Runtime Speedup Research Articles

Hyperdimensional Brain-Inspired Learning for Phoneme Recognition With Large-Scale Inferior Colliculus Neural Activities.

Develop a novel and highly efficient framework that decodes Inferior Colliculus (IC) neural activities for phoneme recognition. We propose using Hyperdimensional Computing (HDC) to support an efficient phoneme recognition algorithm, in contrast to widely applied Deep Neural Networks (DNN). The high-dimensional representation and operations in HDC are rooted in human brain functionalities and naturally parallelizable, showing the potential for efficient neural activity analysis. Our proposed method includes a spatial and temporal-aware HDC encoder that effectively captures global and local patterns. As part of our framework, we deploy the lightweight HDC-based algorithm on a highly customizable and flexible hardware platform, i.e., Field Programmable Gate Arrays (FPGA), for optimal algorithm speedup. To evaluate our method, we record IC neural activities on gerbils while playing the sound of different phonemes. We compare our proposed method with multiple baseline machine learning algorithms in recognition quality and learning efficiency, across different hardware platforms. The results show that our method generally achieves better classification quality than the best-performing baseline. Compared to the Deep Residual Neural Network (i.e., ResNet), our method shows a speedup up to 74×, 67×, 210× on CPU, GPU, and FPGA respectively. We achieve up to 15% (10%) higher accuracy in consonant (vowel) classification than ResNet. By leveraging brain-inspired HDC for IC neural activity encoding and phoneme classification, we achieve orders of magnitude runtime speedup while improving accuracy in various challenging task settings. Decoding IC neural activities is an important step to enhance understanding about human auditory system. However, these responses from the central auditory system are noisy and contain high variance, demanding large-scale datasets and iterative model fine-tuning. The proposed HDC-based framework is more scalable and viable for future real-world deployment thanks to its fast training and overall better quality.

Linear Sources and Anisotropic Scattering in the Random Ray Method

An enhanced implementation of the random ray method (TRRM), a stochastic adaptation of the method of characteristics, is presented. This implementation generalizes from a traditional flat source (FS) approximation to a linear source (LS) approximation, and from the standard isotropic scattering approximation to an anisotropic approximation, up to P3. On the two-dimensional C5G7 benchmark, LS alone enables a 1.4× and 1.7× enhancement in run time for parallel and serial executions, respectively, without compromising accuracy. Further testing on the three-dimensional C5G7 Rodded B benchmark demonstrates that LS enables significant axial coarsening and reduction in ray populations. This results in a 5.3× speedup in run time while still offering comparable accuracy. On the Babcock and Wilcox 1484 Core II benchmark, incorporating anisotropic scattering in TRRM up to P3 with a FS decreased keff errors (~110 pcm) and maximum and average pin power errors. LS with anisotropic scattering (LSPN), showed similarly improved keff errors while benefiting from a coarser mesh. The linear isotropic flat anisotropic (LIFA) method, relative to LSPN, offers further run time and memory savings of 4.1× and 1.6×, respectively, for P3, while maintaining comparable accuracy to anisotropic FS on more refined meshes.

CAVE: Concurrency-Aware Graph Processing on SSDs

Large-scale graph analytics has become increasingly common in areas like social networks, physical sciences, transportation networks, and recommendation systems. Since many such practical graphs do not fit in main memory, graph analytics performance depends on efficiently utilizing underlying storage devices. These out-of-core graph processing systems employ sharding and sub-graph partitioning to optimize for storage while relying on efficient sequential access of traditional hard disks. However, today's storage is increasingly based on solid-state drives (SSDs) that exhibit high internal parallelism and efficient random accesses. Yet, state-of-the-art graph processing systems do not explicitly exploit those properties, resulting in subpar performance. In this paper, we develop CAVE, the first graph processing engine that optimally exploits underlying SSD-based storage by harnessing the available storage device parallelism via carefully selecting which I/Os to graph data can be issued concurrently. Thus, CAVE traverses multiple paths and processes multiple nodes and edges concurrently, achieving parallelization at a granular level. We identify two key ways to parallelize graph traversal algorithms based on the graph structure and algorithm: intra-subgraph and inter-subgraph parallelization. The first identifies subgraphs that contain vertices that can be accessed in parallel, while the latter identifies subgraphs that can be processed in their entirety in parallel. To showcase the benefit of our approach, we build within CAVE parallelized versions of five popular graph algorithms (Breadth-First Search, Depth-First Search, Weakly Connected Components, PageRank, Random Walk) that exploit the full bandwidth of the underlying device. CAVE uses a blocked file format based on adjacency lists and employs a concurrent cache pool that is essential to the parallelization of graph algorithms. By experimenting with different types of graphs on three SSD devices, we demonstrate that CAVE utilizes the available parallelism, and scales to diverse real-world graph datasets. CAVE achieves up to one order of magnitude speedup compared to the popular out-of-core systems Mosaic and GridGraph, and up to three orders of magnitude speedup in runtime compared to GraphChi.

