The abundance of data created in science, engineering, business, and everyday human activity is simply staggering. This data is often complex and high-dimensional, taking the form of video or time-dependent sensor data. Machine learning methods allow us to understand such data, automatically identifying patterns and making important data-driven decisions without human intervention. Machine learning methods have found a wide variety of applications, including providing new scientific insights and the development of self-driving cars.
One machine learning method in particular, neural networks, has emerged as the preeminent tool for the supervised learning tasks of regression and classification. Loosely modeled after the human brain and the basis for deep learning, Neural Networks use composition to develop complex representations of data. In recent years, researchers using Neural Networks have made tremendous breakthroughs in topics as varied as image processing, natural language processing, and playing board games such as Go.
Undergraduate students participating in this SRI stream will be introduced to neural networks and learn how they're used. Students will learn about the mathematics that forms the basis for neural networks and the optimization methods used to train them. They'll learn how to program in python and use machine learning packages such as scikit-learn and TensorFlow to analyze data. Working in teams, students will use Neural networks to solve real-world classification problems like object recognition in images, detecting falsified financial transactions, and controlling for manufacturing defects. They'll also learn to effectively communicate and visualize results.