“DEEP learning for INverse problems (DEEP-IN) in Sciences” working group (April 1st, 2024 - )

DEEP-IN Working Group Website


The essence of discovery in sciences has always been rooted in the reverse engineering of natural phenomena and observational data. This paradigm of deducing the underlying laws of nature from observable outcomes forms the cornerstone of our scientific inquiry. The DEEP-IN working group is established with the recognition that the elucidation of such complex phenomena demands a fusion of physics insights and advanced deep learning methodologies. Historically, the exploration of sceinces has relied heavily on intuition and empirical exploration, with methods like inference playing a significant role in our understanding. 

In response to the evolving landscape of scientific research, our objective is to integrate state-of-the-art deep learning techniques, alongside generative models and other advanced statistical learning methods, into the toolkit of scientists. These modern approaches are indispensable not only for analyzing the vast datasets encountered in fields such as particle and nuclear physics, astrophysics, quantum many-body systems, and complex systems, but also for building meaningful physics from data where noise may obscure the underlying signals in fields such as biological systems.

The DEEP-IN working group aims to build a multidisciplinary platform that leverages the power of artificial intelligence to tackle inverse problems that span a diverse spectrum of sciences. By advancing the application of artificial intelligence(AI), including but not limited to deep learning(DL), in both data-rich and data-sparse scenarios, we aspire to unveil new patterns, principles, and understandings for sciences. To enhance the efficacy and reach of the DEEP-IN working group, we are committed to establishing an inclusive group that bridges deep learning methodologies with the biologists and physicists. Our promise is to collaborate by regularly hosting lectures, organizing seminars, and convening workshops. These initiatives are designed not only to facilitate knowledge exchange and skill development but also to catalyze the formation of interdisciplinary partnerships.

Methodologically, our support is structured around three principal pillars: statistical learning methods, deep learning techniques, and generative models. This comprehensively ensures that we cater to the diverse needs and challenges faced by scientists, whether they are part of iTHEMS or from the broader scientific community. Our goal is to offer direct collaborations and assistances that empower memerbers to refine the tools for understanding the physical world, paving the way for groundbreaking discoveries that stir excitement and curiosity.

The DEEP-IN working group aspires to contribute significantly to the evolution of how we explore, decipher, and interpret the mysteries that define our world.

Lingxiao Wang (RIKEN iTHEMS) *Contact at lingxiao.wang@riken.jp
Catherine Beauchemin (RIKEN iTHEMS)
Enrico Rinaldi (Quantinuum K.K./RIKEN iTHEMS)

Related Link