A Cycle-Consistency Constrained Framework for Dynamic Solution Space Reduction in Noninjective Regression

TL;DR

This paper introduces a cycle-consistency framework for non-injective regression that jointly optimizes forward and backward models, reducing manual assumptions and improving reconstruction accuracy in synthetic and simulated datasets.

Contribution

It proposes a novel cycle consistency-based training framework that eliminates manual rule design and prior assumptions in non-injective regression tasks.

Findings

Cycle reconstruction error below 0.003

Approximately 30% improvement over baseline models

Supports unsupervised learning and reduces manual intervention

Abstract

To address the challenges posed by the heavy reliance of multi-output models on preset probability distributions and embedded prior knowledge in non-injective regression tasks, this paper proposes a cycle consistency-based data-driven training framework. The method jointly optimizes a forward model {\Phi}: X to Y and a backward model {\Psi}: Y to X, where the cycle consistency loss is defined as L _cycleb equal L(Y reduce {\Phi}({\Psi}(Y))) (and vice versa). By minimizing this loss, the framework establishes a closed-loop mechanism integrating generation and validation phases, eliminating the need for manual rule design or prior distribution assumptions. Experiments on normalized synthetic and simulated datasets demonstrate that the proposed method achieves a cycle reconstruction error below 0.003, achieving an improvement of approximately 30% in evaluation metrics compared to baseline models without cycle consistency. Furthermore, the framework supports unsupervised learning and significantly reduces reliance on manual intervention, demonstrating potential advantages in non-injective regression tasks.