flâneur

A Few Words on Representation Learning - Thalles' blog

sthalles.github.io · 2,981 words · saved by 1 readers

In the last two decades, the field of Artificial Intelligence (AI) has seen substantial development in research and industrial applications. Machine learning algorithms such as Logistic Regression and Naive Bayes can recognize patterns from a set of features and solve problems that seemed otherwise impossible to be solved by hard-coding knowledge into expert systems. These shallow learning algorithms can perform relatively complex tasks such as product recommendation or learn to distinguish between spam from not-spam emails. More interesting, the performance of these shallow machine learning algorithms massively depends on the representations they receive as input. For instance, if we decide to build a spam email detector using Naive Bayes, passing a large body of raw unstructured email data to the classifier will not help. Instead, we need to find a different way to represent the text before feeding it to the classifier. Notably, one commonly used text representation for this kind of

Source Image: Introducing Activation Atlases Introduction In the last two decades, the field of Artificial Intelligence (AI) has seen substantial development in research and industrial applications. Machine learning algorithms such as Logistic Regression and Naive Bayes can recognize patterns from a set of features and solve problems that seemed otherwise impossible to be solved by hard-coding knowledge into expert systems. These shallow learning algorithms can perform relatively complex tasks such as product recommendation or learn to distinguish between spam from not-spam emails. More intere

related reading