Search NASA⌕ Search

SEARCH · Search NASA

Results for “nonlinear mappings”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 253 records · Page 14
All sourcesNASA NTRSDOE OSTI
NASA NTRS2023

Investigating the Response of Land-Atmosphere Interactions and Feedbacks to Spatial Representation of Irrigation in a Coupled Modeling Framework

The transport of water, heat, and momentum from the surface to the atmosphere is dependent in part on the 10 characteristics of the land surface. Together with the model physics, parameterization schemes, and parameters employed, land datasets determine the spatial variability in land surface states (i.e., soil moisture and temperature) and fluxes. Despite the importance of these datasets, they are often chosen out of convenience or regional limitations without due assessment of their impacts on model results. Irrigation is an anthropogenic form of land heterogeneity that has been shown to alter the land surface energy balance, ambient weather, and local circulations. As such, irrigation schemes are becoming more 15 prevalent in weather and climate models with rapid developments in dataset availability and parameterization scheme complexity. Thus, to address pragmatic issues related to modeling irrigation, this study uses a high-resolution, regional coupled modeling system to investigate the impacts of irrigation dataset selection on land-atmosphere (L-A) coupling using a case study from the Great Plains Irrigation Experiment (GRAINEX) field campaign. The simulations are assessed in the context of irrigated versus non-irrigated regions, subregions across the irrigation gradient, and sub-grid scale process 20 representation in coarser scale models. The results show that L-A coupling is sensitive to the choice of irrigation dataset and resolution and that the irrigation impact on surface fluxes and near surface meteorology can be dominant, conditioned on the details of the irrigation map (i.e., boundaries, heterogeneity, etc), or minimal. A consistent finding across several analyses was that even a low percentage of irrigation fraction (i.e., 4-16%) can have significant local and downstream atmospheric impacts (e.g., lower PBL height), suggesting that representation of boundaries and heterogeneous areas within irrigated 25 regions is particularly important for the modeling of irrigation impacts on the atmosphere in this model. When viewing the simulations presented here as a proxy for ‘ideal’ tiling in a Earth System Model scale gridbox, the results show that some ‘tiles’ will reach critical nonlinear moisture and planetary boundary layer (PBL) thresholds that could be important for clouds and convection, implying that heterogeneity resulting from irrigation should be taken into consideration in new sub-grid land-atmosphere exchange parameterizations.

Patricia Lawston-Parker↗
NASA NTRS1987

Abundance and distribution of ultramafic microbreccia in Moses Rock dike - Quantitative application of mapping spectroscopy

Data from the Airborne Imaging Spectrometer were used to map the distribution and abundance of the serpentized ultramafic microbreccia (SUM) component in the Moses Rock dike, which is a Tertiary diatreme located on the Colorado Plateau in Utah. The geologic setting and composition of Moses Rock dike are discussed together with its texture and the relationship to the bedrock of surface materials. These observations along with laboratory spectroscopic data are used to interpret surface mineralogy of the dike and the surrounding regions from the imaging spectometer data. The spatial distribution and the abundance of the primary surface components were calculated using a nonlinear model for the mixing of spectra from multicomponent surfaces. The derived SUM distribution and abundance data support McGetchin's (1968) model for the emplacement of Moses Rock dike as a fluidized system.

Mustard, John F.↗
NASA NTRS1984

Deformation measurements of composite multi-span beam shear specimens by Moire interferometry

Experimental analyses were performed for determination of in plane deformations and shear strains in unidirectional and quasi-isotropic graphite-epoxy beams. Forty-eight ply beams were subjected to 5 point and 3 point flexure. Whole field measurements were recorded at load levels from about 20% to more than 90% of failure loads. Contour maps of U and W displacement fields were obtained by moire interferometry, using reference gratings of 2400 lines/mm. Clearly defined fringes with fringe orders exceeding 1000 were obtained. Whole field contour maps of shear strains were obtained by a method developed for these tests. Various anomalous effects were detected in the displacement fields. Their analysis indicated excess shear strains in resin rich zones in regions of shear tractions; free edge shear strains in quasi-isotropic specimens in regions of normal stresses; and shear stresses associated with cyclic shear compliances of quasi-isotropic plies in regions of shear tractions. Their contributions could occur independently or in superposition. Qualitative analyses addressed questions of relaxation; influence of contact stress distribution; specimen failure; effect of specimen overhang; nonlinearity; and qualities of 5 and 3 point flexure tests.