Graph network surrogate model for subsurface flow optimization

The optimization of well locations and controls is an important step in the design of subsurface flow operations such as oil production or geological CO2 storage. These optimization problems can be computationally expensive, however, as many potential candidate solutions must be evaluated. In this study, we propose a graph network surrogate model (GNSM) for optimizing well placement and controls. The GNSM transforms the flow model into a computational graph that involves an encoding-processing-decoding architecture. Separate networks are constructed to provide global predictions for the pressure and saturation state variables. Model performance is enhanced through the inclusion of the single-phase steady-state pressure solution as a feature. A multistage multistep strategy is used for training. The trained GNSM is applied to predict flow responses in a 2D unstructured model of a channelized reservoir. Results are presented for a large set of test cases, in which five injection wells and five production wells are placed randomly throughout the model, with a random control variable (bottom-hole pressure) assigned to each well. Median relative error in pressure and saturation for 300 such test cases is 1-2%. The ability of the trained GNSM to provide accurate predictions for new (geologically similar) permeability realizations is demonstrated. Finally, the trained GNSM is used to optimize well locations and controls with a differential evolution algorithm. GNSM-based optimization results are comparable to those from simulation-based optimization, with a runtime speedup factor of 36. Much larger speedups are expected if the method is used for robust optimization, in which each candidate solution is evaluated on multiple geological models.

D 3 PBO: D ynamic D omain D ecomposition-based P arallel B ayesian O ptimization for Large-scale Analog Circuit Sizing

Bayesian optimization (BO) is an efficient global optimization method for expensive black-box functions, but the expansion for high-dimensional problems and large sample budgets still remains a severe challenge. In order to extend BO for large-scale analog circuit synthesis, a novel computationally efficient parallel BO method, D 3 PBO, is proposed for high-dimensional problems in this work. We introduce the dynamic domain decomposition method based on maximum variance between clusters. The search space is decomposed into subdomains progressively to limit the maximal number of observations in each domain. The promising domain is explored by multi-trust region-based batch BO with the local Gaussian process (GP) model. As the domain decomposition progresses, the basin-shaped domain is identified using a GP-assisted quadratic regression method and exploited by the local search method BOBYQA to achieve a faster convergence rate. The time complexity of D 3 PBO is constant for each iteration. Experiments demonstrate that D 3 PBO obtains better results with significantly less runtime consumption compared to state-of-the-art methods. For the circuit optimization experiments, D 3 PBO achieves up to 10× runtime speedup compared to TuRBO with better solutions.

Combustion simulations of scramjet combustor using reduced mechanism of surrogate fuel for regenerative cooling pyrolysis products

Taking the surrogate fuel (64% ethylene and 36% methane in mole percentage) for regenerative cooling pyrolysis products used in HIFiRE-2 scramjet combustor as an example, present work systematically explores the workflow of the integrated mechanism reduction for surrogate fuel of pyrolysis products, the kinetic performance verification of the preferred reduced mechanism, and the combustion simulation application of the reduced mechanism in scramjet combustor. A static integrated reduction strategy is performed to obtain reduced mechanism for the surrogate fuel with the NUIGMech1.2 as detailed mechanism under wide conditions for temperature range of 900 - 1800 K, pressure range of 1 - 4 atm, and equivalence ratio range of 0.25 - 5.0. A reduced mechanism (34 species and 181 reactions)with remarkably reduced size is obtained, which presents favorable performance incomprehensive kinetic validations. With this compact and high-fidelity reduced mechanism, the combustion simulations for the scramjet combustor are carried out combining with tabulation of dynamic adaptive chemistry for run-time speed-up. The simulation results of static pressure profiles obtained for cold and hot states match well with the experimental measurements for the two conditions with flight Mach number of 5.84 and 6.50. Meanwhile, the flow and combustion characteristics of the two conditions are investigated based on simulation results. The integrated reduction strategy and systematic kinetic verification used in present work provide reference values for the application of more complex surrogate fuel mechanisms in scramjet combustor combustion simulation.

