MetaMIML: Meta Multi-Instance Multi-Label Learning

TL;DR

MetaMIML introduces a novel meta-learning and network embedding framework to improve multi-instance multi-label learning by capturing inter-object dependencies and enabling fast adaptation to new tasks, outperforming existing methods.

Contribution

The paper proposes MetaMIML, a new approach combining network embedding and meta-learning to handle interdependent MIML objects across different types with limited labeled data.

Findings

MetaMIML outperforms state-of-the-art algorithms on benchmark datasets.

The approach effectively captures semantic information of multi-type objects.

MetaMIML enables rapid adaptation to new MIML tasks.

Abstract

Multi-Instance Multi-Label learning (MIML) models complex objects (bags), each of which is associated with a set of interrelated labels and composed with a set of instances. Current MIML solutions still focus on a single-type of objects and assumes an IID distribution of training data. But these objects are linked with objects of other types, %(i.e., pictures in Facebook link with various users), which also encode the semantics of target objects. In addition, they generally need abundant labeled data for training. To effectively mine interdependent MIML objects of different types, we propose a network embedding and meta learning based approach (MetaMIML). MetaMIML introduces the context learner with network embedding to capture semantic information of objects of different types, and the task learner to extract the meta knowledge for fast adapting to new tasks. In this way, MetaMIML can naturally deal with MIML objects at data level improving, but also exploit the power of meta-learning at the model enhancing. Experiments on benchmark datasets demonstrate that MetaMIML achieves a significantly better performance than state-of-the-art algorithms.