We are delighted to announce that a group from MLJC including Luca Bottero, Francesco Calisto and Valerio Pagliarino will present their work at the world-leading conference NeurIPS 2022 (Neural Information Processing Systems) in the Neur-Reps (https://www.neurreps.org/) (Symmetries and Geometry in Neural Representations) workshop. The conference will take place in New Orleans, Louisiana (USA) from November, 27th to December 3rd.
The project is inserted in the landscape of Geometric Deep Learning (GDL), a relatively recent branch of Machine Learning aiming at studying how to enforce geometrical structures and priors into ML models, whit the final goal of increasing performances, generalization capabilities and explainability.
This specific work, titled Unsupervised Learning of Geometrical Features from Images by explicit Group Action Enforcement, has the ability of automatically disentangling the geometric rototranslational and scaling features from the intrinsic ones, when creating a latent representation of a dataset of input images. This line of research may lead in future to the development of more powerful and efficient architectures for machine vision, with applications ranging from self-driving vehicles to medical imaging, with better generalization capabilities.
In this work we propose an autoencoder architecture capable of automatically learning meaningful geometric features of objects in images, achieving a disentangled representation of 2D objects. It is made of a standard dense autoencoder that captures the deep features identifying the shapes and an additional encoder that extracts geometric latent variables regressed in an unsupervised manner. These are then used to apply a transformation on the output of the deep features decoder. The promising results show that this approach performs better than a non-constrained model having more degrees of freedom.
Keep following our website for updates and reporting from the conference!
Special thanks for the support to: University of Turin, HPC4AI, NPO Torino, Pompei Student Lab