Conference Agenda

In this talk, we review the recent developments in machine learning applications to ultrafast photonics with emphasis on the study of complex dynamics and transient instabilities.

We introduce a new approach based on two neural network architectures for mimicking the nonlinear propagation dynamics of ultrashort pulses in optical fibers for supercontinuum generation, allowing for significant memory and speed improvements compared to the conventional approach of numerically integrating the generalized nonlinear Schrödinger equation.

We demonstrate a technique for the programmable generation and multi-millijoule amplification of ultrashort pulse bursts, which can be applied to any master-oscillator regenerative-amplifier system with very low implementation complexity and high stability in burst-mode operation.

Laser-Induced Aerosol Formation (LIAF), driven by UV and near-IR filaments, relies on the nitrogen photo-oxidative chemistry, triggered by photoionization and leading to the production of HNO3, stabilizing the growth of aerosol. Mid-IR filaments were expected to be less efficient due to their lower photoionization rates. However, we observed surprisingly high yields of aerosols, generated by mid-IR laser pulses, which cannot be fully explained by the HNO3-pathway. Therefore, we suggest a new mechanism of LIAF, based on the resonant excitation of volatile organic compounds, enabled by the spectral broadening during filamentation.

Experiments on molecules in solution showed that the first step of vision consists in an ultrafast photo-isomerization that can be coherently controlled by femtosecond pulse shaping. Here, we measure the electric signals fired from the retina of living mice upon femtosecond multi-pulse and single-pulse visual stimulation. We show that the electrophysiological response is sensitive to the manipulation of the pulse spectral-phase and the control mechanism relies on multiple interactions in proximity of the conical intersection, leading to pump-dump and pump/re-pump processes. Our interpretation is supported by quantum dynamics simulations.

In the framework of the Laser Lightning Rod project, we study over 140 m the filaments created by a laser system with J-range pulses of 1 ps duration at 1 kHz repetition rate. We investigate the spatial evolution of the multiple filamentation regime and its ability to control high-voltage discharges at different distance. The measurements were made using both a collimated beam and a focused beam.

Our work contributes to integrate artificial intelligence into laser devices to make them versatile, adaptable and programmable. Within a fiber laser cavity, we incorporate interfaced liquid-crystal components driven by an evolution algorithm that optimizes merit functions leading to user-defined mode-locked regimes. To illustrate the possibility to generate on-demand complex ultrashort-pulse dynamics, we demonstrate the generation of 2-soliton molecules with pre-determined temporal separation.