Meet the Fellow: Florentin Guth
This entry is a part of our Meet the Fellow blog series, which introduces and highlights Faculty Fellows who have recently joined CDS.
Meet Florentin Guth, who will join CDS this fall as a Faculty Fellow. Guth is currently finishing his PhD at Ecole Normale Supérieure in Paris, under the supervision of Stéphane Mallat. In his research, Guth seeks to understand how deep learning manages to exploit the structure of real-world learning problems and work so effectively in practice.
Guth’s work stands out for its focus on unraveling the mysteries of deep learning. One recent article, “A Rainbow in Deep Network Black Boxes,” addresses the “black box” problem, offering a novel method to understand the internal workings of deep neural networks. Guth and his co-authors introduce a new method to analyze the internal weights of a deep network, which they call the “Rainbow model” — a tool that explains why rotation symmetries arise in hidden layers, and how these symmetries can be taken into account to describe the behavior of the network.
Guth has also recently published with Joan Bruna, who is Joint Faculty at CDS. Guth and Bruna’s collaboration was “Conditionally Strongly Log-Concave Generative Models,” which proposed a new image generation approach that partially bridges the gap between deep image generative models and classical algorithms, addressing the shortcomings of both while retaining their strengths.
“I’m excited to join CDS because it is a truly multidisciplinary environment where everyone brings a different perspective on the same questions,” said Guth. “I’m especially looking forward to interesting conversations around the many mysteries of deep learning.”
More of Guth’s research has been published in Advances in Neural Information Processing Systems. Before focusing on deep learning, Guth also wrote a paper on smart contracts. His unique perspective and expertise will undoubtedly enrich the CDS community.
To view all our current faculty fellows, please visit the CDS Faculty Fellow page on our website.
By Stephen Thomas
