I have been trying to figure out whether if I train a model and then while predicting is it possible to train images too just like humans Somehow converting valid images to the dataset by asking us when an object is shown Like the Google Photos somehow they ask us if they predicted a face correctly and then reinforces on it


1 Answer 1


Yes, this method of training a model is commonly known as online learning and specific learning algorithms have been designed for this purpose, such as, Stochastic Gradient Descent(SGD). As opposed to Batch Gradient descent, which computes gradients over the entire training set at each step, the SGD algorithm computes gradients for individual samples and updates the model's parameters.

The online learning strategy is not specific to Reinforcement Learning(RL) methods but is also suitable for the broader classification of Machine Learning algorithms. Suppose, we are given a simple model pretrained on a large Face Recognition dataset, and wish the model to continuously adapt to new faces, we may keep feeding in the new input images with their correct labels (obtained from feedback), as they appear, and apply SGD to incrementally update the model. To go about this using an RL based approach, we may penalize the model with negative rewards for every label misclassified and update it.

  • $\begingroup$ During prediction, if we are using the camera to detect the person instead of uploading picture to predict. Can we still make the algorithm learn? And what if the model would be detecting incorrectly $\endgroup$
    – Mikul Rai
    Oct 15, 2020 at 7:43
  • $\begingroup$ Yes, a camera can be thought of as a device capturing a sequence of images. As long as there is some mechanism to tell that the model detected incorrectly, you should be able to make it learn. $\endgroup$ Oct 15, 2020 at 10:04

Not the answer you're looking for? Browse other questions tagged .