What’s new in BNNS Graph

About this session

The BNNS Graph Builder API now enables developers to write graphs of operations using the familiar Swift language to generate pre- and post-processing routines and small machine-learning models. BNNS compiles graphs ahead of execution and supports real-time and latency-sensitive use cases such as audio processing. In this session, we revisit last year's bit-crusher example and simplify the Swift component by removing the reliance on a separate Python file and instead implement the audio effect entirely in Swift. The BNNS Graph Builder API is also suited to pre-processing image data before passing that data to a machine learning model. The session also includes a demonstration of clipping the transparent pixels from an image with an alpha channel.