Daniel Dimanov (PhD, FST) with this poster entitled: MONCAE: Multi-Objective Neuroevolution of Convolutional Autoencoders.
Click the poster below to enlarge.
With this poster, we present a novel neuroevolutionary method to identify the architecture and hyperparameters of convolutional autoencoders, which has been published in an ICLR workshop. Remarkably, we used a hypervolume indicator employing neuroevolution for in the context of neural architecture search for autoencoders, for the first time to the best of our knowledge. We rely on novel decoding of the architecture to automatically reconstruct the decoder from the encoding. We tested our approach with MNIST, Fashion-MNIST and CIFAR10 to verify the performance of the approach. Results show that images were compressed by a factor of more than 10, while still retaining enough information to achieve image classification for the majority of the tasks. Thus, this new approach can be used to speed up the AutoML pipeline for image compression and much more.
You can view the full poster exhibition and pre-recorded presentations on the conference webpage.