datasets. We address three challenges in measuring transferability between two heterogeneous multivariate datasets: reducing time, minimizing domain gap, and extracting meaningful homogeneous representations. To overcome the above issues, we utilize a pre-trained source model, an adversarial network, and an encoder-decoder architecture. Extensive experiments on heterogeneous multivariate datasets show that Transmeter gives the most accurate transferability measurement with up to 10.3 times faster performance than its competitor. We also show that selecting the best source data with Transmeter followed by a full transfer leads to the best transfer accuracy and the fastest running time.