Human in the loop

In classical ML approaches, we assume that the data is gathered, and we use well-known paradigms of supervised or unsupervised learning to train our models. However, often, there is a need for humans to be more involved in the process to guide it. One example of this process is the use of active learning.

Active learning

Active learning is an ML paradigm where we don’t just use the dataset as it is, but also use the knowledge of a human to provide labels to selected examples. The human actor in this system is often called an oracle. An oracle provides labels to the data examples that we select based on predefined criteria and thereby, we enable efficiency in data labeling, optimizing the performance of the model using minimal human feedback. The following figure shows a simple schema of an active learning system:

Figure 16.6 – Active learning actors, and steps of data transfer and processing

We often need the time...