Active learning

What is Active Learning?

Active learning is used in artificial intelligence training. Practically, Active Learning is a machine learning framework where the learning algorithm can directly interactively query a user (teacher or oracle) and use the result to create new data points and label them with the true labels. The process of active learning can also be called optimal experimental design.

The motivation for such learning is a scenario where one has a large pool of unlabelled data. For example, consider the problem of training image classification models to distinguish cats from dogs. There are millions of images of each in the process, but not all of them are needed to train a good model. It is the case that some images contain more clarity and information than others. Similar applications classify the content of YouTube videos, where the data is naturally dense in large numbers.

Whereas in Passive Learning this is a standard framework and with the large amount of labelled data that is passed to the algorithm, a significant effort is required in labelling the entire dataset. On the other hand, with Active Learning, it is like crowd-sourcing that we can selectively use the system to ask the human experts to selectively label some elements in the dataset and it does not need to label the entirety. An algorithm simply iteratively selects the most informative examples based on the value metric. In doing so, it sends the unlabelled examples to a labelling oracle. The latter returns the true labels for the queried examples to this algorithm.

What is Active Learning?

This learning is a teaching method that demands the students' full attention and participation in learning. It involves hands-on learning as opposed to lecture learning or presentation learning. There are different activities and programmes for Active Learning. For example, there is "Intel AI for Youth", "Intel Future Skills" and "e-sports competitions". Technology is certainly an important factor for the active learning approaches and it helps students to acquire new skills with the help of technology and to fundamentally prepare for a career in the fourth industrial revolution.

What are the areas of application?

Artificial intelligence is used to classify land cover types. A manual, complete evaluation is practically no longer feasible because the immense amount of data would be unmanageable. An oracle is also used in this active learning.
The labelling effort of large amounts of data can be reduced by asking the user about specific information and thus this makes the greatest contribution to the learning process. Artificial intelligence is an area of research shared by philosophy, biology, psychology and computer science.

What types of learning methods are there in artificial intelligence?

With Active Learning, the algorithm can respond to the input data and use specified questions to specifically ask for the relevant results. The questions are selected based on the relevance of the results by the algorithm used. The origin of the data is irrelevant. The data can be available offline or online. Data may be used several times for the learning process. In addition, there is also reinforcement learning, unsupervised learning, supervised learning (Supervised Learning) and semi-supervised learning.

Data Navigator Newsletter