Search NASA⌕ Search

SEARCH · Search NASA

Results for “Computational optimization”

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 1,531 records · Page 85

The potential of quantum annealing for rapid solution structure identification

Abstract The recent emergence of novel computational devices, such as quantum computers, coherent Ising machines, and digital annealers presents new opportunities for hardware-accelerated hybrid optimization algorithms. Unfortunately, demonstrations of unquestionable performance gains leveraging novel hardware platforms have faced significant obstacles. One key challenge is understanding the algorithmic properties that distinguish such devices from established optimization approaches. Through the careful design of contrived optimization tasks, this work provides new insights into the computation properties of quantum annealing and suggests that this model has the potential to quickly identify the structure of high-quality solutions. A meticulous comparison to a variety of algorithms spanning both complete and local search suggests that quantum annealing’s performance on the proposed optimization tasks is distinct. This result provides new insights into the time scales and types of optimization problems where quantum annealing has the potential to provide notable performance gains over established optimization algorithms and suggests the development of hybrid algorithms that combine the best features of quantum annealing and state-of-the-art classical approaches.

97 MATHEMATICS AND COMPUTING↗

Algorithmic and GPU enhancements for molecular dynamics in Cabana and LAMMPS

The focus of this milestone was to implement and test new algorithms and optimizations for improved GPU performance by molecular dynamics (MD) and related particle codes. The work was done in the CoPA Cabana library, the CabanaMD mini-app which uses Cabana to implement and benchmark prototypical MD models, and the LAMMPS MD code which is used by the ECP EXAALT project as well as for materials modeling generally. Efforts included (a) implementation of new or optimized algorithms for fundamental MD kernels like neighbor-list building and FFTs which are also widely used in other particle methods, (b) performance testing of data structure layouts for both simple pairwise and computationally intense neural network potentials, (c) optimizations aimed at improving GPU performance for models with small atom counts, and (d) optimizations specific to the SNAP interatomic potential due to its importance for the EXAALT project.

36 MATERIALS SCIENCE↗

Adaptive Circuit Learning for Quantum Metrology

Quantum sensing is an important application of emerging quantum technologies. We explore whether a hybrid system of quantum sensors and quantum circuits can surpass the classical limit of sensing. In particular, we use optimization techniques to search for encoder and decoder circuits that scalably improve sensitivity under given application and noise characteristics. Furthermore, our approach uses a variational algorithm that can learn a quantum sensing circuit based on platform-specific control capacity, noise, and signal distribution. The quantum circuit is composed of an encoder which prepares the optimal sensing state and a decoder which gives an output distribution containing information of the signal. We optimize the full circuit to maximize the Signal-to-Noise Ratio (SNR). Furthermore, this learning algorithm can be run on real hardware scalably by using the "parameter-shift" rule which enables gradient evaluation on noisy quantum circuits, avoiding the exponential cost of quantum system simulation. We demonstrate up to 13.12x SNR improvement over existing fixed protocol (GHZ), and 3.19x Classical Fisher Information (CFI) improvement over the classical limit on 15 qubits using IBM quantum computer. More notably, our algorithm overcomes the decreasing performance of existing entanglement-based protocols with increased system sizes.

42 ENGINEERING↗

Using Differential Evolution to Optimize Learning from Signals and Enhance Network Security

Computer and communication network attacks are commonly orchestrated through Wireless Access Points (WAPs). This paper summarizes proof-of-concept research activity aimed at developing a physical layer Radio Frequency (RF) air monitoring capability to limit unauthorizedWAP access and mprove network security. This is done using Differential Evolution (DE) to optimize the performance of a Learning from Signals (LFS) classifier implemented with RF Distinct Native Attribute (RF-DNA) fingerprints. Performance of the resultant DE-optimized LFS classifier is demonstrated using 802.11a WiFi devices under the most challenging conditions of intra-manufacturer classification, i.e., using emissions of like-model devices that only differ in serial number. Using identical classifier input features, performance of the DE-optimized LFS classifier is assessed relative to a Multiple Discriminant Analysis / Maximum Likelihood (MDA/ML) classifier that has been used for previous demonstrations. The comparative assessment is made using both Time Domain (TD) and Spectral Domain (SD) fingerprint features. For all combinations of classifier type, feature type, and signal-to-noise ratio considered, results show that the DEoptimized LFS classifier with TD features is uperior and provides up to 20% improvement in classification accuracy with proper selection of DE parameters.

