DeepTeach

MLDB is an open-source data collection for machine learning. DeepTeach is a web app based on MLDB that lets a person train a machine learning model what kinds of pictures they are searching for through a repetitive process. It’s an example of human augmentation because it is used to make people more productive. It showcases the power of intelligence augmentation (IA) since it uses machine learning to augment the intelligence of humans.

DeepTeach was built using JavaScript, Python, HTML, and TypeScript. It uses a deep convolutional neural network (CNN) to generate features. It then uses the input of the user to train a specific decision tree to learn the kind of picture that the user is searching for.

The app shows a small random quantity of unlabelled pictures to the user. The user can then label a number of them by putting them in the positive label section (“Target”) or the negative label area (“Not Target”). The app then trains a model that attempts to learn what each category appears as and recommends new pictures for each label. The user can accept or improve the predictions of the model, and start over from the beginning. Using DeepTeach, a person can train a binary (2) image classifier from unsupervised data in around a minute.