Consider a scenario where I have a model trained on gesture videos (say a 3D ResNet). I am looking for a technique (or a combination) that allows me to further train the model every time I have a new unlabelled datapoint while not resorting to the entire original dataset the model was trained on. In other words I want to utilize the incrementally collected unlabelled data as fast and efficiently as possible.

To my understanding I need a combination of incremental and semi-supervised or unsupervised learning. The latter one to label the unlabelled data and the former to train the model to the new classes.

Any insight is welcome. Thank you.



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.