99 GENERAL AND MISCELLANEOUS↗

Data Locality Enhancement of Dynamic Simulations for Exascale Computing (Final Report)

The development of modern processors exhibits two trends that complicate the optimizations of modern software. The first is the increasing sensitivity of processors' throughput to irregularities in computation. With more processors produced through a massive integration of simple cores, future systems will increasingly favor regular data-level parallel computations, but deviate from the needs of applications with complex patterns. Some evidences are already shown on Graphic Processing Units (GPU): Irregular data accesses (e.g., indirect references A[D[i]]) and conditional branches are limiting many GPU applications' performance at a level an order of magnitude lower than the peak of GPU. The second hardware trend is the growing gap between memory bandwidth and the aggregate speed—that is, the sum of all cores' computing power—of a Chip Multiprocessor (CMP). Despite the capped growth of the peak CPU speed, the aggregate speed of a CMP keeps increasing as more cores get into a single chip. It is expected that by 2018, node concurrency in an exascale system will increase by hundreds of times, whereas, memory bandwidth will expand by only 10 to 20 times. Consequently, data movement and storage is expected to consume more than 70% of the total system power. Bridging this gap is difficult; the complexities of modern CMP memory hierarchy make it even harder: Data cache becomes shared among computing units, and the sharing is often non-uniform—whether two computing units share a cache depends on their proximity and the level of the cache. On the recent IBM Power7 architecture, for instance, four hardware contexts (or SMT threads) in a core share the entire memory hierarchy, all cores in one chip share an on-chip L3 cache, and cores across chips share L3 and main memory through off-chip connections. These two trends complicate the translation of computing power into performance, especially for a program with either intensive data accesses or complex patterns in data accesses or control flow paths. Unfortunately, both attributes present and will persist in a class of important applications. For instance, many scientific simulations deal with a large volume of data. And meanwhile, as most real-world processes are non-uniform and evolving (e.g., the evolution of a galaxy or the process of a drug injection), both the computations and data accesses of these programs tend to be irregular and dynamically changing. Currently, the lack of support to these applications on modern CMP severely limits their performance. On GPU, as our recent study shows and other studies echo, performance enhancement of a factor of integers is possible when memory accesses or control flows are streamlined for a set of GPU applications. On multicore CPU, our studies show that traditional locality enhancement, for being oblivious to the new features of multicore memory hierarchy, may even cause large slowdown to data-intensive dynamic applications. The severity of the issues is expected to worsen as the two hardware trends continue. Some recent studies try to match software with the trends, but in a limited scope or manner. For irregularities on GPU, most studies focus on irregularities analyzable through static analysis (e.g., data accesses in regular loops). Dynamic irregularities are harder to address because the needed analysis and transformations typically have to happen at run time. Some other research resorts to hardware extensions, an actual adoption of which is unclear for the entailed space cost and complexity. For data locality, recent years have seen some exploitations of the new memory hierarchy on multicore for performance, but most of them are on process or thread scheduling, rather than program transformations. Our recent study reveals that program-level transformations may magnify the scheduling benefits by a factor of seven, concluding that program-level transformation should play a central role for data locality enhancement on modern CMP. But research in this direction has been sparse, and most have focused on data layout or cache performance modeling, rather than program transformations to match with the new memory hierarchy features. Overall, it is still an open question how to bridge the gap between dynamic computations and the two prominent properties of modern processors. The goal of this project is to develop a set of techniques and software tools to enhance the matching between memory accesses in dynamic simulations and the prominent features of modern and future CMP, alleviating the memory performance issues for petascale and exascale computing. This report summarizes the discoveries and products produced throughout this project. It includes free launch, a new software approach to overcoming the shortcomings of both methods; coherence-free multiview, an approach that allows multiple views of a single data object to co-exist on GPU memory during a GPU kernel execution; algorithmic optimizations to data analytics problems, especially those that involve lots of distance calculations. Supporting dynamic parallelism is important for GPU to benefit a broad range of applications. There are currently two fundamental ways for programs to exploit dynamic parallelism on GPU: a software-based approach with software-managed worklists, and a hardware-based approach through dynamic subkernel launches. Neither is satisfactory. The former is complicated to program and is often subject to some load imbalance; the latter suffers large runtime overhead. In this work, we propose free launch, a new software approach to overcoming the shortcomings of both methods. It allows programmers to use subkernel launches to express dynamic parallelism. It employs a novel compiler-based code transformation named subkernel launch removal to replace the subkernel launches with the reuse of parent threads. Coupled with an adaptive task assignment mechanism, the transformation reassigns the tasks in the subkernels to the parent threads with a good load balance. The technique requires no hardware extensions, immediately deployable on existing GPUs. It keeps the programming convenience of the subkernel launch-based approach while avoiding its large runtime overhead. Meanwhile, its superior load balancing makes it outperform manual worklist-based techniques by 3X on average. The work was published at the 48th Annual IEEE/ACM International Symposium on Microarchitecture (Micro'2015). A Graphic Processing Unit (GPU) system is typically equipped with many types of memory (e.g., global, constant, texture, shared, cache). Data placement determines what data are placed on which type of memory, essential for GPU memory performance. Prior optimizations of data placement always require a single view of a data object on memory, which limits the optimization effectiveness. In this work, we propose coherence-free multiview, an approach that allows multiple views of a single data object to co-exist on GPU memory during a GPU kernel execution. We demonstrate that under certain conditions, the multiple views can remain incoherent while facilitating enhanced data placement. We present a theorem and some compiler support to ensure the soundness of the usage of coherence-free multiview. We further develop reference-discerning data placement, a new way to enhance data placements on GPU. It enables more flexible data placements by using coherence-free multiview to leverage the slack in coherence requirement of some GPU programs. Experiments on three types of GPU systems show that, with less than 200KB space cost, the new data placement technique can provide a 1.6X average (up to 4.27X) speedup. In addition, we have examined algorithmic optimizations to data analytics problems, especially those that involve lots of distance calculations. Computing distances among data points is an essential part of many important algorithms in data analytics, graph analysis, and other domains. In each of these domains, developers have spent significant manual effort optimizing algorithms, often through novel applications of the triangle equality, in order to minimize the number of distance computations in the algorithms. In this work, we observe that many algorithms across these domains can be generalized as an instance of a generic distance-related abstraction. Based on this abstraction, we derive seven principles for correctly applying the triangular inequality to optimize distance-related algorithms. Guided by the findings, we develop Triangular OPtimizer (TOP), the first software framework that is able to automatically produce optimized algorithms that either matches or outperforms manually designed algorithms for solving distance-related problems. TOP achieves up to 237x speedups and 2.5X on average. The work has been published at the 32nd International Conference on Machine Learning (ICML'15) and the 41st International Conference on Very Large Data Bases (VLDB'15).