Post, D.↗
NASA NTRS1997

Detection of chaotic determinism in time series from randomly forced maps

Time series from biological system often display fluctuations in the measured variables. Much effort has been directed at determining whether this variability reflects deterministic chaos, or whether it is merely "noise". Despite this effort, it has been difficult to establish the presence of chaos in time series from biological sytems. The output from a biological system is probably the result of both its internal dynamics, and the input to the system from the surroundings. This implies that the system should be viewed as a mixed system with both stochastic and deterministic components. We present a method that appears to be useful in deciding whether determinism is present in a time series, and if this determinism has chaotic attributes, i.e., a positive characteristic exponent that leads to sensitivity to initial conditions. The method relies on fitting a nonlinear autoregressive model to the time series followed by an estimation of the characteristic exponents of the model over the observed probability distribution of states for the system. The method is tested by computer simulations, and applied to heart rate variability data.

Non-NASA Center↗
NASA NTRS1973

Automated thematic mapping and change detection of ERTS-A images

The author has identified the following significant results. For the recognition of terrain types, spatial signatures are developed from the diffraction patterns of small areas of ERTS-1 images. This knowledge is exploited for the measurements of a small number of meaningful spatial features from the digital Fourier transforms of ERTS-1 image cells containing 32 x 32 picture elements. Using these spatial features and a heuristic algorithm, the terrain types in the vicinity of Phoenix, Arizona were recognized by the computer with a high accuracy. Then, the spatial features were combined with spectral features and using the maximum likelihood criterion the recognition accuracy of terrain types increased substantially. It was determined that the recognition accuracy with the maximum likelihood criterion depends on the statistics of the feature vectors. Nonlinear transformations of the feature vectors are required so that the terrain class statistics become approximately Gaussian. It was also determined that for a given geographic area the statistics of the classes remain invariable for a period of a month but vary substantially between seasons.

Gramenopoulos, N.↗
NASA NTRS2002

Induced Moment Effects of Formation Flight Using Two F/A-18 Aircraft

Previous investigations into formation flight have shown the possibility for significant fuel savings through drag reduction. Using two F/A-18 aircraft, NASA Dryden Flight Research Center has investigated flying aircraft in autonomous formation. Positioning the trailing airplane for best drag reduction requires investigation of the wingtip vortex effects induced by the leading airplane. A full accounting of the vortex effect on the trailing airplane is desired to validate vortex-effect prediction methods and provide a database for the design of a formation flight autopilot. A recent flight phase has mapped the complete wingtip vortex effects at two flight conditions with the trailing airplane at varying distances behind the leading one. Force and moment data at Mach 0.56 and an altitude of 25,000 ft and Mach 0.86 and an altitude of 36,000 ft have been obtained with 20, 55, 110, and 190 ft of longitudinal distance between the aircraft. The moments induced by the vortex on the trailing airplane were well within the pilot's ability to control. This report discusses the data analysis methods and vortex-induced effects on moments and side force. An assessment of the impact of the nonlinear vortex effects on the design of a formation autopilot is offered.

Hansen, Jennifer L.↗
NASA NTRS1985

A new model of resonance in the winter stratosphere

It is generally accepted that the planetary waves observed in the winter stratosphere are primarily a response to dynamical forcing from the troposphere. Nevertheless, the mechanism by which wave amplitudes sometimes become larger remains uncertain. It is possible that anomalously large waves in the stratosphere might simply be the result of anomalously large tropospheric forcing. However, it has also been suggested that they are a response to a stratospheric-tropospheric cavity being in a near-resonant configuration. It has been suggested that nonlinear self-tuning effects could play an important role in the behavior of such a cavity. Self-tuning may occur when a system starts to one side of resonance, such that the mean-state change induced by growing waves brings the system closer to resonance. A new model of the stratospheric cavity is introduced and is then used to re-examine the possibility of wave growth in the real atmosphere and in atmospheric models due to self-tuning effects. The new model is based on the picture of the winter-time stratosphere which has been revealed by the observations of Ertel's potential vorticity, Q. Isentropic maps of Q show two rather distinct regions, the first containing the circumpolar vortex, where gradients of Q are large and Rossby waves may propagate easily. Surrounding this is a second, low-latitude region where the gradients are generally weak and where, because the Eulerian-mean flow is comparable with their phase speed, Rossby waves must be continually breaking. As the waves are observed to grow the relative sizes of these two regions change in time. This leads to the interesting possibility that self-tuning, mainly due to irreversible changes in the size of the polar vortex, is taking place.

