The Energy Landscape of Machine Learning Algorithms

Machine learning is a fascinating subject both from the theoretical and practical point of view. Although deep neural networks were introduced decades ago, only recently their implementation became possible thanks to an increased computational power combined with a huge amount of data (used as learning set). This lead to a real revolution that is still bursting right now. Almost every day a new ground-breaking application of machine learning–thought to be impossible just a few years ago–comes out. Despite their success, the theoretical understanding of deep neural networks is quite poor. Practitioners have developed recipes to construct and train them but fundamental questions remain opened. Answering them not only has the potentiality of leading to great improvements as well as avoiding dramatic pitfalls but it is a fascinating scientific subject at…


Link to Full Article: The Energy Landscape of Machine Learning Algorithms