Absolute Zero-Shot Learning

TL;DR

This paper introduces Absolute Zero-Shot Learning (AZSL), a data-free classification paradigm using a teacher model to guide training without data leakage, achieving competitive zero-shot learning performance.

Contribution

It proposes a novel AZSL framework involving a generator and student network, enabling data-free knowledge transfer with security considerations in black-box and white-box scenarios.

Findings

Achieves state-of-the-art ZSL and GZSL performance in white-box scenarios.

Demonstrates promising results under black-box deployment.

Provides analysis of security levels in AZSL settings.

Abstract

Considering the increasing concerns about data copyright and privacy issues, we present a novel Absolute Zero-Shot Learning (AZSL) paradigm, i.e., training a classifier with zero real data. The key innovation is to involve a teacher model as the data safeguard to guide the AZSL model training without data leaking. The AZSL model consists of a generator and student network, which can achieve date-free knowledge transfer while maintaining the performance of the teacher network. We investigate `black-box' and `white-box' scenarios in AZSL task as different levels of model security. Besides, we also provide discussion of teacher model in both inductive and transductive settings. Despite embarrassingly simple implementations and data-missing disadvantages, our AZSL framework can retain state-of-the-art ZSL and GZSL performance under the `white-box' scenario. Extensive qualitative and quantitative analysis also demonstrates promising results when deploying the model under `black-box' scenario.