Data-driven surrogate modelling of residual stresses in Laser Powder-Bed Fusion

ABSTRACT In order to enable the industrialization of additive manufacturing, it is necessary to develop process simulation models that can rapidly predict part quality. Although multi-physics simulations have shown success at predicting residual stress, distortion, microstructure and mechanical properties of additively manufactured parts, they are generally too computationally expensive to be directly used in applications, such as optimization, controls, or digital twinning. In this study, a critical evaluation is made of how data-driven surrogate models can be used to model the residual stress of parts fabricated by Laser Powder-Bed Fusion. Residual stress data is generated by using an inherent-strain based process simulation for two families of part geometries. Three different models using varying levels of sophistication are compared: a multilayer perceptron (MLP), a convolutional neural network (CNN) based on the U-Net architecture, and an interpolation-based method based on mapping geometries onto a reference. All three methods were found to be sufficient for part design, providing mechanical predictions for a CPU time below 0.2 s, representing a runtime speed-up of at least 3900 × . Neural network-based models are significantly more expensive to train compared to using interpolation. However, the generality of models based on the U-Net architecture is attractive for applications in optimization.

Exploiting Functional Constraints in Automatic Dominance Breaking for Constraint Optimization

Dominance breaking is a powerful technique in improving the solving efficiency of Constraint Optimization Problems (COPs) by removing provably suboptimal solutions with additional constraints. While dominance breaking is effective in a range of practical problems, it is usually problem specific and requires human insights into problem structures to come up with correct dominance breaking constraints. Recently, a framework is proposed to generate nogood constraints automatically for dominance breaking, which formulates nogood generation as solving auxiliary Constraint Satisfaction Problems (CSPs). However, the framework uses a pattern matching approach to synthesize the auxiliary generation CSPs from the specific forms of objectives and constraints in target COPs, and is only applicable to a limited class of COPs. This paper proposes a novel rewriting system to derive constraints for the auxiliary generation CSPs automatically from COPs with nested function calls, significantly generalizing the original framework. In particular, the rewriting system exploits functional constraints flattened from nested functions in a high-level modeling language. To generate more effective dominance breaking nogoods and derive more relaxed constraints in generation CSPs, we further characterize how to extend the system with rewriting rules exploiting function properties, such as monotonicity, commutativity, and associativity, for specific functional constraints. Experimentation shows significant runtime speedup using the dominance breaking nogoods generated by our proposed method. Studying patterns of generated nogoods also demonstrates that our proposal can reveal dominance relations in the literature and discover new dominance relations on problems with ineffective or no known dominance breaking constraints.

Better Defunctionalization through Lambda Set Specialization

Higher-order functions pose a challenge for both static program analyses and optimizing compilers. To simplify the analysis and compilation of languages with higher-order functions, a rich body of prior work has proposed a variety of defunctionalization techniques, which can eliminate higher-order functions from a program by transforming the program to a semantically-equivalent first-order representation. Several modern languages take this a step further, specializing higher-order functions with respect to the functions on which they operate, and in turn allowing compilers to generate more efficient code. However, existing specializing defunctionalization techniques restrict how function values may be used, forcing implementations to fall back on costly dynamic alternatives. We propose lambda set specialization (LSS), the first specializing defunctionalization technique which imposes no restrictions on how function values may be used. We formulate LSS in terms of a polymorphic type system which tracks the flow of function values through the program, and use this type system to recast specialization of higher-order functions with respect to their arguments as a form of type monomorphization. We show that our type system admits a simple and tractable type inference algorithm, and give a formalization and fully-mechanized proof in the Isabelle/HOL proof assistant showing soundness and completeness of the type inference algorithm with respect to the type system. To show the benefits of LSS, we evaluate its impact on the run time performance of code generated by the MLton compiler for Standard ML, the OCaml compiler, and the new Morphic functional programming language. We find that pre-processing with LSS achieves run time speedups of up to 6.85x under MLton, 3.45x for OCaml, and 78.93x for Morphic.