42 ENGINEERING↗

Anthropogenic emissions of NOx over China: Reconciling the difference of inverse modeling results using GOME-2 and OMI measurements

Inverse modeling using satellite observations of nitrogen dioxide (NO2) columns has been extensively used to estimate nitrogen oxides (NOx) emissions in China. Recently, the Global Ozone Monitoring Experiment-2 (GOME-2) and Ozone Monitoring Instrument (OMI) provide independent global NO2 column measurements on a nearly daily basis at around 9:30 and 13:30 local time across the equator, respectively. Anthropogenic NOx emission estimates by applying previously developed monthly inversion (MI) or daily inversion (DI) methods to these two sets of measurements show substantial differences. We improve the DI method by conducting model simulation, satellite retrieval, and inverse modeling sequentially on a daily basis. After each inversion, we update anthropogenic NOx emissions in the model simulation with the newly obtained a posteriori results. Consequently, the inversion-optimized emissions are used to compute the a priori NO2 profiles for satellite retrievals. As such, the a priori profiles used in satellite retrievals are now coupled to inverse modeling results. The improved procedure was applied to GOME-2 and OMI NO2 measurements in 2011. The new daily retrieval-inversion (DRI) method estimates an average NOx emission of 6.9 Tg N/yr over China, and the difference between using GOME-2 and OMI measurements is 0.4 Tg N/yr, which is significantly smaller than the difference of 1.3 Tg N/yr using the previous DI method. Using the more consistent DRI inversion results, we find that anthropogenic NOx emissions tend to be higher in winter and summer than spring (and possibly fall) and the weekday-to-weekend emission ratio tends to increase with NOx emission in China.

