Loss

The mission

It is my passion to blend the research and technical side of A.I with the creative arts in order to produce high quality visualizations of deep learning optimization processes that can both inspire and help trigger new insights. As the project continues, new collaborations are arising with researchers at different institutions. In this way, we continue the mission of bringing a sense of beauty to A.I, combining rigurious research with very high resolution and quality visualizations that can bring new light and perspective onto some of the mysteries of deep learning training processes.

Studying the morphology and dynamics of the loss landscape is very expensive in different ways. It is a process that requires enormous amounts of computing power as well as time. In order to scale and take this project even further, I will need way more computing power and way more time. That’s why I welcome all suggestions and ideas regarding possibilities that may eventually allow me to dedicate more time and resources in order to take the loss landscape project way further as well as to share in parallel visualizations and related research with the community. Recently, after receiving some requests for mounted wall art of these visualizations, I have also launched the web ideami.pixels.com, where people may acquire some of these visualizations as wall art and in other forms, this will also contribute to sustain the project.

The second phase of the Loss Landscape project is currently ongoing. The sky is the limit! Thank you for any suggestions. You can contact me at ideami@ideami.com. My central website is at ideami.com and ideami.com/ideami

Javier Ideami

  

MEDIUM

Writings by Ideami at Medium

Gratitude

This project exists because I was inspired by two important groups of people to which I am grateful: the phenomenal Fast.ai community, and the team formed by Hao Li, Zheng Xu, Gavin Taylor, Christoph Studer and Tom Goldstein, authors of the paper that sparked my initial interest in the world of Loss Landscapes. To you both, thank you for inspiring the deep learning community with your teachings, research, spirit and ventures.

xyz

Going deep