Welcome to the Pixel Proficiency Training Series!
Instructors: Valentina Staneva, Joseph Hellerstein, Nicoleta Cristea
Session Dates: Oct 15 - Nov 19, Wednesdays from 12:30-1:50 p.m.
Oct 15: Intro to neural networks for image classification
Oct 22: Model selection and evaluation
Oct 29: Model tuning and transfer learning
Nov 5: Self-supervised learning through autoencoders
Nov 12: Segmentation and object detection
Nov 19: Addressing common pitfalls
In-person and zoom options available. While we offer a remote option, we encourage in-person participation, when possible.
The series of 6 tutorials will demonstrate how to build neural networks capable of addressing common computer vision tasks such as classifying patterns in images, detecting objects, identifying the boundaries of those objects. These tutorials will be focused on providing more than just a brief introduction to technical tools; attendees will also learn methods to rigorously validate the accuracy of their models and assess how their results generalize in the presence of new data. We will focus on convolutional network architectures and will teach the skills to adopt more complex ones through the Python Keras library.
Participant Expectations: No prior experience with neural networks or related software packages is necessary, though attendees are expected to have some basic experience with Python code and should have some familiarity with one or more basic machine learning approaches, such as logistic regression or random forests. Attendees will use the Keras library to do their work, but learn concepts that are broadly useful regardless of the technology.