A Spiking Neuromorphic Algorithm for Markov Reward Processes
Explore the source record for details and available documents.
Luca, Sarah Katherine↗
SEARCH · Search NASA
Results for “spiking algorithm”
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.
Building Scalable, Composable Spiking Neural Algorithms with Fugu
Explore the source record for details and available documents.
Musuvathy, Srideep↗
Spiking Neuromorphic Algorithms for Markov Decision Processes
Explore the source record for details and available documents.
Luca, Sarah Katherine [Sandia National Laboratorie↗
A Spiking Neuromorphic Algorithm for Markov Reward Processes
Explore the source record for details and available documents.
Luca, Sarah Katherine↗
Orion MPCV Touchdown Detection Threshold Development and Testing
A robust method of detecting Orion Multi ]Purpose Crew Vehicle (MPCV) splashdown is necessary to ensure crew and hardware safety during descent and after touchdown. The proposed method uses a triple redundant system to inhibit Reaction Control System (RCS) thruster firings, detach parachute risers from the vehicle, and transition to the post ]landing segment of the Flight Software (FSW). The vehicle crew is the prime input for touchdown detection, followed by an autonomous FSW algorithm, and finally a strictly time based backup timer. RCS thrusters must be inhibited before submersion in water to protect against possible damage due to firing these jets under water. In addition, neglecting to declare touchdown will not allow the vehicle to transition to post ]landing activities such as activating the Crew Module Up ]righting System (CMUS), resulting in possible loss of communication and difficult recovery. A previous AIAA paper gAssessment of an Automated Touchdown Detection Algorithm for the Orion Crew Module h concluded that a strictly Inertial Measurement Unit (IMU) based detection method using an acceleration spike algorithm had the highest safety margins and shortest detection times of other methods considered. That study utilized finite element simulations of vehicle splashdown, generated by LS ]DYNA, which were expanded to a larger set of results using a Kriging surface fit. The study also used the Decelerator Systems Simulation (DSS) to generate flight dynamics during vehicle descent under parachutes. Proto ]type IMU and FSW MATLAB models provided the basis for initial algorithm development and testing. This paper documents an in ]depth trade study, using the same dynamics data and MATLAB simulations as the earlier work, to further develop the acceleration detection method. By studying the combined effects of data rate, filtering on the rotational acceleration correction, data persistence limits and values of acceleration thresholds, an optimal configuration was determined. The lever arm calculation, which removes the centripetal acceleration caused by vehicle rotation, requires that the vehicle angular acceleration be derived from vehicle body rates, necessitating the addition of a 2nd order filter to smooth the data. It was determined that using 200 Hz data directly from the vehicle IMU outperforms the 40 Hz FSW data rate. Data persistence counter values and acceleration thresholds were balanced in order to meet desired safety and performance. The algorithm proved to exhibit ample safety margin against early detection while under parachutes, and adequate performance upon vehicle splashdown. Fall times from algorithm initiation were also studied, and a backup timer length was chosen to provide a large safety margin, yet still trigger detection before CMUS inflation. This timer serves as a backup to the primary acceleration detection method. Additionally, these parameters were tested for safety on actual flight test data, demonstrating expected safety margins.
Daum, Jared↗
The backpropagation algorithm implemented on spiking neuromorphic hardware
Not Available
97 MATHEMATICS AND COMPUTING↗
An analysis of the structure of gamma ray burst time histories
If gamma-ray bursts (GRB) arise from a small number of distinctly different physical phenomena, then this might be revealed by a clustering of time profile characteristics into a small number of groups. A 'spike' counting algorithm was applied to 107 GRB profiles. Graphs of spike frequency and spike amplitude versus burst intensity and duration are presented. So far, no evidence of grouping is seen.
Lestrade, John Patrick↗
Solving sparse finite element problems on neuromorphic hardware
The finite element method (FEM) is one of the most important and ubiquitous numerical methods for solving partial differential equations (PDEs) on computers for scientific and engineering discovery. Applying the FEM to larger and more detailed scientific models has driven advances in high-performance computing for decades. Here we demonstrate that scalable spiking neuromorphic hardware can directly implement the FEM by constructing a spiking neural network that solves the large, sparse, linear systems of equations at the core of the FEM. We show that for the Poisson equation, a fundamental PDE in science and engineering, our neural circuit achieves meaningful levels of numerical accuracy and close to ideal scaling on modern, inherently parallel and energy-efficient neuromorphic hardware, specifically Intel’s Loihi 2 neuromorphic platform. We illustrate extensions to irregular mesh geometries in both two and three dimensions as well as other PDEs such as linear elasticity. Our spiking neural network is constructed from a recurrent network model of the brain’s motor cortex and, in contrast to black-box deep artificial neural network-based methods for PDEs, directly translates the well-understood and trusted mathematics of the FEM to a natively spiking neuromorphic algorithm.
