11–26 Nov 2021
Europe/Budapest timezone

VPNet: Variable Projection Networks

Not scheduled
20m
Online lecture

Speaker

Dr Péter Kovács (Department of Numerical Analysis, Faculty of Informatics, Eötvös Loránd University, Budapest, Hungary )

Description

Analysis of signals by means of mathematical transformations proved to be an effective method in various aspects, including filtering, system identification, feature extraction, classification, etc. The most widely used method in transform-domain techniques operates with fixed basic functions like trigonometric functions in the Fourier transform, Walsh functions in Walsh–Fourier transform, mother wavelet function for wavelet transforms, etc. In these cases, the flexibility of the method is in the proper choice of the function system. Once the system is set it is used regardless the difference between the individual signals. In other words, the system can be adjusted on the problem level, but not on the individual signal level. This limitation turned to be significant, especially in dynamically changing environments, when we need to adjust the system to the signal. One way to surpass this limitation is to use adaptive orthogonal transformations. In recent years, we generalized this concept, developed various adaptive mathematical models, and successfully applied them in a range of applications, including ECG, EEG signal processing, telecommunication, CT-, photoacoustic-, and thermographic imaging. In these applications the transformation step was followed by machine learning techniques. Even though the interaction of these two phases were studied and considered they were not integrated into a uniform method. In our recent project, based on our former work, we incorporate the representation abilities of adaptive orthogonal transformations and the prediction abilities of neural networks (NNs) in form of the model called VPNet. This is a novel model-driven NN architecture for 1D signal-processing which utilize variable projection (VP). Applying VP operators to neural networks has the advantage of learnable features, interpretable parameters, and compact network structures. We show that, compared to fully connected and one-dimensional convolutional networks, VPNet offers fast learning ability and good accuracy at a low computational cost of both training and inference.

Title

VPNet: Variable Projection Networks

affiliation Department of Numerical Analysis, Faculty of Informatics, Eötvös Loránd University, Budapest, Hungary
authors Péter Kovács, Gergő Bognár, Sándor Fridli

Primary authors

Prof. Sándor Fridli (Department of Numerical Analysis, Faculty of Informatics, Eötvös Loránd University, Budapest, Hungary ) Dr Péter Kovács (Department of Numerical Analysis, Faculty of Informatics, Eötvös Loránd University, Budapest, Hungary ) Dr Gergő Bognár (Department of Numerical Analysis, Faculty of Informatics, Eötvös Loránd University, Budapest, Hungary )

Presentation materials

There are no materials yet.