Explore the implementation of a U-Net Convolutional Neural Network (CNN) in APL without using any frameworks or libraries in this 24-minute conference talk from ARRAY 2023. Delve into how APL notation can be effectively applied to complex neural network architectures like U-Net. Compare the performance and language design of APL against specialized frameworks such as PyTorch for neural network programming. Discover how a modern APL compiler with GPU support, Co-dfns, performs in comparison to current specialized neural network frameworks. Gain insights into the conciseness, clarity, and transparency of APL code for machine learning applications. Consider the implications of this approach for future machine learning language design, pedagogy, and implementation, both within and outside the APL community.
Overview
Syllabus
[ARRAY'23] U-Net CNN in APL: Exploring Zero-Framework, Zero-Library Machine Learning
Taught by
ACM SIGPLAN