🤖 AI Summary
Addressing the challenge of multi-task learning (MTL) with highly heterogeneous and weakly correlated classification and regression tasks, this paper proposes a label-transformation-driven task integration framework that unifies disparate tasks into a single-network regression formulation. Unlike conventional MTL approaches, our method avoids shared backbone architectures and intricate loss weighting schemes; instead, it leverages task-specific label mapping functions and end-to-end joint optimization to enable efficient cross-task collaboration. Evaluated on the Advanced Steel Property dataset (4,536 images of size 224×224), our approach achieves 99.54% classification accuracy and a regression mean absolute percentage error of only 2.4%, substantially outperforming traditional MTL baselines while reducing both inference latency and training time. To the best of our knowledge, this is the first work to adopt label transformation as the core mechanism for task integration—establishing a concise, efficient, and scalable paradigm for weakly related heterogeneous MTL.
📝 Abstract
FastCAR is a novel task consolidation approach in Multi-Task Learning (MTL) for a classification and a regression task, despite task heterogeneity with only subtle correlation. It addresses object classification and continuous property variable regression, a crucial use case in science and engineering. FastCAR involves a labeling transformation approach that can be used with a single-task regression network architecture. FastCAR outperforms traditional MTL model families, parametrized in the landscape of architecture and loss weighting schemes, when learning of both tasks are collectively considered (classification accuracy of 99.54%, regression mean absolute percentage error of 2.4%). The experiments performed used an Advanced Steel Property dataset https://github.com/fastcandr/Advanced-Steel-Property-Dataset contributed by us. The dataset comprises 4536 images of 224x224 pixels, annotated with object classes and hardness properties that take continuous values. With our designed approach, FastCAR achieves reduced latency and time efficiency.