Memory-efficient Method Research Articles

Edge illumination x-ray phase contrast simulations using the CAD-ASTRA toolbox.

Edge illumination x-ray phase contrast imaging (XPCI) provides increased contrast for low absorbing materials compared to attenuation images and sheds light on the material microstructure through dark field contrast. To apply XPCI in areas such as non-destructive testing and inline inspection, where scanned samples are increasingly compared to simulated reference images, accurate and efficient simulation software is required. However, currently available simulators rely on expensive Monte Carlo techniques or wave-optics frameworks, resulting in long simulation times. Furthermore, these simulators are often not optimized to work with computer-aided design (CAD) models, a common and memory-efficient method to represent manufactured objects, hindering their integration in an inspection pipeline. In this work, we address these shortcomings by introducing an edge illumination XPCI simulation framework built upon the recently developed CAD-ASTRA toolbox. CAD-ASTRA allows for the efficient simulation of x-ray projections from CAD models through GPU-accelerated ray tracing and supports ray refraction in a geometric optics framework. The edge illumination implementation is validated and its performance is benchmarked against GATE, a state-of-the-art Monte Carlo simulator, revealing a simulation speed increase of up to three orders of magnitude, while maintaining high accuracy in the resulting images.

An efficient algorithm for the computation of the First Passage Time maps of a given signal

The First Passage Time (FPT) of a given signal is the time required to reach a level Xf for the first time, when starting from a level X0. For a given set (X0,Xf), this time can be obtained by moving along the given signal and clocking the time on when passing through level X0, then off when passing through level Xf for the first time. For random signals, the FPT is a random variable. Several samples of the first passage time are collected while moving along the given signal to compute the corresponding statistical moments and probability density function. The FPT map associates each statistical moment with every couple (X0,Xf). A map is constructed for every desired moment. A naive implementation of the signal processing method to compute the FPT map would require to move along the signal as many times as the desired pairs (X0,Xf). Instead, this paper introduces a new and optimized algorithm to establish the FPT maps of any statistical moment and, optionally, of the whole probability distribution. This algorithm combines different features to produce a computationally effective and memory-efficient method. The algorithm estimates the entire map by iterating only once across the signal. The accuracy of the algorithm has been assessed with particular cases where theoretical solutions exist as well as for lab data. Furthermore the algorithm’s performance has also been evaluated in terms of computational burden. A specific attention is given to the computation time scaling for the evaluation of an FPT map, with respect to the signal size.

μ- PBWT: a lightweight r-indexing of the PBWT for storing and querying UK Biobank data.

The Positional Burrows-Wheeler Transform (PBWT) is a data structure that indexes haplotype sequences in a manner that enables finding maximal haplotype matches in h sequences containing w variation sites in O(hw) time. This represents a significant improvement over classical quadratic-time approaches. However, the original PBWT data structure does not allow for queries over Biobank panels that consist of several millions of haplotypes, if an index of the haplotypes must be kept entirely in memory. In this article, we leverage the notion of r-index proposed for the BWT to present a memory-efficient method for constructing and storing the run-length encoded PBWT, and computing set maximal matches (SMEMs) queries in haplotype sequences. We implement our method, which we refer to as μ-PBWT, and evaluate it on datasets of 1000 Genome Project and UK Biobank data. Our experiments demonstrate that the μ-PBWT reduces the memory usage up to a factor of 20% compared to the best current PBWT-based indexing. In particular, μ-PBWT produces an index that stores high-coverage whole genome sequencing data of chromosome 20 in about a third of the space of its BCF file. μ-PBWT is an adaptation of techniques for the run-length compressed BWT for the PBWT (RLPBWT) and it is based on keeping in memory only a succinct representation of the RLPBWT that still allows the efficient computation of set maximal matches (SMEMs) over the original panel. Our implementation is open source and available at https://github.com/dlcgold/muPBWT. The binary is available at https://bioconda.github.io/recipes/mupbwt/README.html.

Fast inference of genetic recombination rates in biobank scale data.

Although rates of recombination events across the genome (genetic maps) are fundamental to genetic research, the majority of current studies only use one standard map. There is evidence suggesting population differences in genetic maps, and thus estimating population-specific maps, are of interest. Although the recent availability of biobank-scale data offers such opportunities, current methods are not efficient at leveraging very large sample sizes. The most accurate methods are still linkage disequilibrium (LD)-based methods that are only tractable for a few hundred samples. In this work, we propose a fast and memory-efficient method for estimating genetic maps from population genotyping data. Our method, FastRecomb, leverages the efficient positional Burrows-Wheeler transform (PBWT) data structure for counting IBD segment boundaries as potential recombination events. We used PBWT blocks to avoid redundant counting of pairwise matches. Moreover, we used a panel-smoothing technique to reduce the noise from errors and recent mutations. Using simulation, we found that FastRecomb achieves state-of-the-art performance at 10-kb resolution, in terms of correlation coefficients between the estimated map and the ground truth. This is mainly because FastRecomb can effectively take advantage of large panels comprising more than hundreds of thousands of haplotypes. At the same time, other methods lack the efficiency to handle such data. We believe further refinement of FastRecomb would deliver more accurate genetic maps for the genetics community.

