Work

Leverage Score Sampling for Complete Mode Coverage in Generative Adversarial Networks

Year: 2021

Type: preprint

Abstract: Commonly, machine learning models minimize an empirical expectation. As a result, the trained models typically perform well for the majority of the data but the performance may deteriorate in less dense regions of the dataset. This issue also arises in generative modeling. A generative model may overlook underrepresented modes that are less frequent in the empirical data distribution. This problem is known as complete mode coverage. We propose a sampling procedure based on ridge leverage scores which significantly improves mode coverage when compared to standard methods and can easily be combined with any GAN. Ridge leverage scores are computed by using an explicit feature map, associated with the next-to-last layer of a GAN discriminator or of a pre-trained network, or by using an implicit feature map corresponding to a Gaussian kernel. Multiple evaluations against recent approaches of complete mode coverage show a clear improvement when using the proposed sampling strategy.

Source: arXiv (Cornell University)

DOI: 10.48550/arxiv.2104.02373

Authors: Schreurs, Joachim, De Meulemeester, Hannes, Fanuel, Michaël, De Moor, Bart, Suykens, Johan A. K.

Institutions: KU Leuven, Centre National de la Recherche Scientifique, Université de Lille, Centre de Recherche en Informatique, Signal et Automatique de Lille, Centrale Lille Institut

Language: English

Topic: Generative Adversarial Networks and Image Synthesis

Subfield: Computer Vision and Pattern Recognition

Field: Computer Science

Domain: Physical Sciences

Open Access status: green

Funders: Agentschap Innoveren en Ondernemen, Fonds Wetenschappelijk Onderzoek, KU Leuven, Kom op tegen Kanker, Vlaamse regering

Awards: G1214117848, G1371775621, G1431749042, The Library of Voices - Unlocking the Alamire Foundation’s Music Heritage Resources Collection through Visual and Sound Technology, G30912928, G3092313991, Structured low-rank matrix / tensor approximation: numerical optimization-based algorithms and applications, G338995740, G3438164672, Next-generation pathology by MILAN: multiplex immunohistochemistry and advanced trainable image analysis, G4466228027, nD: Multi-dimensional dynamical models, (multi-)linear numerical algorithms and tensor data., G4829754272, The Leuven Chansonnier: Researching and Valorizing Franco-Flemish Polyphony with High-Technology Tools, G5103706795, G5109153945, G5167584069, Innovative research and valorization of plainchant through digital technology (THE SOUND OF MUSIC)., G5397124988, AI Tools for Curation of Clinical Registries, Optimization frameworks for deep kernel machines, G6219748686, Globally optimal least-squares system identification and model reduction, Direct detection and identification of neuro-oncology markers in brain tumours, G7418247666, G894091104, G907169927

OpenAlex ID: W4287239759

Cites:

Cited by:

Related to: 10