Smart phoTonic souRces harnEssing Advanced Multidimensional Light optimization towards machIne-learNing-Enhanced imaging (STREAMLINE)

Modern photonic systems increasingly rely on complex nonlinear optical processes at the foundation of demanding applications spanning advanced light source development, metrology and imaging. Importantly, current flagship imaging systems are based on nonlinear light-matter interactions provided by specialized lasers requiring complex operation and lacking flexibility: means of controlling nonlinear phenomena and interactions are restricted, and reaching the ideal settings for a specific application can prove extremely challenging.

In this context, optical excitations can be inefficient (with e.g. excessive power or spectral coverage) and versatile means to drive coherent control of light properties are highly sought-after, for they provide the main building blocks for advanced imaging techniques. However, such control is currently constrained to few degrees of freedom provided by complex components ultimately hindering the accessible optical parameter space. The realization of versatile, efficient and practical optical sources in compact forms would thus represent a fundamental revolution.

STREAMLINE constitutes an ambitious multidisciplinary program aiming to push forwards the development of 'smart photonic sources' for the creation of a promising new research field merging ultrafast nonlinear optics and computational imaging.

The envisioned architecture, combining integrated and fibered components, will explore new multimode and input-dependent nonlinear dynamics via dedicated machine-learning schemes.

Together with suitable monitoring techniques, fully reconfigurable and tailored optical wavepackets (with ‘on-demand’ spectral, temporal and spatial properties), will be exploited towards disruptive nonlinear imaging and metrology techniques. Besides providing user-friendly operation with improved performances, blueprint dynamical imaging with custom light-matter interactions is expected to unlock access to novel deep-learning strategies towards biological sample histology.