Gu, Dasa↗

Fuel utilization effects on system efficiency in solid oxide fuel cell gas turbine hybrid systems

Here, a computational analysis was conducted to optimize the design of a solid oxide fuel cell - gas turbine hybrid power generator, focusing on the impact that fuel utilization within the fuel cell has on system efficiency and installed costs. This is the first ever design-study considering the effect of fuel utilization on performance, as well as on the optimum power split. This hybrid system attained high electric generation efficiencies (>70%) over a wide range of operating conditions (60% < fuel utilization < 90%) while the fuel cell stack size decreased in proportion to decreasing the fuel utilization. A one-dimensional fuel cell model was used to simulate the fuel cell while GateCycle® was used to simulate the performance of the associated recuperated turbine and various subsystems necessary for thermal management. For each test case, the size of the solid oxide fuel cell, gas turbine, and recuperator, as well as the fuel and air flow rates, hot-air bypass set point, and heat exchange effectiveness in the solid oxide fuel cell manifold were varied to obtain 550 MWe output. In addition, anode recycle, turbomachinery efficiency, and various thermal management options were tested. The maximum system efficiency (75.6%) was attained for the single-pass solid oxide fuel cell with highly efficient turbomachinery when the solid oxide fuel cell used 80% of the incoming fuel. Efficiency was essentially flat from 75% fuel utilization through 85% fuel utilization. Employing anode recycle starting at 65% resulted in roughly 1 percentage point efficiency decrease for each percent increase in fuel utilization. For minimized solid oxide fuel cell degradation, a near 50:50 power split case was studied resulting in 68.6% efficiency and the solid oxide fuel cell using 55% of the incoming fuel. Because of shifting half of the power generation to the gas turbine, the size of the fuel cell stack was reduced by 25% as compared to that at maximum efficiency (80% fuel utilization).

42 ENGINEERING↗

Lyapunov-based nonlinear control of nonautonomous systems with individual input constraints

A control algorithm that can locally stabilize a specific class of multi-input multi-output nonautonomous nonlinear dynamical systems while satisfying individual input constraints is developed. The proposed Lyapunov-based state-feedback control law inherently accounts for the actuator amplitude saturation limits without the need for computationally expensive real-time optimization techniques. In addition to the control law, a formal definition for the local “controllable region” within which the controller can asymptotically drive the system states to the origin and satisfy the input saturation limits is also presented. The nonautonomous nature of the system dynamics implies that the “controllable region” continuously evolves with time. Therefore, a sufficient condition to maintain the system states within the “controllable region” is proposed in this work to make practical implementation feasible. The effectiveness of the controller is tested for a specific control problem arising in tokamaks, which are toroidal devices that use strong magnetic fields to confine a plasma (hot ionized gas). Here, the primary emphasis of tokamak research is to regulate the plasma properties around predetermined values to achieve stable plasma confinement. Nonlinear simulations show that the proposed controller can achieve the desired plasma control objectives in a DIII-D tokamak scenario.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

A Flexible Transition State Searching Method for Atmospheric Reaction Systems

The precise and rapid exploration of transition states (TSs) is a major challenge when studying atmospheric reactions due to their complexity. In this work, a Monte Carlo Transition State Search Method (MCTSSM), which integrates Monte Carlo sampling technique with transition state optimization methods using an efficient computer script, has been developed for transition state searches. The efficiency and the potential application in atmospheric reactions of this method have been demonstrated by three types of test suits related to the reactions of atmospheric volatile organic compounds (VOCs): (1) OH addition, (2) OH hydrogen-abstraction, and (3) the other reactive group (e.g. Cl, O3, NO3), especially for the reaction of β-pinene-sCI (stabilized Criegee Intermediates) with water. It was shown that the application of this method with effective restricted parameters has greatly simplified the time-consuming and tedious manual search procedure for transition state (TS) of the bimolecular reaction systems.

Aerosol↗

Deep-freeze graph training for latent learning

Scientific and engineering advances are primarily driven by multi-tier conceptual constructs and conditional theoretical frameworks. The theories allow predictions of hypothetical system responses, given a set of approximate conditions (ranges of applicability) imposed on latent parameters that cannot be measured directly. Learning to estimate the latent variables (Latent Learning) helps to pinpoint the anticipated range-edge anomalies and improves the confidence in interpretation, interpolation and extrapolation of limited experimental data. Due to high dimensionality and extreme non-linearity of the materials science problems, very large datasets are typically required for conventional data-driven model development. The vital experimental data collection, particularly on microstructural phases, is very challenging, which makes it difficult to compile a high-quality database. Incorporation of the domain knowledge into the computational graph structure, initialization and optimization processes presents a viable mechanism for developing accurate models, with limited datasets. Furthermore, this study successfully utilized the approach to build the Deep Freeze Graph (DeepFreG) by mapping known causality relationships and by digitizing empirical domain knowledge for Latent Learning (LL), with specific applications in materials science.