Applied mathematics↗
Motion Cueing Algorithm Development: Initial Investigation and Redesign of the Algorithms
In this project four motion cueing algorithms were initially investigated. The classical algorithm generated results with large distortion and delay and low magnitude. The NASA adaptive algorithm proved to be well tuned with satisfactory performance, while the UTIAS adaptive algorithm produced less desirable results. Modifications were made to the adaptive algorithms to reduce the magnitude of undesirable spikes. The optimal algorithm was found to have the potential for improved performance with further redesign. The center of simulator rotation was redefined. More terms were added to the cost function to enable more tuning flexibility. A new design approach using a Fortran/Matlab/Simulink setup was employed. A new semicircular canals model was incorporated in the algorithm. With these changes results show the optimal algorithm has some advantages over the NASA adaptive algorithm. Two general problems observed in the initial investigation required solutions. A nonlinear gain algorithm was developed that scales the aircraft inputs by a third-order polynomial, maximizing the motion cues while remaining within the operational limits of the motion system. A braking algorithm was developed to bring the simulator to a full stop at its motion limit and later release the brake to follow the cueing algorithm output.
Telban, Robert J.↗
When in-memory computing meets spiking neural networks—A perspective on device-circuit-system-and-algorithm co-design
This review explores the intersection of bio-plausible artificial intelligence in the form of spiking neural networks (SNNs) with the analog in-memory computing (IMC) domain, highlighting their collective potential for low-power edge computing environments. Through detailed investigation at the device, circuit, and system levels, we highlight the pivotal synergies between SNNs and IMC architectures. Additionally, we emphasize the critical need for comprehensive system-level analyses, considering the inter-dependencies among algorithms, devices, circuit, and system parameters, crucial for optimal performance. An in-depth analysis leads to the identification of key system-level bottlenecks arising from device limitations, which can be addressed using SNN-specific algorithm–hardware co-design techniques. This review underscores the imperative for holistic device to system design-space co-exploration, highlighting the critical aspects of hardware and algorithm research endeavors for low-power neuromorphic solutions.
Physics↗
Predicting Band-Gap of Inorganic Materials Using Neuromorphic Graph Learning
Predicting properties of inorganic materials is a heavily researched topic, with several new prediction approaches emerging as competitors. One such competitor is graph neural networks, which leverage the structure of the graph to aid in the prediction process. In this work, we propose integration of neuromorphic computation into the graph neural network pipeline. We call this approach Neuromorphic Graph Learning (NGL). We utilize the NGL approach to leverage evolutionary algorithms and a novel Spike Pipeline for Raster Analysis (SPIRE) for the prediction of band gap in inorganic materials.
Mulet, Ian [University of Tennessee (UT)]↗
Binary operations on neuromorphic hardware with application to linear algebraic operations and stochastic equations
Abstract Non-von Neumann computational hardware, based on neuron-inspired, non-linear elements connected via linear, weighted synapses—so-called neuromorphic systems—is a viable computational substrate. Since neuromorphic systems have been shown to use less power than CPUs for many applications, they are of potential use in autonomous systems such as robots, drones, and satellites, for which power resources are at a premium. The power used by neuromorphic systems is approximately proportional to the number of spiking events produced by neurons on-chip. However, typical information encoding on these chips is in the form of firing rates that unarily encode information. That is, the number of spikes generated by a neuron is meant to be proportional to an encoded value used in a computation or algorithm. Unary encoding is less efficient (produces more spikes) than binary encoding. For this reason, here we present neuromorphic computational mechanisms for implementing binary two’s complement operations. We use the mechanisms to construct a neuromorphic, binary matrix multiplication algorithm that may be used as a primitive for linear differential equation integration, deep networks, and other standard calculations. We also construct a random walk circuit and apply it in Brownian motion simulations. We study how both algorithms scale in circuit size and iteration time.
