Reservoir computing (RC) is a neuromorphic machine learning paradigm that is ideal for temporal signal processing and is suitable for analog implementations in physical substrates, including photonic devices. RC has been extended to quantum systems due to the enhanced capabilities provided by an enlarged Hilbert space. In that regard, quantum reservoir computing (QRC) has the advantage of avoiding barren plateaus during training. Photonic architectures have already been studied for QRC applications. In our research, we propose a scalable quantum photonic platform for QRC that is suitable for solving temporal tasks. The physical substrate of our reservoir is an optical pulse, which recirculates through an optical cavity with losses, thus creating a quantum memory. The dissipation device (a beam-splitter) also allows the injection of external information and the weak monitoring of the reservoir. Our work focuses on the ability to process classical signals in real time by creating a physical ensemble of identical pulses inside a fiber and the noise robustness of our architecture by tuning the squeezing produced inside the optical cavity.
Photonic quantum technologies are noteworthy candidates in the achievement of quantum advantage for quantum information processing. Moreover, their capabilities for fast signal processing have attracted the interest of researchers in the field of quantum reservoir computing (QRC). In our research, we propose a scalable quantum photonic platform for QRC suitable for solving temporal tasks. In our platform, an optical pulse recirculating through an optical cavity creates a quantum memory, thus not needing external classical storage. A classical signal is sequentially encoded in the quantum field fluctuations of external optical pulses, which interact with the cavity pulse using a beam-splitter (BS). A nonlinear crystal is placed inside the cavity to generate non-trivial dynamics and create a quantum network of entangled modes. A homodyne detector is placed at one of the output paths of the BS for sequential data collecting. Our work focuses on the ability to process classical signals in real time and the noise robustness of our architecture.
KEYWORDS: Semiconductor lasers, Data processing, Systems modeling, Mathematical optimization, Laser systems engineering, Education and training, Computing systems, Transceivers, Signal processing, Photonics systems
Photonic systems, exhibiting multi-gigahertz bandwidth, facilitate data transmission at gigabit-per-second rates. While traditionally used in optical communication for data transfer, semiconductor lasers are now being explored for their potential in optical computation and signal processing. Injecting information into these lasers leads to nonlinear transformations and high-speed processing. Experimentally, a single semiconductor laser shows essential features for versatile computation, such as high-dimensional and nonlinear responses within sub-nanoseconds. To boost computational power, we study numerically the training of delay-coupled laser networks. The objective is, akin to training artificial neural networks, optimizing laser network's to improve performance and computational efficiency in challenging machine learning tasks. However, relying on offline optimization methods and physical models raises challenges due to device variability and limited system observability. Here, we propose evolutionary strategies to optimize physical systems without needing precise model knowledge, offering a promising approach for online system optimization.
Convolutional Neural Networks (CNNs) are employed in a plethora of fields, including computer vision, natural language processing, and speech recognition. We present an integrated photonic accelerator for CNNs based on the temporal-spatial interleaving of signals. This architecture supports 1D kernels, and can be extended to 2D convolutional kernels, providing scalability for complex networks. A supervised on-chip learning algorithm is employed to guarantee a reliable setting of convolutional weights against fabrication tolerances, thermal cross-talks, and changes in operating conditions. Overall, by leveraging photonics technology, the proposed accelerator significantly reduces hardware complexity while enabling high-speed processing and parallelism.
Proposed for analog computing, multimode fibers have limitations due to slow spatial-domain encoding. Our work showcases instead the computational prowess of a scheme employing a step-index few-mode fiber (FMF) segment, for high-speed spatiotemporal coincidence detection by leveraging the FMF’s dispersive optical properties. The FMF is a custom-made fabrication, with NA = 0.15, a core diameter of 22 μm, and a length of 13 m, introducing delay to temporal input pulses through the supported propagation of higher-order fiber modes. The temporal mixing of these modes creates short-term memory for time-encoded information which we exploit for coincidence detection. By slightly misaligning the input beam with the FMF’s longitudinal axis, we can modify the impact of the different modes on the overall spatial pattern distribution. Our experimental system operates at 1550 nm and encodes 6-bit header patterns with 35.1 ps pulses per bit. With four distinct 40 GHz photodetected points at the output speckle pattern of the FMF, we capture four different time series that correspond to different power integrals and use them to train a logistic regression classifier. Eventually, every header classification is performed with the sampling of only one pulse time window, thus our system operates at 28.5 Gb/s. Remarkably, under various input misalignment conditions, our system demonstrates error rates below 1/5000. This level of performance could not be obtained with a standard step-index multimode fiber of the appropriate length.