Haynes, P. H.↗
NASA NTRS1997

A Comparison of PETSC Library and HPF Implementations of an Archetypal PDE Computation

Two paradigms for distributed-memory parallel computation that free the application programmer from the details of message passing are compared for an archetypal structured scientific computation a nonlinear, structured-grid partial differential equation boundary value problem using the same algorithm on the same hardware. Both paradigms, parallel libraries represented by Argonne's PETSC, and parallel languages represented by the Portland Group's HPF, are found to be easy to use for this problem class, and both are reasonably effective in exploiting concurrency after a short learning curve. The level of involvement required by the application programmer under either paradigm includes specification of the data partitioning (corresponding to a geometrically simple decomposition of the domain of the PDE). Programming in SPAM style for the PETSC library requires writing the routines that discretize the PDE and its Jacobian, managing subdomain-to-processor mappings (affine global- to-local index mappings), and interfacing to library solver routines. Programming for HPF requires a complete sequential implementation of the same algorithm, introducing concurrency through subdomain blocking (an effort similar to the index mapping), and modest experimentation with rewriting loops to elucidate to the compiler the latent concurrency. Correctness and scalability are cross-validated on up to 32 nodes of an IBM SP2.

Hayder, M. Ehtesham↗
NASA NTRS2018

Advances in Hyperspectral Image Classification Methods for Vegetation and Agricultural Cropland Studies

Hyperspectral data are becoming more widely available via sensors on airborne and unmanned aerial vehicle (UAV) platforms, as well as proximal platforms. While space-based hyperspectral data continue to be limited in availability, multiple spaceborne Earth-observing missions on traditional platforms are scheduled for launch, and companies are experimenting with small satellites for constellations to observe the Earth, as well as for planetary missions. Land cover mapping via classification is one of the most important applications of hyperspectral remote sensing and will increase in significance as time series of imagery are more readily available. However, while the narrow bands of hyperspectral data provide new opportunities for chemistry-based modeling and mapping, challenges remain. Hyperspectral data are high dimensional, and many bands are highly correlated or irrelevant for a given classification problem. For supervised classification methods, the quantity of training data is typically limited relative to the dimension of the input space. The resulting Hughes phenomenon, often referred to as the curse of dimensionality, increases potential for unstable parameter estimates, overfitting, and poor generalization of classifiers. This is particularly problematic for parametric approaches such as Gaussian maximum likelihood–based classifiers that have been the backbone of pixel-based multispectral classification methods. This issue has motivated investigation of alternatives, including regularization of the class covariance matrices, ensembles of weak classifiers, development of feature selection and extraction methods, adoption of nonparametric classifiers, and exploration of methods to exploit unlabeled samples via semi-supervised and active learning. Data sets are also quite large, motivating computationally efficient algorithms and implementations. This chapter provides an overview of the recent advances in classification methods for mapping vegetation using hyperspectral data. Three data sets that are used in the hyperspectral classification literature (e.g., Botswana Hyperion satellite data and AVIRIS airborne data over both Kennedy Space Center and Indian Pines) are described in Section 3.2 and used to illustrate methods described in the chapter. An additional high-resolution hyperspectral data set acquired by a SpecTIR sensor on an airborne platform over the Indian Pines area is included to exemplify the use of new deep learning approaches, and a multiplatform example of airborne hyperspectral data is provided to demonstrate transfer learning in hyperspectral image classification. Classical approaches for supervised and unsupervised feature selection and extraction are reviewed in Section 3.3. In particular, nonlinearities exhibited in hyperspectral imagery have motivated development of nonlinear feature extraction methods in manifold learning, which are outlined in Section 3.3.1.4. Spatial context is also important in classification of both natural vegetation with complex textural patterns and large agricultural fields with significant local variability within fields. Approaches to exploit spatial features at both the pixel level (e.g., co-occurrence–based texture and extended morphological attribute profiles [EMAPs]) and integration of segmentation approaches (e.g., HSeg) are discussed in this context in Section 3.3.2. Recently, classification methods that leverage nonparametric methods originating in the machine learning community have grown in popularity. An overview of both widely used and newly emerging approaches, including support vector machines (SVMs), Gaussian mixture models, and deep learning based on convolutional neural networks is provided in Section 3.4. Strategies to exploit unlabeled samples, including active learning and metric learning, which combine feature extraction and augmentation of the pool of training samples in an active learning framework, are outlined in Section 3.5. Integration of image segmentation with classification to accommodate spatial coherence typically observed in vegetation is also explored, including as an integrated active learning system. Exploitation of multisensor strategies for augmenting the pool of training samples is investigated via a transfer learning framework in Section 3.5.1.2. Finally, we look to the future, considering opportunities soon to be provided by new paradigms, as hyperspectral sensing is becoming common at multiple scales from ground-based and airborne autonomous vehicles to manned aircraft and space-based platforms.

