Single-Task Learning (STL)

Single-Task Learning (STL) is the regular machine learning that trains one task on one dataset.

Definition

In CLArena, single-task learning is specifically designed for classification problems and follows the formal definition below.

Definition 1 (Single-Task Learning Classification) Given:

• An initialized neural network model $f$
  • A shared backbone network $B$
  • An output head
• Training data: ${\mathcal{D}}_{\text{train}}=\{({\mathbf{x}}_i,y_i){\}}_{i=1}^N\in (\mathcal{X},\mathcal{Y})$
• Validation data: ${\mathcal{D}}_{\text{val}}\in (\mathcal{X},\mathcal{Y})$
• Test data: ${\mathcal{D}}_{\text{test}}\in (\mathcal{X},\mathcal{Y})$

Objective: Develop an algorithm that trains the model $f$ to perform well on test dataset ${\mathcal{D}}_{\text{test}}$.

Supported Pipelines

CLArena supports the following experiment and evaluation pipelines for single-task learning:

• Single-Task Learning Experiment: The experiment for training and evaluating single-task learning models. See Single-Task Learning Experiment.
• Single-Task Learning Evaluation: The evaluation phase for assessing the performance of the trained single-task learning models. See Save and Evaluate Model.