Periodic Reporting for period 4 - SMARTIES (Scattering Media as a Resource Towards Information Processing and Sensing)
Período documentado: 2021-10-01 hasta 2022-09-30
The project has revolved about exploiting the complexity of light propagation in complex media with advanced signal processing tools, in order to develop a new paradigm for optics in complex media : using scattering not as a nuisance, but as a resource, by exploiting the randomness of the scattering process to one's advantage.
During the project, we have explored multiple aspects of the paradigm : For imaging, we have shown for instance that we can record neuronal activity at depth, by demixing the fluorescence emitted by neurons when they emit an action potential, but also developed many techniques to computationally reconstruct images of object buried deep behind a scatterer, using multiple contrast mechanism, from linear fluorescence and Raman to non-linear contrast mechanisms such as 3-photon fluorescence. In Optical computing, we have demonstrated the use of optical reservoir computing for large scale spatiotemporal time series prediction, as well as for simulating an Ising Hamiltonian. In the quantum realm, we have shown how a complex medium (a multimode fiber) can be used a reconfigurable linear circuit for multi-photon interference, which is of interest in quantum information processing and computing.
Overall, the project has therefore delivered important results in multiple directions, of high fundamental and practical importance, with potential societal impact in the long run, from biology and medecine to information technologies.
During the project, we have explored multiple aspects of the paradigm : For imaging, we have shown for instance that we can record neuronal activity at depth, by demixing the fluorescence emitted by neurons when they emit an action potential, but also developed many techniques to computationally reconstruct images of object buried deep behind a scatterer, using multiple contrast mechanism, from linear fluorescence and Raman to non-linear contrast mechanisms such as 3-photon fluorescence. In Optical computing, we have demonstrated the use of optical reservoir computing for large scale spatiotemporal time series prediction, as well as for simulating an Ising Hamiltonian. In the quantum realm, we have shown how a complex medium (a multimode fiber) can be used a reconfigurable linear circuit for multi-photon interference, which is of interest in quantum information processing and computing.
Overall, the project has therefore delivered important results in multiple directions, of high fundamental and practical importance, with potential societal impact in the long run, from biology and medecine to information technologies.
The work has been progressing quite well, all three major axis have been developing, with strong synergies and interactions between the people involved on the various projects.
-On the imaging side, we have been focusing on linear fluorescence imaging, with several computational methods to focus light on fluorescence targets, and on functional activity reconstruction, simulating fluorescence pattern from neuronal activity. In all this work, we have strongly borrowed from the optical computing effort on the algorithmic front. We have also taken inspiration and insight from mesoscopic physics, for instance on our work on the stability of the focus in a dynamical work, and trying to expand our imaging capability beyond what is possible with the memory effect.
-On the optical computing, we have developped a novel setup to explore optical reservoir computing, and are now trying to tackle more challenging machine learning problem than the simple time-series. Results are encouraging.
-On the quantum computing part, we have reached a major milestone by demonstrating the programmability of simple linear circuits in a multimode fibers are and now trying to expand it further and apply it to simple computing problem well adapted to our approach, in particular variational Eigenvalue solver problems.
-On the imaging side, we have been focusing on linear fluorescence imaging, with several computational methods to focus light on fluorescence targets, and on functional activity reconstruction, simulating fluorescence pattern from neuronal activity. In all this work, we have strongly borrowed from the optical computing effort on the algorithmic front. We have also taken inspiration and insight from mesoscopic physics, for instance on our work on the stability of the focus in a dynamical work, and trying to expand our imaging capability beyond what is possible with the memory effect.
-On the optical computing, we have developped a novel setup to explore optical reservoir computing, and are now trying to tackle more challenging machine learning problem than the simple time-series. Results are encouraging.
-On the quantum computing part, we have reached a major milestone by demonstrating the programmability of simple linear circuits in a multimode fibers are and now trying to expand it further and apply it to simple computing problem well adapted to our approach, in particular variational Eigenvalue solver problems.
We have already demonstrated several advances, well beyond the current state of the art, in all 3 axes of the project. We plan to push this further in the second half of the project, with the goal to demonstrate:
-for imaging: an unifying approach to computational imaging in scattering media, based on algorithmic developments (phase retrieval, Non-negative matrix factorization, deep learning), and the development of a proof of principle experiment of non-invasive neuronal activity recording at depth beyond what is currently achievable by optical means.
-for computing, a speedup of our experimental setup and its application to large datasets, that are not easily amenable to in-silico calculation using GPUs or CPUs due to their large size. If possible, we will explore the technology transfer of these results to our spin-off company.
-for quantum computing, we plan to achieve reconfigurable linear networks at the state of the art of on-chip circuits, and to the implementation of quantum algorithms on this platform.
-for imaging: an unifying approach to computational imaging in scattering media, based on algorithmic developments (phase retrieval, Non-negative matrix factorization, deep learning), and the development of a proof of principle experiment of non-invasive neuronal activity recording at depth beyond what is currently achievable by optical means.
-for computing, a speedup of our experimental setup and its application to large datasets, that are not easily amenable to in-silico calculation using GPUs or CPUs due to their large size. If possible, we will explore the technology transfer of these results to our spin-off company.
-for quantum computing, we plan to achieve reconfigurable linear networks at the state of the art of on-chip circuits, and to the implementation of quantum algorithms on this platform.