Contrasting quadratic assignments for set-based representation learning
This work addresses a bottleneck in contrastive learning for representation learning, offering an incremental improvement for tasks like metric learning and self-supervised classification.
The paper tackles the limitation of standard contrastive learning in handling intra-set and inter-set similarities by proposing a set-contrastive objective based on combinatorial quadratic assignment theory. The result shows improved representations for metric learning and self-supervised classification tasks, though no concrete numbers are provided in the abstract.
The standard approach to contrastive learning is to maximize the agreement between different views of the data. The views are ordered in pairs, such that they are either positive, encoding different views of the same object, or negative, corresponding to views of different objects. The supervisory signal comes from maximizing the total similarity over positive pairs, while the negative pairs are needed to avoid collapse. In this work, we note that the approach of considering individual pairs cannot account for both intra-set and inter-set similarities when the sets are formed from the views of the data. It thus limits the information content of the supervisory signal available to train representations. We propose to go beyond contrasting individual pairs of objects by focusing on contrasting objects as sets. For this, we use combinatorial quadratic assignment theory designed to evaluate set and graph similarities and derive set-contrastive objective as a regularizer for contrastive learning methods. We conduct experiments and demonstrate that our method improves learned representations for the tasks of metric learning and self-supervised classification.