36 MATERIALS SCIENCE↗

Machine learning based approach to predict ductile damage model parameters for polycrystalline metals

Damage models for ductile materials typically need to be parameterized, often with the appropriate parameters changing for a given material depending on the loading conditions. This can make parameterizing these models computationally expensive, since an inverse problem must be solved for each loading condition. Using standard inverse modeling techniques typically requires hundreds or thousands of high-fidelity computer simulations to estimate the optimal parameters. Additionally, the time of a human expert is required to set up the inverse model. Machine learning has recently emerged as an alternative approach to inverse modeling in these settings, where the machine learning model is trained in an offline manner and new parameters can be quickly generated on the fly, after training is complete. Here, this work utilizes such a workflow to enable the rapid parameterization of a ductile damage model called TEPLA with a machine learning inverse model. The machine learning model can efficiently estimate the model parameters much faster, as compared to previously employed methods, such as Bayesian calibration. The results demonstrate good accuracy on a synthetic test dataset and is validated against experimental data.

36 MATERIALS SCIENCE↗

Two-layer cathode architecture for high-energy density and high-power density solid state batteries

Solid state batteries with high-energy density and high-power density require the development of thick and energy dense cathodes. Structured cathode electrodes with a double-layer configuration were enabled using a freeze tape casting technique. A bottom dense layer was utilized to enhance the energy density whereas a top porous layer with vertically aligned walls was utilized to enhance the power density. The porous structure of the power layer was achieved by ice templating this layer on top of the densified energy layer of the cathode. This configuration was found to better utilize the active material of the cathodes and was optimized using numerical simulation and computer modeling. Cells with Li metal anode and LiNi 0.6 Mn 0.2 Co 0.2 O 2 (NMC622) at approximately 5 and 20 mg/cm 2 were cycled at 70 °C at different C-rates. Poly(ethylene oxide) (PEO) with lithium bis-trifluoromethanesulfonimide (LiTFSI) was used for the catholyte and the solid-state electrolyte. The structured cathodes exhibited more than double capacity values as well as better Coulombic efficiency compared to non-structured (single-layer) thick cathodes. In conclusion, synchrotron X-ray tomography and scanning electron microscopy were used to characterize the microstructure of the cathodes.

25 ENERGY STORAGE↗

JetGP: A derivative enhanced Gaussian process library

Derivative enhanced Gaussian Processes (DEGPs) can significantly improve surrogate model accuracy over standard Gaussian Process (GP) formulations by incorporating derivative information. However, standard implementations scale poorly with dimension, limiting their use in high dimensional engineering problems. JetGP is a Python framework that unifies existing derivative enhanced GP methodologies into a single library and extends them to support arbitrary order derivative information. The library implements four complementary formulations: standard derivative enhanced Gaussian Processes (DEGP), directional DEGP (DDEGP), generalized directional DEGP (GDDEGP), and weighted DEGP (WDEGP). By unifying these approaches in a consistent interface with robust numerical implementations, JetGP enables practitioners to balance predictive accuracy and computational efficiency for high dimensional optimization, uncertainty quantification, and sensitivity analysis in engineering design.

Derivative enhanced Gaussian process↗

Deep Learning Coordinate-Free Quantum Chemistry

Computing quantum chemical properties of small molecules and polymers can provide insights valuable to physicists, chemists and biologists when designing new materials, catalysts, biological probes and drugs. Deep learning can compute quantum chemical properties accurately in a fraction of the time required by commonly used methods such as density functional theory (DFT). However, many of these deep learning architectures require energy minimized molecular geometries as input, which is also computationally expensive, and decreasing the reproducibility and throughput of these methods. In this study, we demonstrate that accurate quantum chemical computations can be performed without optimized geometries by operating in the coordinate-free domain using deep learning on graph encodings. Furthermore, we also find that the choice of graph-encoding architecture substantially affects the performance of these methods. The Wave architecture outperforms graph convolution architectures, particularly on complex molecules. Furthermore, the structures of these graph encoding architectures provide an opportunity to probe an important, outstanding question in quantum mechanics: What types of quantum chemical properties can be represented by local-variable models? We find that Wave, a local-variable model, is more accurately calculates quantum chemical properties. Graph convolutional architectures require global variables, and are not as effective as as Wave. We anticipate that coordinate-free, deep-learning models of quantum chemistry will become valuable tools in chemistry and biology, enabling researchers to rapidly screen chemical databases or identify new molecules using automated, de-novo design algorithms.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Energy Minimization of Discrete Protein Titration State Models Using Graph Theory

