Artificial intelligence and deep learning are transforming photonics by enabling the modeling of complex light-matter interactions, the design of advanced photonic structures, and the extraction of meaningful signals from high-dimensional data. In our group, we apply deep learning to ultrashort laser pulse characterization and to the data-driven design and optimisation of plasmonic and nanophotonic structures.
Ultrashort Laser Pulse Characterisation
Ultrashort laser pulses underpin applications from ultrafast imaging to high-harmonic spectroscopy, but their characterization is a fundamental challenge: pulses too brief for direct electronic measurement must be reconstructed indirectly. We apply deep learning to the FROG (Frequency-Resolved Optical Gating) technique, training models to retrieve the full amplitude and phase of a pulse from its spectrogram — delivering fast, accurate characterization that keeps pace with experimental demands.
Design and Optimisation of Nanophotonic Structures
Designing plasmonic and nanophotonic structures for targeted optical performance involves navigating a large, multi-dimensional parameter space. We combine electromagnetic simulations (FDTD) with artificial neural networks trained on simulation data, creating surrogate models capable of predicting optical responses — such as near-field enhancement or SERS efficiency — across parameter combinations orders of magnitude faster than direct simulation. This approach is applicable to a broad range of plasmonic and nanophotonic systems, enabling systematic, computationally efficient optimisation that guides experimental fabrication.
Collaboration:
Debrecen University, Theoretical Physics Department, Debrecen, Hungary