97 MATHEMATICS AND COMPUTING↗
A Hardware and Software Co-design Framework for Energy Efficient Neuromorphic Systems
Neuromorphic systems can be realized by a variety of algorithms and architectures. A common understanding is that spiking neuromorphic designs, which encode information into spatio-temporal spiking events, are both a biologically-accurate and efficient way of processing information. However, representing the information through timing relationships induces sophisticated circuit designs in traditional CMOS-based implementations. In recent years, high-capacity resistive memory (RRAM, aka, memristor) has demonstrated great potential in mimicking synaptic behaviors. Several RRAM-based spiking neuromorphic designs exist, most of which focus on rate coding schemes. These designs simplify circuit implementations of neuron models and explore challenges such as unsatisfactory speed, resolution, and performance. As an alternative, we will explore temporal coding spiking neuromorphic systems that encode information as the relative timing of neuron activations (spikes), which have been proven to be more adaptive and energy-efficient. Developing a neuromorphic system for spiking neural network (SNN) inference and online training, however, faces some major technical challenges: (1) It lacks circuit implementation support for temporal-coding SNN to achieve satisfying power efficiency and accuracy; (2) Although existing research works have investigated memristive synapse and neuron designs for spike-timing-dependent plasticity, the non-ideal conditions in implementation, such as device variations and signal degradation, degrade online learning accuracy of large scale systems; and (3) Non-optimized, inter-layer data traffic in SNNs, leads to unnecessary data communication costs. In this project, we plan to address these challenges by a hardware and software co-design framework that incorporates solutions at the circuit, architecture, and algorithm levels. At the circuit-level, we will elaborate on the in-situ SNN processing element designs for supporting both inference and online training modes. Variation-aware schemes will be studied to improve reliability. At the architecture level, we propose a pipelined, asynchronous architecture to retain the timing resolution of spikes. At the algorithm level, we will investigate an innovative SNN training algorithm for enabling activation sparsification and reducing unnecessary data communication costs. This neuromorphic system will provide an effective solution to real-life energy-constrained applications and significantly contribute to the exploration of next-generation high-performance computing systems under the DOE context.
97 MATHEMATICS AND COMPUTING↗
The CCD/Transit Instrument (CTI) data-analysis system
The automated software system for archiving, analyzing, and interrogating data from the CCD/Transit Instrument (CTI) is described. The CTI collects up to 450 Mbytes of image-data each clear night in the form of a narrow strip of sky observed in two colors. The large data-volumes and the scientific aims of the project make it imperative that the data are analyzed within the 24-hour period following the observations. To this end a fully automatic and self evaluating software system has been developed. The data are collected from the telescope in real-time and then transported to Tucson for analysis. Verification is performed by visual inspection of random subsets of the data and obvious cosmic rays are detected and removed before permanent archival is made to the optical disc. The analysis phase is performed by a pair of linked algorithms, one operating on the absolute pixel-values and the other on the spatial derivative of the data. In this way both isolated and merged images are reliably detected in a single pass. In order to isolate the latter algorithm from the effects of noise spikes a 3x3 Hanning filter is applied to the raw data before the analysis is run. The algorithms reduce the input pixel-data to a database of measured parameters for each image which has been found. A contrast filter is applied in order to assign a detection-probability to each image and then x-y calibration and intensity calibration are performed using known reference stars in the strip. These are added to as necessary by secondary standards boot-strapped from the CTI data itself. The final stages involve merging the new data into the CTI Master-list and History-list and the automatic comparison of each new detection with a set of pre-defined templates in parameter-space to find interesting objects such as supernovae, quasars and variable stars. Each stage of the processing from verification to interesting image selection is performed under a data-logging system which both controls the pipe-lining of data through the system and records key performance monitor parameters which are built into the software. Furthermore, the data from each stage are stored in databases to facilitate evaluation, and all stages offer the facility to enter keyword-indexed free-format text into the data-logging system. In this way a large measure of certification is built into the system to provide the necessary confidence in the end results.
