Hi! I saw this video and it is still unclear for me when and what should I use
Hi, when you want to have multiple outputs from you model also refered to as multitasking here, check out YOLO algorithm, it gives quite a few outputs at the same time (the class, the boubding box coordinates etc.)
Trasfer learning is used when you already have available a pretrained model on a large dataset, then you can slightly change it (usually towards the last few layers) and train it again with your limited dataset. This is normally done when you a have a pretty small dataset to train on.
Dear @someone555777 ,
Thank you for asking this question. In my reply I will do my best to explain Multi-task learning and Transfer learning along with examples about when they should be used.
- Multi-task learning trains a model on multiple tasks simultaneously. This can be beneficial if the tasks are related, as the model can learn shared features that can be used to improve performance on all of the tasks. For example, a model trained to classify images of cats and dogs could also be used to classify images of other animals, such as horses and cows.
- Transfer learning trains a model on a source task and then uses that knowledge to improve the performance of a model on a target task. This can be beneficial if the source task is similar to the target task, but there is not enough data available to train a model from scratch on the target task. For example, a model trained to classify images of cats on a large dataset of cat images could be used to improve the performance of a model on a smaller dataset of dog images.
Please feel free to ask a followup question if you feel uncertain about when each method should be used.
Best,
Can
so, what is the main difference between them in short? That multi-task gives opportunity to work with data, that not very similar with main task?