Pasolli, Edoardo↗
NASA NTRS2005

Coherent Anti-Stokes Raman Spectroscopy (CARS) Measurements in Supersonic Combustors at NASA Langley Research Center

This paper describes the recent use of coherent anti-Stokes Raman spectroscopy (CARS) to study supersonic combustion at NASA Langley Research Center. CARS is a nonlinear optical measurement technique used to measure temperature and species mole fractions remotely in harsh environments. A CARS system has been applied to two different combustor geometries at NASA Langley. Both experiments used the same vitiated wind-tunnel facility to create an air flow that simulates flight at Mach numbers of 6 and 7 for the combustor inlet and both experiments used hydrogen fuel. In the first experiment, the hydrogen was injected supersonically at a 30-degree angle with respect to the incoming flow. In the second experiment, the hydrogen was injected sonically at normal incidence. While these injection schemes produced significantly different flow features, the CARS method provided mean temperature, N2, O2 and H2 maps at multiple downstream locations for both. The primary aim of these measurements was to provide detailed flowfield information for computational fluid dynamics (CFD) code validation.

Danehy, Paul M.↗
NASA NTRS1999

A Compact, Tunable Near-UV Source for Quantitative Microgravity Combustion Diagnostics

There is a need for improved optical diagnostic methods for use in microgravity combustion research. Spectroscopic methods with fast time response that can provide absolute concentrations and concentration profiles of important chemical species in flames are needed to facilitate the understanding of combustion kinetics in microgravity. Although a variety of sophisticated laser-based diagnostics (such as planar laser induced fluorescence, degenerate four wave mixing and coherent Raman methods) have been applied to the study of combustion in laboratory flames, the instrumentation associated with these methods is not well suited to microgravity drop tower or space station platforms. Important attributes of diagnostic systems for such applications include compact size, low power consumption, ruggedness, and reliability. We describe a diode laser-based near-UV source designed with the constraints of microgravity research in mind. Coherent light near 420 nm is generated by frequency doubling in a nonlinear crystal. This light source is single mode with a very narrow bandwidth suitable for gas phase diagnostics, can be tuned over several 1/cm and can be wavelength modulated at up to MHz frequencies. We demonstrate the usefulness of this source for combustion diagnostics by measuring CH radical concentration profiles in an atmospheric pressure laboratory flame. The radical concentrations are measured using wavelength modulation spectroscopy (WMS) to obtain the line-of-sight integrated absorption for different paths through the flame. Laser induced fluorescence (LIF) measurements are also demonstrated with this instrument, showing the feasibility of simultaneous WMS absorption and LIF measurements with the same light source. LIF detection perpendicular to the laser beam can be used to map relative species densities along the line-of-sight while the integrated absorption available through WMS provides a mathematical constraint on the extraction of quantitative information from the LIF data. Combining absorption with LIF - especially if the measurements are made simultaneously with the same excitation beam - may allow elimination of geometrical factors and effects of intensity fluctuations (common difficulties with the analysis of LIF data) from the analysis.

Peterson, K. A.↗
NASA NTRS2003

Stochastic Optimal Control via Bellman's Principle

This paper presents a method for finding optimal controls of nonlinear systems subject to random excitations. The method is capable to generate global control solutions when state and control constraints are present. The solution is global in the sense that controls for all initial conditions in a region of the state space are obtained. The approach is based on Bellman's Principle of optimality, the Gaussian closure and the Short-time Gaussian approximation. Examples include a system with a state-dependent diffusion term, a system in which the infinite hierarchy of moment equations cannot be analytically closed, and an impact system with a elastic boundary. The uncontrolled and controlled dynamics are studied by creating a Markov chain with a control dependent transition probability matrix via the Generalized Cell Mapping method. In this fashion, both the transient and stationary controlled responses are evaluated. The results show excellent control performances.