Emulating microstructural evolution during spinodal decomposition using a tensor decomposed convolutional and recurrent neural network

Phase-field (PF) models are one of the most powerful tools to simulate microstructural evolution in metallic materials, polymers, and ceramics. However, existing PF approaches rely on rigorous mathematical model development, sophisticated numerical schemes, and high-performance computing for accuracy. Although recently developed surrogate microstructure models employ deep-learning techniques and reconstruction of microstructures from lower-dimensional data, their accuracy is fairly limited as spatiotemporal information is lost in the pursuit of dimensional reduction. Given these limitations, we present a novel data-driven emulator (DDE) for predicting microstructural evolution, which combines an image-based convolutional and recurrent neural network (CRNN) with tensor decomposition, while leveraging previously obtained PF datasets for training. To assess the robustness of DDE, we also compare the emulation sequence and the scaling behavior with phase-field simulations for several noisy initial states. Finally, we discuss the effectiveness of our microstructure emulation technique in the context of runtime speed-up while also highlighting its trade-off with accuracy.

A Hybrid Embedded Discrete Fracture Model and Dual-Porosity, Dual-Permeability Workflow for Hierarchical Treatment of Fractures in Practical Field Studies

Summary Natural fracture systems comprise numerous small features and relatively few large ones. At field scale, it is impractical to treat all fractures explicitly. We represent the largest fractures using an embedded discrete fracture model (EDFM) and account for smaller ones using a dual-porosity, dual-permeability (DPDK) idealized representation of the fracture network. The hybrid EDFM + DPDK approach uses consistent discretization schemes and efficiently simulates realistic field cases. Further speedup can be obtained using aggregation-based upscaling. Capabilities to visualize and post-process simulation results facilitate understanding for effective management of fractured reservoirs. The proposed approach embeds large discrete fractures as EDFM within a DPDK grid (which contains both matrix and idealized fracture continua for smaller fractures) and captures all connections among the triple media. In contrast with existing EDFM formulations, we account for discrete fracture spacing within each matrix cell via a new matrix-fracture transfer term and use consistent assumptions for classical EDFM and DPDK calculations. In addition, the workflow enables coarse EDFM representations using flow-based cell-aggregation upscaling for computational efficiency. Using a synthetic case, we show that the proposed EDFM + DPDK approach provides a close match of simulation results from a reference model that represents all fractures explicitly, while providing runtime speedup. It is also more accurate than previous standard EDFM and DPDK models. We demonstrate that the matrix-fracture transfer function agrees with flow-based upscaling of high-resolution fracture models. Next, the automated workflow is applied to a waterflooding study for a giant carbonate reservoir, with an ensemble of stochastic fracture realizations. The overall workflow provides the computational efficiency needed for performance forecasts in practical field studies, and the 3D visualization allows for the derivation of insights into recovery mechanisms. Finally, we apply a finite-volume tracer-based flux post-processing scheme on simulation results to analyze production allocation and sweep for understanding expected waterflood performance.

QAOA-in-QAOA: Solving Large-Scale MaxCut Problems on Small Quantum Machines

The design of fast algorithms for combinatorial optimization greatly contributes to a plethora of domains such as logistics, finance, and chemistry. Quantum approximate optimization algorithms (QAOAs), which utilize the power of quantum machines and inherit the spirit of adiabatic evolution, are approaches to tackle combinatorial problems with potential runtime speedups. However, hurdled by the limited quantum resources nowadays, QAOAs are infeasible to manipulate large-scale problems. To address this issue, here we revisit the MaxCut problem via the divide-and-conquer heuristic: seek the solutions of subgraphs in parallel and then merge these solutions to obtain the global solution. Because of the ${\mathbb{Z}}_{2}$ symmetry in MaxCut, we prove that the merging process can be further cast into a new MaxCut problem and thus be addressed by QAOAs or other MaxCut solvers. In view of this, we propose QAOA-in-QAOA (${\mathrm{QAOA}}^{2}$) to solve arbitrary large-scale MaxCut problems using small quantum machines. We also prove that the performance of ${\mathrm{QAOA}}^{2}$ is lower bounded with respect to the divide-and-conquer process. Experiment results illustrate that, under different graph settings, ${\mathrm{QAOA}}^{2}$ attains a competitive or even better performance over classical algorithms when the node count is around $2000$. Our method can be seamlessly embedded into other advanced strategies to enhance the capability of QAOAs in large-scale combinatorial optimization problems.

