Explore the implementation of the U-Net convolutional neural network architecture in APL through this conference talk from Dyalog '22. Dive into the collaboration between Rodrigo Girão Serrão and Aaron Hsu as they demonstrate how APL's unique features make this neural network highly effective for experimentation and problem-solving in AI. Learn about the U-Net architecture, its applications in image segmentation, and how to arrange it using matrices and rectangular arrays. Gain insights into inner products, axes, and optimized cases of the stencil operator. Discover how understanding APL solutions can enhance your programming skills and problem-solving abilities in machine learning and deep learning tasks.
Overview
Syllabus
Introduction
What is a U-net Convolutional Neutral Network CNN?
Motivation and method
How to arrange a u-net as matrices rectangular arrays
What else did I learn?
Inner products and axes
Optimised cases of the stencil operator
Conclusion: Understanding APL solutions gives you super powers
Taught by
Dyalog User Meetings