KEYWORDS: Tunable filters, Optical filters, Neurons, Frequency combs, Multiplexing, Integrated optics, Linear filtering, Photonics, Signal processing, Signal attenuation
Reservoir Computers (RCs) are brain-inspired algorithms based on recurrent neural networks where only output weights are tuned, while internal weights remain untrained. We recently demonstrated a photonic frequency-multiplexing RC encoding neurons in the lines of a frequency comb. We also demonstrated a single-layer feed-forward neural network based on a similar frequency-multiplexing principle. Here we present the design for an integrated optical output layer for such frequency multiplexing based photonic neural networks. The all-optical output layer uses wavelength (de)multiplexers and wavelength converters to apply signed weights to neurons encoded in comb lines.
We show the computational power of few-mode fibers (FMF) in a 40 Gbps spatiotemporal coincidence detector scheme. We consider a 5.5 m step-index FMF, with a 16.6 μm core diameter, as the medium that introduces various delays to a temporal input pulse, via the supported propagation fiber modes. In our representation, the different group velocities of the excited fiber modes define equivalent optical dendritic branches. A 1550 nm laser’s optical output is modulated by a 40 Gbps binary sequence and coupled to the FMF. The output optical pattern is photodetected by a 3×3 array and used to solve successfully a 6-bit header classification task.
KEYWORDS: Analog electronics, Optoelectronics, Electronic filtering, Photodiodes, Modulators, Nonlinear filtering, Data processing, Computing systems, Digital filtering, Signal generators
We present a systematic approach to the design of an analog implementation of photonic reservoir computing. The scheme builds on the idea that, thanks to time-multiplexing, a single nonlinear node subject to a delayed feedback loop can emulate a network with ring-like topology. We go beyond previous approaches for analog photonic reservoir computers by considering a ow model (continuous time) of the corresponding optoelectronic implementation, instead of the usually considered map limit, as the continuous time approach allows for operating at faster modulations. We focus on the implementation of an analog output layer made of a modulator and a second order filter that makes any digital post-processing unnecessary. Numerical simulations of the system show that the suggested analog design of the analog output layer is robust towards potential experimental deviations such as time jitter. The combination of the optoelectronic implementation with an analog output layer allows for high-speed information processing.
Quantum reservoir computing is an unconventional computing approach that exploits the quantumness of physical systems used as reservoirs to process information, combined with an easy training strategy. An overview is presented about a range of possibilities including quantum inputs, quantum physical substrates and quantum tasks. Recently, the framework of quantum reservoir computing has been proposed using Gaussian quantum states that can be realized e.g. in linear quantum optical systems. The universality and versatility of the system makes it particularly interesting for optical implementations. In particular, full potential of the proposed model can be reached even by encoding into quantum fluctuations, such as squeezed vacuum, instead of classical intense fields or thermal fluctuations. Some examples of the performance of this linear quantum reservoir in temporal tasks are reported.
Forecasting the dynamics of chaotic systems from the analysis of their output signals is a challenging problem with applications in most fields of modern science. In this work, we use a laser model to compare the performance of several machine learning algorithms for forecasting the amplitude of upcoming emitted chaotic pulses. We simulate the dynamics of an optically injected semiconductor laser that presents a rich variety of dynamical regimes when changing the parameters. We focus on a particular regime where the intensity shows a chaotic pulsing dynamics, and occasionally an ultra-high pulse, reminiscent of a rogue wave, is emitted. Our goal is to predict the amplitude (height) of the next pulse, knowing the amplitude of the three preceding pulses. We compare the performance of several machine learning methods, namely neural networks, support vector machine, nearest neighbors and reservoir computing. We analyze how their performance depends on the length of the time-series used for training.
Semiconductor lasers subject to external feedback are known to exhibit a wide variety of dynamical regimes desired for some applications such as chaos cryptography, random bit generation, and reservoir computing. Low-frequency fluctuations is one of the most frequently encountered regimes. It is characterized by a fast drop in laser intensity followed by a gradual recovery. The duration of this recovery process is irregular and of the order of hundred nanoseconds. The average time between dropouts is much larger than the laser system characteristic time-scales. Semiconductor ring lasers are currently the focus of a rapidly thriving research activity due to their unique feature of directional bistability. They can be employed in systems for all-optical switching, gating, wavelength-conversion functions, and all-optical memories. Semiconductor ring lasers do not require cleaved facets or gratings for optical feedback and are thus particularly suited for monolithic integration. We experimentally and numerically address the issue of low-frequency fluctuations considering a semiconductor ring laser in a feedback configuration where only one directional mode is re-injected into the same directional mode, a so-called single self-feedback. We have observed that the system is very sensitive to the feedback strength and the injection current. In particular, the power dropouts are more regular when the pump current is increased and become less frequent when the feedback strength is increased. In addition, we find two different recovery processes after the power dropouts of the low-frequency fluctuations. The recovery can either occur via pulses or in a stepwise manner. Since low-frequency fluctuations are not specific to semiconductor ring lasers, we expect these recovery processes to appear also in VCSELs and edge-emitting lasers under similar feedback conditions. The numerical simulations also capture these different behaviors, where the representation in the phase space of the carriers versus the round trip phase difference gives additional insight into these phenomena. This proceedings paper gives a short overview of the results of L. Mashal et al. [L. Mashal et al., IEEE J. Quantum. Electron. 49, 790, 2013].
