What should be cleaned up or changed:
Add a module of task allocation and move related codes from task mining to task allocation.
Why is this needed:
In multi-task learning, The module of task allocation is missing, which serves as the predictor of the task definition.
Given samples, task allocation assigns the corresponding task to the sample. Besides, lifelong learning also needs to use this multi-task learning library.
In fact, the task allocation codes exist in the library but are currently mixing up with the module of task mining.
Given tasks. the task mining is to reveal the task relation in the inference, e.g., the predictor of task relationship discovery.
As we can see, the input and output of the task allocation are both different from task mining. Mixing up the two modules will directly lead to the mixing up of tasks and task groups (relative tasks). That would be followed by the contamination of samples in a task, where other task samples that do not belong to this task are put in the task, making the lifelong process for task management chaotic or even meaningless.
What should be cleaned up or changed: Add a module of task allocation and move related codes from task mining to task allocation.
Why is this needed: In multi-task learning, The module of task allocation is missing, which serves as the predictor of the task definition. Given samples, task allocation assigns the corresponding task to the sample. Besides, lifelong learning also needs to use this multi-task learning library.
In fact, the task allocation codes exist in the library but are currently mixing up with the module of task mining. Given tasks. the task mining is to reveal the task relation in the inference, e.g., the predictor of task relationship discovery.
As we can see, the input and output of the task allocation are both different from task mining. Mixing up the two modules will directly lead to the mixing up of tasks and task groups (relative tasks). That would be followed by the contamination of samples in a task, where other task samples that do not belong to this task are put in the task, making the lifelong process for task management chaotic or even meaningless.