Convolutional – recurrent neural network proxy for robust optimization and closed-loop reservoir management

Production optimization under geological uncertainty is computationally expensive, as a large number of well control schedules must be evaluated over multiple geological realizations. In this work, a convolutional-recurrent neural network (CNN-RNN) proxy model is developed to predict well-by-well oil and water rates, for given time-varying well bottom-hole pressure (BHP) schedules, for each realization in an ensemble. This capability enables the estimation of the objective function and nonlinear constraint values required for robust optimization. The proxy model represents an extension of a recently developed long short-term memory (LSTM) RNN proxy designed to predict well rates for a single geomodel. A CNN is introduced here to processes permeability realizations, and this provides the initial states for the RNN. The CNN-RNN proxy is trained using simulation results for 300 different sets of BHP schedules and permeability realizations. We demonstrate proxy accuracy for oil-water flow through multiple realizations of 3D multi-Gaussian permeability models. The proxy is then incorporated into a closed-loop reservoir management (CLRM) workflow, where it is used with particle swarm optimization and a filter-based method for nonlinear constraint satisfaction. History matching is achieved using an adjoint-gradient-based procedure. The proxy model is shown to perform well in this setting for five different (synthetic) ‘true’ models. Improved net present value along with constraint satisfaction and uncertainty reduction are observed with CLRM. For the robust production optimization steps, the proxy provides O(100) runtime speedup over simulation-based optimization.

Hybrid Quantum-Classical Generative Adversarial Network for synthesizing chemically feasible molecules

Current drug discovery processes can cost billions of dollars and usually take five to ten years. People have been researching and implementing various computational approaches to search for molecules and compounds from the chemical space, which can be on the order of 1060 molecules. One solution involves deep generative models, which are artificial intelligence models that learn from nonlinear data by modeling the probability distribution of chemical structures and creating similar data points from the trends it identifies. These generative models can extract salient features that characterize the molecules. However, they often suffer from increased time complexity. Aiming for faster runtime and greater robustness when analyzing high-dimensional data, we designed and implemented a Hybrid Quantum-Classical Generative Adversarial Network (QGAN) to synthesize molecules. There are two parts in the QGAN: a quantum generator that creates molecules based on the probability distributions of likely combinations and a classical discriminator that differentiates real molecules from the generated ones. We hypothesized that a quantum generator would be more impactful, offering a runtime speedup and increasing generated molecule possibilities, because we could use quantum mechanical phenomena superposition, entanglement, and interference to analyze more atom-bond combinations than a classical generator. The PyTorch-PennyLane implementation of the QGAN generated seven chemically stable molecules out of 300, a 2.3% success rate. Although the QGAN is a work in progress, it has demonstrated a path towards more efficient drug development, which would accelerate the development of medicines and reduce costs for the whole R&D process.

Building a post-layout simulation performance model with global mapping model fusion technique

Building a post-layout simulation performance model is essential in closing the loop of analog circuits, but it is a challenging task because of the high-dimensional space and expensive simulation cost. To facilitate efficient modeling, this paper proposes a Global Mapping Model Fusion (GMMF) technique. The key idea of GMMF is to reuse the schematic-level model trained by the Artificial Neural Network (ANN) algorithm, and combine it with few mapping coefficients to build the post-simulation model. Furthermore, as an efficient global optimization algorithm, differential evolution is applied to determine the optimal mapping coefficients with few samples. In GMMF, only a small number of mapping coefficients are unknown, so the number of post-layout samples needed is significantly reduced. To enhance practical utility of the proposed GMMF technique, two specific mapping relations, i.e., linear or weakly no-linear and nonlinear, are carefully considered in this paper. We conduct experiments on two topologies of two-stage operational amplifier and comparator in different commercial processes. All the simulation data for modeling are obtained from a parametric design framework. A more than 5× runtime speedup is achieved over ANN without surrendering any accuracy.

GPR-Based Framework for Statistical Analysis of Gate Delay under NBTI and Process Variation Effects

With the aggressive scaling of feature size, Negative Bias Temperature Instability (NBTI) and Process Variation (PV) have become major issues for circuit reliability and yield. In this paper, we analyze the variation of gate delay by jointly considering NBTI and PV effects. Using Gaussian Process Regression (GPR) learning interface, a Statistical Gate Delay Extraction (SGDE) framework is proposed. Typical types of logic gates are simulated with commercial 28 nm technology for verifying the performance of SGDE in the experiment. Compared to the golden data, the results show that our proposed approach achieves minimal loss of accuracy with significant runtime speed-up.