Reservoir computing has recently been introduced as a new paradigm in the eld of machine learning. It is
based on the dynamical properties of a network of randomly connected nodes or neurons and shows to be very
promising to solve complex classication problems in a computationally ecient way. The key idea is that an
input generates nonlinearly transient behavior rendering transient reservoir states suitable for linear classication.
Our goal is to study up to which extent systems with delay, and especially photonic systems, can be used as
reservoirs.
Recently an new architecture has been proposed1 , based on a single nonlinear node with delayed feedback.
An electronic1 and an opto-electronic implementation2, 3 have been demonstrated and both have proven to be
very successful in terms of performance. This simple conguration, which replaces an entire network of randomly
connected nonlinear nodes with one single hardware node and a delay line, is signicantly easier to implement
experimentally. It is no longer necessary to construct an entire network of hundreds or even thousands of circuits,
each one representing a node. With this approach one node and a delay line suce to construct a computational
unit.
In this manuscript, we present a further investigation of the properties of delayed feedback congurations
used as a reservoir. Instead of quantifying the performance as an error obtained for a certain benchmark, we
now investigate a task-independent property, the linear memory of the system.
The time evolution of the output of a semiconductor laser subject to optical feedback can exhibit high-dimensional
chaotic fluctuations. In this contribution, our aim is to quantify the complexity of the chaotic time-trace generated
by a semiconductor laser subject to delayed optical feedback. To that end, we discuss the properties of two
recently introduced complexity measures based on information theory, namely the permutation entropy (PE)
and the statistical complexity measure (SCM). The PE and SCM are defined as a functional of a symbolic
probability distribution, evaluated using the Bandt-Pompe recipe to assign a probability distribution function to
the time series generated by the chaotic system. In order to evaluate the performance of these novel complexity
quantifiers, we compare them to a more standard chaos quantifier, namely the Kolmogorov-Sinai entropy. Here,
we present numerical results showing that the statistical complexity and the permutation entropy, evaluated at
the different time-scales involved in the chaotic regime of the laser subject to optical feedback, give valuable
information about the complexity of the laser dynamics.
Vertical Cavity Surface Emitting Lasers (VCSELs) often present switching between two orthogonal polarization states when varying parameters like e.g. current or temperature. Around such a switching point, the system randomly jumps between these two polarization states (mode hopping), driven by noise. In this contribution, we present experimental and numerical results showing the effect of coloured noise, externally added to the current, on the switching characteristics of a VCSEL.
The distribution and playback of digital images and other multimedia products are easily and fast done. Thus, its processing in order to achieve satisfactory copyright protection is a challenging problem for the research community. Encrypting the data only offers protection as long as the data remains encrypted, since once an authorized but fraudulent user decrypts it, nothing stops him from redistributing the data without having to worry about being caught. A watermarking scheme, which embeds some owner information (mark) into host images, is regarded as a possible solution to this problem. Nevertheless, digital watermarking is not strong enough to offer protection against illegal distributors. In this environment, digital fingerprinting techniques provide a good solution to dissuade illegal copying. To make such distribution systems work securely, the embedded marks in those system must be resistant to powerful attacks such as common image processing operations, lossy image compression, geometric transforms, combination addition of random noise (errors) and/or collusion attacks.
The work presented in this paper consists on the development of an empirical and portable JAVA platform where digital video (in MPEG2 format) can be protected against redistribution by dishonest users. The platform allows to verify at a practical level the strength properties of digital watermarking and fingerprinting marks. More precisely, it can be used to compare the performance of different watermarking algorithms (spread-spectrum and QIM). Moreover, it also offers the capability of embedding different digital fingerprinting codes, and verify its behaviour.
Low Frequency Fluctuations (LFF) are defined by an abrupt (1 ns)
drop-out of the emitted power followed by a gradual (50 ns)
build-up of the power until the next drop-out event, when the
laser with feedback is biased close to threshold. In this paper
experimental and theoretical results on a vertical-cavity
surface-emitting laser (VCSEL) with polarized optical feedback are
presented. Experimentally, we observe single-mode low frequency
dynamics when the VCSEL is biased below the solitary laser
threshold. We can choose one of the two typical polarization modes
(PM) of the VCSEL to be lasing, by an adequate choice of the
polarization direction in the external cavity. Our theoretical
analysis is based on a model developed by Loiko et al. which is an
extension of the Spin-Flip model. We confirm the appearance of
single-mode LFF and also reproduce the response of the orthogonal
polarization mode above the solitary laser threshold, both
deterministically and in presence of noise. This analysis shows
that aiming the feedback at the passive mode (in absence of
feedback) forces the active mode to react with short pulses, due
to parasitic carrier theft, while targeting the feedback at the
active mode induces a smaller response from the orthogonal
polarization mode. This difference in response allows us to
conclude that the secondary polarization does not play an
essential role in the LFF dynamics.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.