Crespo, Luis G.↗
NASA NTRS1993

On the dynamics of approximating schemes for dissipative nonlinear equations

Since one can rarely write down the analytical solutions to nonlinear dissipative partial differential equations (PDE's), it is important to understand whether, and in what sense, the behavior of approximating schemes to these equations reflects the true dynamics of the original equations. Further, because standard error estimates between approximations of the true solutions coming from spectral methods - finite difference or finite element schemes, for example - and the exact solutions grow exponentially in time, this analysis provides little value in understanding the infinite time behavior of a given approximating scheme. The notion of the global attractor has been useful in quantifying the infinite time behavior of dissipative PDEs, such as the Navier-Stokes equations. Loosely speaking, the global attractor is all that remains of a sufficiently large bounded set in phase space mapped infinitely forward in time under the evolution of the PDE. Though the attractor has been shown to have some nice properties - it is compact, connected, and finite dimensional, for example - it is in general quite complicated. Nevertheless, the global attractor gives a way to understand how the infinite time behavior of approximating schemes such as the ones coming from a finite difference, finite element, or spectral method relates to that of the original PDE. Indeed, one can often show that such approximations also have a global attractor. We therefore only need to understand how the structure of the attractor for the PDE behaves under approximation. This is by no means a trivial task. Several interesting results have been obtained in this direction. However, we will not go into the details. We mention here that approximations generally lose information about the system no matter how accurate they are. There are examples that show certain parts of the attractor may be lost by arbitrary small perturbations of the original equations.

Jones, Donald A.↗
NASA NTRS1996

Guidance of Nonlinear Nonminimum-Phase Dynamic Systems

The research work has advanced the inversion-based guidance theory for: systems with non-hyperbolic internal dynamics; systems with parameter jumps; and systems where a redesign of the output trajectory is desired. A technique to achieve output tracking for nonminimum phase linear systems with non-hyperbolic and near non-hyperbolic internal dynamics was developed. This approach integrated stable inversion techniques, that achieve exact-tracking, with approximation techniques, that modify the internal dynamics to achieve desirable performance. Such modification of the internal dynamics was used (a) to remove non-hyperbolicity which is an obstruction to applying stable inversion techniques and (b) to reduce large preactuation times needed to apply stable inversion for near non-hyperbolic cases. The method was applied to an example helicopter hover control problem with near non-hyperbolic internal dynamics for illustrating the trade-off between exact tracking and reduction of preactuation time. Future work will extend these results to guidance of nonlinear non-hyperbolic systems. The exact output tracking problem for systems with parameter jumps was considered. Necessary and sufficient conditions were derived for the elimination of switching-introduced output transient. While previous works had studied this problem by developing a regulator that maintains exact tracking through parameter jumps (switches), such techniques are, however, only applicable to minimum-phase systems. In contrast, our approach is also applicable to nonminimum-phase systems and leads to bounded but possibly non-causal solutions. In addition, for the case when the reference trajectories are generated by an exosystem, we developed an exact-tracking controller which could be written in a feedback form. As in standard regulator theory, we also obtained a linear map from the states of the exosystem to the desired system state, which was defined via a matrix differential equation.

Devasia, Santosh↗
NASA NTRS1988

Compression creep of filamentary composites

Axial and transverse strain fields induced in composite laminates subjected to compressive creep loading were compared for several types of laminate layups. Unidirectional graphite/epoxy as well as multi-directional graphite/epoxy and graphite/PEEK layups were studied. Specimens with and without holes were tested. The specimens were subjected to compressive creep loading for a 10-hour period. In-plane displacements were measured using moire interferometry. A computer based data reduction scheme was developed which reduces the whole-field displacement fields obtained using moire to whole-field strain contour maps. Only slight viscoelastic response was observed in matrix-dominated laminates, except for one test in which catastrophic specimen failure occurred after a 16-hour period. In this case the specimen response was a complex combination of both viscoelastic and fracture mechanisms. No viscoelastic effects were observed for fiber-dominated laminates over the 10-hour creep time used. The experimental results for specimens with holes were compared with results obtained using a finite-element analysis. The comparison between experiment and theory was generally good. Overall strain distributions were very well predicted. The finite element analysis typically predicted slightly higher strain values at the edge of the hole, and slightly lower strain values at positions removed from the hole, than were observed experimentally. It is hypothesized that these discrepancies are due to nonlinear material behavior at the hole edge, which were not accounted for during the finite-element analysis.

