24 of 33

|

Mode Connectivity. Optima of complex loss functions connected by simple curves over which training and test accuracy are nearly constant. Visualization data generated through a collaboration between Pavel Izmailov (@Pavel_Izmailov), Timur Garipov (@tim_garipov) and Javier Ideami (@ideami). Based on the NeurIPS 2018 paper by Timur Garipov, Pavel Izmailov, Dmitrii Podoprikhin, Dmitry Vetrov, Andrew Gordon Wilson: https://arxiv.org/abs/1802.10026 | Creative visualization and artwork produced by Javier Ideami.

The use of the imagenette dataset in combination with a resnet and a set of specific hyperparameters produces interesting irregular morphology during the training phase in the areas above the two minima.

Real data, resnet-20 no-skip, imagenette, sgd-mom, bs=128, wd=3e-4, mom=0.9, bn, train mod, 1 million pts, log scaled (orig loss nums)

This piece has been created through a multi day process composed of different stages that begin with the training of deep...

Explorer of A.I, AR, advanced visualizations & others, multidisciplinary engineer & creative director. As a multidisciplinary creative director, engineer, artist and entrepreneur, Javier has travelled the world leading projects, exhibiting his work and giving talks, from Silicon Valley to the jungles of Bali, from Stanford University, through the United Nations FAO HQ and the financial center of London, to the International Cultural Diplomacy Conference in Berlin and many others. In his latest works, Javier Ideami produces A.I artworks through a complex process that can last many days. The process begins with training a neural network and capturing relevant data. This multidimensional data is then transformed into a 3D piece which is...