Data that typically collected are the symptoms of real-world processes of interest. However our desire in Machine Learning is to model the real-world process, not it’s symptoms. Recent advances in Machine Learning all point to the fact that we need to model hidden variables (the not visible factors that represent the underlying process. In this talk, I demystify hidden variables as used in Recommender systems, natural language processing, image recognition and network analysis and tie it back to a simple formulation of machine learning as search.
Key Takeaway for the Audience:
- Understanding of the commonality and differences between different methods that exist today in a machine learning engineers toolkit for learning lower dimensionality representations of the domain they model.