Graesser, D. L.↗
NASA NTRS1991

On the theory of the synthesis of offset dual-shaped reflectors - Case examples

In an earlier paper by Galindo-Israel et al. (1987), the geometrical optics (GO) principles, constraints, and requirements of the dual- and single-offset-shaped reflector synthesis problem were collected and developed into a set of nonlinear first-order PDEs. An extension of the methods by which solutions to the PDEs can be obtained is presented, together with several case examples. These examples are independently analyzed by GO and physical optics diffraction methods. The starting point for the integration over each reflector can be taken on the outer rim, at the center, or at an intermediate point-the intermediate starting point being the more general case. The utility of the speed of this synthesis method is demonstrated. For example, the program utilized completes a synthesis in less than 45 s on a 386/20 personal computer and within a fraction of a second on a mainframe Cray. This makes practical the incorporation of the synthesis into a search algorithm that can optimize one or more parameters of the reflector system. As an example, the optimization of the mapping equations for low cross polarization is discussed.

Galindo-Israel, Victor↗
NASA NTRS1990

Dual-shaped offset reflector antenna designs from solutions of the geometrical optics first-order partial differential equations

In obtaining solutions to the first-order nonlinear partial differential equations (PDEs) for synthesizing offset dual-shaped reflectors, it is found that previously observed computational problems can be avoided if the integration of the PDEs is started from an inner projected perimeter and integrated outward rather than starting from an outer projected perimeter and integrating inward. This procedure, however, introduces a new parameter, the main reflector inner perimeter radius p(o), when given a subreflector inner angle 0(o). Furthermore, a desired outer projected perimeter (e.g., a circle) is no longer guaranteed. Stability of the integration is maintained if some of the initial parameters are determined first from an approximate solution to the PDEs. A one-, two-, or three-parameter optimization algorithm can then be used to obtain a best set of parameters yielding a close fit to the desired projected outer rim. Good low cross-polarization mapping functions are also obtained. These methods are illustrated by synthesis of a high-gain offset-shaped Cassegrainian antenna and a low-noise offset-shaped Gregorian antenna.

Galindo-Israel, V.↗
NASA NTRS2007

Identification and Control of Aircrafts using Multiple Models and Adaptive Critics

We compared two possible implementations of local linear models for control: one approach is based on a self-organizing map (SOM) to cluster the dynamics followed by a set of linear models operating at each cluster. Therefore the gating function is hard (a single local model will represent the regional dynamics). This simplifies the controller design since there is a one to one mapping between controllers and local models. The second approach uses a soft gate using a probabilistic framework based on a Gaussian Mixture Model (also called a dynamic mixture of experts). In this approach several models may be active at a given time, we can expect a smaller number of models, but the controller design is more involved, with potentially better noise rejection characteristics. Our experiments showed that the SOM provides overall best performance in high SNRs, but the performance degrades faster than with the GMM for the same noise conditions. The SOM approach required about an order of magnitude more models than the GMM, so in terms of implementation cost, the GMM is preferable. The design of the SOM is straight forward, while the design of the GMM controllers, although still reasonable, is more involved and needs more care in the selection of the parameters. Either one of these locally linear approaches outperform global nonlinear controllers based on neural networks, such as the time delay neural network (TDNN). Therefore, in essence the local model approach warrants practical implementations. In order to call the attention of the control community for this design methodology we extended successfully the multiple model approach to PID controllers (still today the most widely used control scheme in the industry), and wrote a paper on this subject. The echo state network (ESN) is a recurrent neural network with the special characteristics that only the output parameters are trained. The recurrent connections are preset according to the problem domain and are fixed. In a nutshell, the states of the reservoir of recurrent processing elements implement a projection space, where the desired response is optimally projected. This architecture trades training efficiency by a large increase in the dimension of the recurrent layer. However, the power of the recurrent neural networks can be brought to bear on practical difficult problems. Our goal was to implement an adaptive critic architecture implementing Bellman s approach to optimal control. However, we could only characterize the ESN performance as a critic in value function evaluation, which is just one of the pieces of the overall adaptive critic controller. The results were very convincing, and the simplicity of the implementation was unparalleled.

Principe, Jose C.↗