There are several applications in computational biophysics which require the optimization of discrete interacting states; e.g., amino acid titration states, ligand oxidation states, or discrete rotamer angles. Such optimization can be very time-consuming as it scales exponentially in the number of sites to be optimized. In this paper, we describe a new polynomial-time algorithm for optimization of discrete states in macromolecular systems. This algorithm was adapted from image processing and uses techniques from discrete mathematics and graph theory to restate the optimization problem in terms of maximum flow-minimum cut graph analysis. The interaction energy graph, a graph in which vertices (amino acids) and edges (interactions) are weighted with their respective energies, is transformed into a flow network in which the value of the minimum cut in the network equals the minimum free energy of the protein, and the cut itself encodes the state that achieves the minimum free energy. Because of its deterministic nature and polynomial-time performance, this algorithm has the potential to allow for the ionization state of larger proteins to be discovered.

energy minimization↗

Global solutions of restricted open-shell Hartree-Fock theory from semidefinite programming with applications to strongly correlated quantum systems

We present a density matrix approach for computing global solutions of restricted open-shell Hartree-Fock theory, based on semidefinite programming (SDP), that gives upper and lower bounds on the Hartree-Fock energy of quantum systems. While wave function approaches to Hartree-Fock theory yield an upper bound to the Hartree-Fock energy, we derive a semidefinite relaxation of Hartree-Fock theory that yields a rigorous lower bound on the Hartree-Fock energy. We also develop an upper-bound algorithm in which Hartree-Fock theory is cast as a SDP with a nonconvex constraint on the rank of the matrix variable. Equality of the upper- and lower-bound energies guarantees that the computed solution is the globally optimal solution of Hartree-Fock theory. The work extends a previously presented method for closed-shell systems [S. Veeraraghavan and D. A. Mazziotti, Phys. Rev. A 89, 010502–R (2014)]. For strongly correlated systems the SDP approach provides an alternative to the locally optimized Hartree-Fock energies and densities with a certificate of global optimality. Applications are made to the potential energy curves of C{sub 2}, CN, Cr {sub 2}, and NO {sub 2}.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

TeraChem: Accelerating electronic structure and ab initio molecular dynamics with graphical processing units

Developed over the past decade, TeraChem is an electronic structure and ab initio molecular dynamics software package designed from the ground up to leverage graphics processing units (GPUs) to perform large-scale ground and excited state quantum chemistry calculations in the gas and the condensed phase. TeraChem’s speed stems from the reformulation of conventional electronic structure theories in terms of a set of individually optimized high-performance electronic structure operations (e.g., Coulomb and exchange matrix builds, one- and two-particle density matrix builds) and rank-reduction techniques (e.g., tensor hypercontraction). Recent efforts have encapsulated these core operations and provided language-agnostic interfaces. Finally, this greatly increases the accessibility and flexibility of TeraChem as a platform to develop new electronic structure methods on GPUs and provides clear optimization targets for emerging parallel computing architectures.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Tensor decompositions for count data that leverage stochastic and deterministic optimization

There is growing interest to extend low-rank matrix decompositions to multi-way arrays, or tensors. One fundamental low-rank tensor decomposition is the canonical polyadic decomposition (CPD). The challenge of fitting a low-rank, nonnegative CPD model to Poisson-distributed count data is of particular interest. Several popular algorithms use local search methods to approximate the maximum likelihood estimator (MLE) of the Poisson CPD model. Here, this work presents two new algorithms that extend state-of-the-art local methods for Poisson CPD. Hybrid GCP-CPAPR combines Generalized Canonical Decomposition (GCP) with stochastic optimization and CP Alternating Poisson Regression (CPAPR), a deterministic algorithm, to increase the probability of converging to the MLE over either method used alone. Restarted CPAPR with SVDrop uses a heuristic based on the singular values of the CPD model unfoldings to identify convergence toward optimizers that are not the MLE and restarts within the feasible domain of the optimization problem, thus reducing overall computational cost when using a multi-start strategy. We provide empirical evidence that indicates our approaches outperform existing methods with respect to converging to the Poisson CPD MLE.

CPAPR↗