Sometimes, Machine Learning models seem like magic. They can produce new patterns, texts, or images seemingly out of nowhere. Powered by massive parallel computation, these algorithms reveal hidden structures in datasets so large and complex that they look like noise to us. The technology’s ability to see what is invisible to us makes it difficult to grasp what happens inside the 'black box.' Noise to Signal explores the principles behind Machine Learning algorithms to expose their inner workings. Inspired by functions like Classification, Regression, Clustering, Object Detection and Denoising, it aims for a more intuitive relation to the technology and look for beauty in the black box.
Installed in the public area of the BEDI AI datacenter, the work consists of two parts. A large panoramic screen depicts a machine learning landscape, a window into the mind of a machine trying to find patterns in unstructured data. Informed by real time performance of the datacenter (power usage, CPU and GPU usage and data throughput), the back reveals the inner mechanics of the massive parallel computing behind it in more detail, showing on three screens the relationship between sequential and parallel processing, and network traffic as dynamic treemaps depicting the clusters, nodes, and cores that make up the system.