[Seminar] Balanced Mixture of SuperNets for Learning the CNN Pooling Architecture (AutoML 2023 paper)

The next LIVIA seminar will be held on Thursday, June 29 at 12h00 in hybrid mode.

 

Title: Balanced Mixture of SuperNets for Learning the CNN Pooling Architecture  (AutoML 2023 paper)

by Mehraveh Javan Roshtkhari, student at the LIVIA

 

Abstract: 

Down sampling layers, including pooling and strided convolutions, are crucial components of the convolutional neural network architecture that determine both the granularity/scale of image feature analysis as well as the receptive field size of a given layer. To fully understand this problem, we analyse the performance of models independently trained with each pooling configurations and show that the position of the down sampling layers can highly influence the performance of a network and predefined down sampling configurations are not optimal. Network Architecture Search (NAS) might be used to optimize down sampling configurations as an hyperparameter. However, we find that common one-shot NAS based on a single SuperNet does not work for this problem. Finally, we propose a balanced mixture of SuperNets that automatically associates pooling configurations to different weight models and helps to reduce the weight-sharing and inter-influence of pooling configurations on the SuperNet parameters..

 

Paper:  https://arxiv.org/abs/2306.11982

* In person: ETS-LIVIA, room A-3600.

* Zoom link: https://etsmtl.zoom.us/j/84820130813

Meeting ID: 848 2013 0813