|Department||Electrical and Computer Engineering|
Fast changing computer architectures pose a difficult task for developers of high performance software. An important example is the area of DSP (digital signal processing) transforms and algorithms, which are typically used in performance-critical and data-intensive applications. SPIRAL is s system that automatically implements platform-adapted libraries of DSP algorithms by combining methods and knowledge from machine learning, compiler technology, computer architecture, mathematics, and signal processing.
SMART uses methods from algebra and representation theory to answer fundamental questions of digital signal processing such as the interplay between statistical signal models, signal extensions, boundary conditions, transforms, and their fast algorithms. One application is the automatic derivation of DSP transforms and their fast algorithms.