Jeremy Uzan



Artificial Intelligence enthusiast. Following the first computer music algorithms of pioneer Iannis Xenakis, deep music generative models have emerged and are now considered strong and innovative tools for artists. I truly believe that the perfect framework isn’t just represented by powerful software instruments, but rather pairing them with hardware to obtain an augmented instrument. This would keep the sensation of touch while using the infinite potential of the software inside. With the rise of powerful computers and datasets, deep models are now extremely precise and generate realistic sounds; however, there are consequences due to the size and energy consumption of these models. Conscious of the environmental impact of data centres on global warming, we need to find a path to combine powerful tools and energy consumption efficiency. In order to achieve that, we can rethink the architecture of Deep Learning models to compress them as much as possible without losing precision and efficiency. New compression techniques have emerged and I work on implementing powerful embedded models into tiny computers like Raspberry and small devices like smartphones.