The input for a multi-task model is usually one sample with several labels corresponding to different supervised information. Different from the common situation, here we have samples with different single label (i.e., one sample one label), for classification or regression. Thus we design an alternate training method.

At beginning, the training sets containing classification and regression samples are randomly split into multiple batches. Then, each batch of the regression task and the classification task are alternately trained. In other words, alternate training is applied during which the back propagation algorithm selects different branches to update according to different sources of data. In each training batch of regression/classification task, the shared weights and the specific weights are updated. Similar to the training process of single-task model, we select the final models based on the minimum values on their corresponding validation sets.

Note: The content above has been extracted from a research article, so it may not display correctly.



Q&A
Please log in to submit your questions online.
Your question will be posted on the Bio-101 website. We will send your questions to the authors of this protocol and Bio-protocol community members who are experienced with this method. you will be informed using the email address associated with your Bio-protocol account.



We use cookies on this site to enhance your user experience. By using our website, you are agreeing to allow the storage of cookies on your computer.