Universal Virtual Lab: A Fast and Accurate Simulation Tool for Wideband Nonlinear DWDM Systems

We introduce the concept of the universal virtual lab, an extension to the virtual lab platform of [Golani <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">et al.</i> 2016], enabling a fast and accurate simulation of wideband nonlinear DWDM systems. The universal virtual lab is compliant with non-ideal transmitter and receiver architectures, distributed optical filters in the optical link, inter-channel stimulated Raman scattering, and it provides accurate performance predictions even when adaptive equalization methods are applied. In comparison with the conventional full-bandwidth split step Fourier transform method, we show with different test scenarios that the universal virtual lab provides accuracy errors below 0.1 dBQ and 0.09 bit/4D-symb in Q-factor and GMI assessments respectively, with runtime speedup factors exceeding 1000. We also report performance assessments in an ultra-wideband (11 THz) C+L system and discuss equalization gain under different compensation scenarios. The estimated speedup factor with respect to the full-bandwidth split step Fourier transform method is assessed to be greater than 35,000.

The Pareto-optimal temporal aggregation of energy system models

The growing share of intermittent renewable energy sources, storage technologies, and the increasing degree of so-called sector coupling necessitates optimization-based energy system models with high temporal and spatial resolutions, which significantly increases their runtimes and limits their maximum sizes. In order to maintain the computational viability of these models for large-scale application cases, temporal aggregation has emerged as a technique for reducing the number of considered time steps by reducing the original time horizon down to fewer, more representative ones.This study presents advanced but generally applicable clustering techniques that allow for ad-hoc improvements of current approaches without requiring profound knowledge of the individual energy system model. These improvements comprise a method to find the optimal tradeoff between the number of typical days and inner-daily temporal resolutions, and a representation method that can reproduce the value distribution of the original time series. We prove the superiority of these approaches by applying them to two fundamentally different model types and benchmarking them against state-of-the-art approaches. This is performed for a variety of temporal resolutions, which leads to many hundreds of model runs.The results show that the proposed improvements on current methods strictly dominate the status quo with respect to Pareto-optimality in terms of runtime and accuracy. Although a speeding up factor of one magnitude could be achieved using traditional aggregation methods within a cost deviation range of two percent, the algorithms proposed herein achieve this accuracy with a runtime speedup by a factor of two orders of magnitude.

Rank-adaptive structure-preserving model order reduction of Hamiltonian systems

This work proposes an adaptive structure-preserving model order reduction method for finite-dimensional parametrized Hamiltonian systems modeling non-dissipative phenomena. To overcome the slowly decaying Kolmogorov width typical of transport problems, the full model is approximated on local reduced spaces that are adapted in time using dynamical low-rank approximation techniques. The reduced dynamics is prescribed by approximating the symplectic projection of the Hamiltonian vector field in the tangent space to the local reduced space. This ensures that the canonical symplectic structure of the Hamiltonian dynamics is preserved during the reduction. In addition, accurate approximations with low-rank reduced solutions are obtained by allowing the dimension of the reduced space to change during the time evolution. Whenever the quality of the reduced solution, assessed via an error indicator, is not satisfactory, the reduced basis is augmented in the parameter direction that is worst approximated by the current basis. Extensive numerical tests involving wave interactions, nonlinear transport problems, and the Vlasov equation demonstrate the superior stability properties and considerable runtime speedups of the proposed method as compared to global and traditional reduced basis approaches.

An Optimal Approximation for Submodular Maximization Under a Matroid Constraint in the Adaptive Complexity Model

An Exponentially Faster Algorithm for Submodular Maximization Under a Matroid Constraint This paper studies the problem of submodular maximization under a matroid constraint. It is known since the 1970s that the greedy algorithm obtains a constant-factor approximation guarantee for this problem. Twelve years ago, a breakthrough result by Vondrák obtained the optimal 1 − 1/e approximation. Previous algorithms for this fundamental problem all have linear parallel runtime, which was considered impossible to accelerate until recently. The main contribution of this paper is a novel algorithm that provides an exponential speedup in the parallel runtime of submodular maximization under a matroid constraint, without loss in the approximation guarantee.