Shawn’s Blog
  • πŸ—‚οΈ Collections
    • πŸ–₯️ Slides Gallery
    • πŸ§‘β€πŸ³οΈ Cooking Ideas
    • 🍱 Cookbook
    • πŸ’¬ Language Learning
    • 🎼 Songbook
  • βš™οΈ Projects
    • βš› Continual Learning Arena
  • πŸ“„ Papers
    • AdaHAT
    • FG-AdaHAT
  • πŸŽ“ CV
    • CV (English)
    • CV (Mandarin)
  • About
  1. Multi-Task Learning (MTL)
  2. Configure MTL Experiment
  3. MTL Dataset
  • Welcome to CLArena
  • Get Started
  • Continual Learning (CL)
    • Configure CL Main Experiment
      • Experiment Index Config
      • CL Algorithm
      • CL Dataset
      • Backbone Network
      • Optimizer
      • Learning Rate Scheduler
      • Trainer
      • Metrics
      • Lightning Loggers
      • Callbacks
      • Other Configs
    • Save and Evaluate Model
    • Full Experiment
    • Output Results
  • Continual Unlearning (CUL)
    • Configure CUL Main Experiment
      • Experiment Index Config
      • Unlearning Algorithm
      • Callbacks
    • Full Experiment
    • Output Results
  • Multi-Task Learning (MTL)
    • Configure MTL Experiment
      • Experiment Index Config
      • MTL Algorithm
      • MTL Dataset
      • Backbone Network
      • Optimizer
      • Learning Rate Scheduler
      • Trainer
      • Metrics
      • Callbacks
    • Save and Evaluate Model
    • Output Results
  • Single-Task Learning (STL)
    • Configure STL Experiment
      • Experiment Index Config
      • STL Dataset
      • Backbone Network
      • Optimizer
      • Learning Rate Scheduler
      • Trainer
      • Metrics
      • Callbacks
    • Save and Evaluate Model
    • Output Results
  • Implement Your Modules (TBC)
  • API Reference
  1. Multi-Task Learning (MTL)
  2. Configure MTL Experiment
  3. MTL Dataset

Configure MTL Dataset (MTL)

Modified

August 16, 2025

The multi-task learning dataset consists of multiple datasets corresponding to multiple learning tasks, each of which has their own training, validation and test data.

MTL dataset is a sub-config under the experiment index config (MTL). To configure a custom MTL dataset, you need to create a YAML file in mtl_dataset/ folder.

MTL dataset can also be converted from continual learning dataset. To do this, specify /cl_dataset field rather than /mtl_dataset field in the experiment index config (MTL), and create a YAML file in cl_dataset/ folder. Please refer to Configure CL Dataset (CL Main) for detailed instructions.

Below shows examples of the MTL dataset config.

Example

configs
β”œβ”€β”€ __init__.py
β”œβ”€β”€ entrance.yaml
β”œβ”€β”€ experiment
β”‚   β”œβ”€β”€ example_stl_train.yaml
β”‚   └── ...
β”œβ”€β”€ mtl_dataset
β”‚   └── multi_domain_sentiment.yaml
...
configs/experiment/example_mtl_train.yaml
defaults:
  ...
    - /mtl_dataset: multi_domain_sentiment.yaml
  ...
configs/mtl_dataset/multi_domain_sentiment.yaml
_target_: clarena.mtl_datasets.MultiDomainSentiment
(TBC)

Supported MTL Datasets & Required Config Fields

All CL datasets in CLArena can be converted into MTL datasets. Please refer to Supported CL Datasets & Required Config Fields.

In CLArena, we also implemented many MTL datasets as Python classes in clarena.mtl_datasets module that you can use for your experiment.

To choose a MTL dataset, assign the _target_ field to the class name of the MTL dataset. For example, to use the Multi-Domain Sentiment dataset, set _target_ field to clarena.mtl_datasets.MultiDomainSentiment. Each MTL dataset has its own hyperparameters and configurations, which means it has its own required fields. The required fields are the same as the arguments of the class specified by _target_. The arguments of each MTL dataset class can be found in API documentation.

API Reference (MTL Datasets) Source Code (MTL Datasets)

Below is the full list of supported MTL datasets. We only support image classification datasets. Note that the β€œMTL Dataset” is exactly the class name that the _target_ field is assigned.

MTL Dataset Description Required Config Fields

MultiDomainSentiment

(TBC)

TBC. Same as PermutedArabicHandwrittenDigits class arguments
Back to top
MTL Algorithm
Backbone Network
 
 

©️ 2025 Pengxiang Wang. All rights reserved.