To be Continuous, or to be Discrete, Those are Bits of Questions

TL;DR

This paper explores the use of binary representations on the output side of models, extending contrastive hashing to improve structured prediction tasks by bridging continuous and discrete representations.

Contribution

It introduces structured contrastive hashing with bit-level CKY, a new similarity function, and a contrastive loss, advancing binary output modeling in deep learning.

Findings

Achieves competitive performance on structured prediction tasks.

Demonstrates binary representations bridge continuous and discrete properties.

Extends contrastive hashing to output-side binary representations.

Abstract

Recently, binary representation has been proposed as a novel representation that lies between continuous and discrete representations. It exhibits considerable information-preserving capability when being used to replace continuous input vectors. In this paper, we investigate the feasibility of further introducing it to the output side, aiming to allow models to output binary labels instead. To preserve the structural information on the output side along with label information, we extend the previous contrastive hashing method as structured contrastive hashing. More specifically, we upgrade CKY from label-level to bit-level, define a new similarity function with span marginal probabilities, and introduce a novel contrastive loss function with a carefully designed instance selection strategy. Our model achieves competitive performance on various structured prediction tasks, and demonstrates that binary representation can be considered a novel representation that further bridges the gap between the continuous nature of deep learning and the discrete intrinsic property of natural languages.