Gradient Descent: The Workhorse (Algorithm) of Machine Learning

UPDATE: LINKS TO PRESENTATION AND CODE BELOW

A pdf of tonights presentation may be found here (https://github.com/jermwatt/mlrefined/blob/master/Gradient_Descent_talk_5_26/metis_ds_talk_grad_descent.pdf)

The jupyter notebook illustrating gradient descent can be found here (https://github.com/jermwatt/mlrefined/blob/master/Gradient_Descent_talk_5_26/demos/grad_descent_demos.ipynb)

------------------------------------------------------------------------

Every machine learning problem has parameters that must be tuned properly to ensure optimal learning. For example, in fitting a line to a dataset of paired inputs and outputs (i.e., linear regression) we want to tune the slope and intercept of the line properly so that the corresponding line best represents our data.

In this talk, Jeremy Watt -- instructor of the upcoming Metis course titled Machine Learning: Algorithms & Applications (http://www.thisismetis.com/machine-learning) and author of Machine Learning Refined (http://www.amazon.com/Machine-Learning-Refined-Foundations-Applications/dp/1107123526) -- will discuss one of the most fundamental – and arguably most commonly used – algorithms in machine learning today: gradient descent (also known as the backpropagation algorithm). This extremely popular scheme is a part of the broader family of algorithms known collectively as mathematical optimization, which are the primary methods used to solve every machine learning problem in existence today (regression, classification, dimension reduction, reinforcement learning). These methods, including the workhorse algorithm gradient descent, are also used in a variety of other fields as well including operations, logistics, economics, and physics as well.

In this tutorial Jeremy will give a big-picture introduction to the family of mathematical optimization as a whole, and gradient descent in particular. Mathematical equations will be kept to an absolute bare minimum, with concepts being elucidated using a variety of engaging custom-made graphics, geometric intuition, and interactive IPython notebook demonstrations.

6:30 - 7:00 Register, Mingle, Eat & Drink
7:00 - 7:45 Jeremy Watt presents
7:45 - 8:00 Chat with Jeremy, attendees, Metis staff, and enjoy more food & drink

http://photos1.meetupstatic.com/photos/event/a/5/9/e/600_447882398.jpeg

Jeremy Watt Metis Machine Learning Course Instructor Machine Learning Consultant and Author of Machine Learning Refined (http://www.cambridge.org/us/academic/subjects/engineering/communications-and-signal-processing/machine-learning-refined-foundations-algorithms-and-applications?format=HB)

Jeremy holds a PhD in Computer Science and Electrical Engineering from Northwestern University where he conducted research in machine learning and computer vision while actively consulting with partners in finance and insurance, as well as startups in the e-commerce and healthcare space. Jeremy is a seasoned and passionate instructor of data science. In addition to authoring his own textbook on machine learning, titled Machine Learning Refined and published by Cambridge University Press, he has designed and taught several large university courses on machine learning, as well as large tutorial short courses on deep learning at major conferences on artificial intelligence and computer vision.

Connect with us before the event! @thisismetis (https://twitter.com/thisismetis)