Principles and metrics of extreme learning machines using a highly nonlinear fiber.

Optical computing offers potential for ultra high-speed and low-latency computation by leveraging the intrinsic properties of light, such as parallelism and linear as well as nonlinear ultra-high bandwidth signal transformations. Here, we explore the use of highly nonlinear optical fibers (HNLFs) as platforms for optical computing based on the concept of extreme learning machines (ELMs). To evaluate the information processing potential of the system, we consider both task-independent and task-dependent performance metrics.

Limits of nonlinear and dispersive fiber propagation for an optical fiber-based extreme learning machine.

We report a generalized nonlinear Schrödinger equation simulation model of an extreme learning machine (ELM) based on optical fiber propagation. Using the MNIST handwritten digit dataset as a benchmark, we study how accuracy depends on propagation dynamics, as well as parameters governing spectral encoding, readout, and noise. For this dataset and with quantum noise limited input, test accuracies of over 91% and 93% are found for propagation in the anomalous and normal dispersion regimes, respectively.

Spectral optimization of supercontinuum shaping using metaheuristic algorithms, a comparative study.

Supercontinuum generation in optical fiber involves complex nonlinear dynamics, making optimization challenging, and typically relying on trial-and-error or extensive numerical simulations. Machine learning and metaheuristic algorithms offer more efficient optimization approaches. We report here an experimental study of supercontinuum spectral shaping by tuning the phase of the input pulses, different optimization approaches including a genetic algorithm, particle swarm optimizer, and simulated annealing.

Optimization and realignment of OAM mode excitation in ring-core optical fibers using machine learning.

Light beams carrying orbital angular momentum (OAM) in free space or within optical fibers have a wide range of applications in optics; however, exciting these modes with both high purity and low loss generally requires demanding optimization of excitation conditions in a high dimensional space. Furthermore, mechanical drift can significantly degrade the mode purity over time, which may limit practical deployment of OAM modes in concrete applications. Here, combining an iterative wavefront matching approach and a genetic algorithm, we demonstrate rapid and automated excitation of OAM modes with optimized purity and reduced loss.

Tailored supercontinuum generation using genetic algorithm optimized Fourier domain pulse shaping.

We report the generation of a spectrally tailored supercontinuum using Fourier-domain pulse shaping of femtosecond pulses injected into a highly nonlinear fiber controlled by a genetic algorithm. User-selectable spectral enhancement is demonstrated over the 1550-2000-nm wavelength range, with the ability to both select a channel with target central wavelength and bandwidth in the range of 1-5 nm. The spectral enhancement factor relative to unshaped input pulses is typically ∼5-20 in the range 1550-1800 nm and increases for longer wavelengths, exceeding a factor of 160 around 2000 nm.

Genetic algorithm optimization of broadband operation in a noise-like pulse fiber laser.

The noise-like pulse regime of optical fiber lasers is highly complex, and associated with multiscale emission of random sub-picosecond pulses underneath a much longer envelope. With the addition of highly nonlinear fiber in the cavity, noise-like pulse lasers can also exhibit supercontinuum broadening and the generation of output spectra spanning 100's of nm. Achieving these broadest bandwidths, however, requires careful optimization of the nonlinear polarization rotation based saturable absorber, which involves a very large potential parameter space.

Feed-forward neural network as nonlinear dynamics integrator for supercontinuum generation: erratum.

We present an erratum to our Letter [Opt. Lett.47, 802 (2022)10.

Feed-forward neural network as nonlinear dynamics integrator for supercontinuum generation.

The nonlinear propagation of ultrashort pulses in optical fibers depends sensitively on the input pulse and fiber parameters. As a result, the optimization of propagation for specific applications generally requires time-consuming simulations based on the sequential integration of the generalized nonlinear Schrödinger equation (GNLSE). Here, we train a feed-forward neural network to learn the differential propagation dynamics of the GNLSE, allowing emulation of direct numerical integration of fiber propagation, and particularly the highly complex case of supercontinuum generation.