Self-supervised machine learning, the next frontier in AI

Current deep learning approaches require large amounts of labeled data. The creation of labeled data is expensive, error-prone, and time-consuming. Despite these challenges, in the last decades, tremendous successes in machine learning have been achieved in the area of supervised learning that requires the compilation of large datasets with labels (for example, grouping pictures based on the person in the image). In contrast, unsupervised learning algorithms do not require labels and require minimal human participation. However, due to significant technical difficulties, they haven’t been as successful as supervised learning algorithms. Self-supervision overcomes these technical difficulties to extract value from very large unlabeled datasets using machine learning with minimal human intervention in cybersecurity, precision medicine, and predictive maintenance applications. The new method circumvents these difficulties and clears the way to scaling unsupervised learning algorithms to large and complex datasets. We will show applications of these methods in cybersecurity and drug discovery problems.