Cawson, M. G. M.↗
Assessment of an Automated Touchdown Detection Algorithm for the Orion Crew Module
Orion Crew Module (CM) touchdown detection is critical to activating the post-landing sequence that safe?s the Reaction Control Jets (RCS), ensures that the vehicle remains upright, and establishes communication with recovery forces. In order to accommodate safe landing of an unmanned vehicle or incapacitated crew, an onboard automated detection system is required. An Orion-specific touchdown detection algorithm was developed and evaluated to differentiate landing events from in-flight events. The proposed method will be used to initiate post-landing cutting of the parachute riser lines, to prevent CM rollover, and to terminate RCS jet firing prior to submersion. The RCS jets continue to fire until touchdown to maintain proper CM orientation with respect to the flight path and to limit impact loads, but have potentially hazardous consequences if submerged while firing. The time available after impact to cut risers and initiate the CM Up-righting System (CMUS) is measured in minutes, whereas the time from touchdown to RCS jet submersion is a function of descent velocity, sea state conditions, and is often less than one second. Evaluation of the detection algorithms was performed for in-flight events (e.g. descent under chutes) using hi-fidelity rigid body analyses in the Decelerator Systems Simulation (DSS), whereas water impacts were simulated using a rigid finite element model of the Orion CM in LS-DYNA. Two touchdown detection algorithms were evaluated with various thresholds: Acceleration magnitude spike detection, and Accumulated velocity changed (over a given time window) spike detection. Data for both detection methods is acquired from an onboard Inertial Measurement Unit (IMU) sensor. The detection algorithms were tested with analytically generated in-flight and landing IMU data simulations. The acceleration spike detection proved to be faster while maintaining desired safety margin. Time to RCS jet submersion was predicted analytically across a series of simulated Orion landing conditions. This paper details the touchdown detection method chosen and the analysis used to support the decision.
Gay, Robert S.↗
Time-accurate simulation of a self-excited oscillatory supersonic external flow with a multi-block solution-adaptive mesh algorithm
Results are presented of an investigation of the time-accurate simulation of supersonic unsteady flow oscillations over spike-tipped bodies using the multistage Runge-Kutta scheme coupled with a dynamic solution-adaptive grid algorithm modified for multiblock capabilities. The inviscid fluxes are described by a modified advective upwind split method to obviate the need for artificial dissipation. If a time-varying, solution-adaptive mesh algorithm is incorporated, resolution of the details of the unsteady spike-tipped body flow is improved. The adaptive algorithm is also shown to resolve multiple and diverse features of the flow simultaneously, with the adapted regions in the mesh convecting with these features as they translate.
Ingram, Clint L.↗
MODIS 3km Aerosol Product: Algorithm and Global Perspective
After more than a decade of producing a nominal 10 km aerosol product based on the dark target method, the MODIS aerosol team will be releasing a nominal 3 km product as part of their Collection 6 release. The new product differs from the original 10 km product only in the manner in which reflectance pixels are ingested, organized and selected by the aerosol algorithm. Overall, the 3 km product closely mirrors the 10 km product. However, the finer resolution product is able to retrieve over ocean closer to islands and coastlines, and is better able to resolve fine aerosol features such as smoke plumes over both ocean and land. In some situations, it provides retrievals over entire regions that the 10 km product barely samples. In situations traditionally difficult for the dark target algorithm, such as over bright or urban surfaces the 3 km product introduces isolated spikes of artificially high aerosol optical depth (AOD) that the 10 km algorithm avoids. Over land, globally, the 3 km product appears to be 0.01 to 0.02 higher than the 10 km product, while over ocean, the 3 km algorithm is retrieving a proportionally greater number of very low aerosol loading situations. Based on collocations with ground-based observations for only six months, expected errors associated with the 3 km land product are determined to be greater than for the 10 km product: 0.05 0.25 AOD. Over ocean, the suggestion is for expected errors to be the same as the 10 km product: 0.03 0.05 AOD. The advantage of the product is on the local scale, which will require continued evaluation not addressed here. Nevertheless, the new 3 km product is expected to provide important information complementary to existing satellite-derived products and become an important tool for the aerosol community.
Remer, L. A.↗
Real-Time Neuromorphic Processing of Spatiotemporal Data for Scientific Discovery
Spiking Neural Networks (SNNs) are brain-inspired computing models incorporating unique temporal dynamics and event-driven processing. Rich dynamics in both space and time offer great challenges and opportunities for efficient processing of sparse spatiotemporal data compared with conventional artificial neural networks (ANNs). Under this context, the goal of this project is to develop spiking neural network based neuromorphic computing to enable energy-efficient real-time learning and processing of spatiotemporal data. This report summarizes the key results on network architecture design, training methods, and SNN hardware acceleration achieved under this project, demonstrating the promise of spiking neural networks.
97 MATHEMATICS AND COMPUTING↗