Open Internet by MindsNet

Create Self-Supervised Learning Systems

Develop ML systems that can learn powerful representations from unlabeled data without requiring human annotation, mimicking how humans learn from raw sensory experience. Current supervised learning requires enormous amounts of labeled data, which is expensive and time-consuming to create. Self-supervised learning promises to unlock the vast amounts of unlabeled data available in the world. The challenge involves developing pretext tasks and architectures that can extract meaningful signal from unlabeled data across different domains. Success would dramatically reduce the data requirements for ML and enable learning in domains where labeled data is scarce or impossible to obtain.

Computing & Technology, Computer Science, Machine Learning