Memory efficient model based deep learning reconstructions for high spatial resolution 3D non-cartesian acquisitions

Objective. Model based deep learning (MBDL) has been challenging to apply to the reconstruction of 3D non-Cartesian MRI due to GPU memory demand because the entire volume is needed for data-consistency steps embedded in the model. This requirement makes holding even a single unroll in GPU memory difficult meaning memory efficient techniques used to increase unroll number like gradient checkpointing and deep equilibrium learning will not work well for high spatial resolution 3D non-Cartesian reconstructions without modification. Here we develop a memory efficient method called block-wise learning that combines gradient checkpointing with patch-wise training to overcome this obstacle and allow for fast and high-quality 3D non-Cartesian reconstructions using MBDL. Approach. Block-wise learning applied to a single unroll decomposes the input volume into smaller patches, gradient checkpoints each patch, passes each patch iteratively through a neural network regularizer, and then rebuilds the full volume from these output patches for data-consistency. This method is applied across unrolls during training. Block-wise learning significantly reduces memory requirements by tying GPU memory for a single unroll to user selected patch size instead of the full volume. This algorithm was used to train a MBDL architecture to reconstruct highly undersampled, 1.25 mm isotropic, pulmonary magnetic resonance angiography volumes with matrix sizes varying from 300–450 × 200–300 × 300–450 on a single GPU. We compared block-wise learning reconstructions against L1 wavelet compressed reconstructions and proxy ground truth images. Main results. MBDL with block-wise learning significantly improved image quality relative to L1 wavelet compressed sensing while simultaneously reducing average reconstruction time 38x. Significance. Block-wise learning allows for MBDL to be applied to high spatial resolution, 3D non-Cartesian datasets with improved image quality and significant reductions in reconstruction time relative to traditional iterative methods.

Benchmarking automated cell type annotation tools for single-cell ATAC-seq data.

As single-cell chromatin accessibility profiling methods advance, scATAC-seq has become ever more important in the study of candidate regulatory genomic regions and their roles underlying developmental, evolutionary, and disease processes. At the same time, cell type annotation is critical in understanding the cellular composition of complex tissues and identifying potential novel cell types. However, most existing methods that can perform automated cell type annotation are designed to transfer labels from an annotated scRNA-seq data set to another scRNA-seq data set, and it is not clear whether these methods are adaptable to annotate scATAC-seq data. Several methods have been recently proposed for label transfer from scRNA-seq data to scATAC-seq data, but there is a lack of benchmarking study on the performance of these methods. Here, we evaluated the performance of five scATAC-seq annotation methods on both their classification accuracy and scalability using publicly available single-cell datasets from mouse and human tissues including brain, lung, kidney, PBMC, and BMMC. Using the BMMC data as basis, we further investigated the performance of these methods across different data sizes, mislabeling rates, sequencing depths and the number of cell types unique to scATAC-seq. Bridge integration, which is the only method that requires additional multimodal data and does not need gene activity calculation, was overall the best method and robust to changes in data size, mislabeling rate and sequencing depth. Conos was the most time and memory efficient method but performed the worst in terms of prediction accuracy. scJoint tended to assign cells to similar cell types and performed relatively poorly for complex datasets with deep annotations but performed better for datasets only with major label annotations. The performance of scGCN and Seurat v3 was moderate, but scGCN was the most time-consuming method and had the most similar performance to random classifiers for cell types unique to scATAC-seq.

Erasable Virtual HyperLogLog for Approximating Cumulative Distribution over Data Streams

Many real-world datasets are given in the stream of entity-identifier pairs, and measuring data distribution on these datasets is fundamental for applications such as privacy protection. In this paper, we study the problem of computing the cumulative distribution for different cardinalities (i.e., the number of distinct entities owning the same identifier). However, previous sketch-based methods cost large memory space especially when there are a large number of identifiers, and sampling-based methods require much time for cardinality estimation. A recent work KHyperLogLog combines both sketch and sampling methods but it is wasteful to separately build a HyperLogLog sketch of large size for identifiers with small cardinalities. To address these challenges, we propose a memory-efficient method EV-HLL, which designs a shared structure to store all sampled identifiers and their entities and utilizes additional sketches to track value updates during the sampling procedure. Meanwhile, EV-HLL provides real-time unbiased estimations according to value changes whenever a new entity-identifier pair arrives. We evaluate the performance of EV-HLL and other state-of-the-arts on real-world available datasets. Experimental results demonstrate that comparing to other methods, EV-HLL effectively reduces their memory usage with the same estimation accuracy and has higher accuracy with the same memory usage.

A robust and memory-efficient transition state search method for complex energy landscapes.

Locating transition states is crucial for investigating transition mechanisms in wide-ranging phenomena, from atomistic to macroscale systems. Existing methods, however, can struggle in problems with a large number of degrees of freedom, on-the-fly adaptive remeshing and coarse-graining, and energy landscapes that are locally flat or discontinuous. To resolve these challenges, we introduce a new double-ended method, the Binary-Image Transition State Search (BITSS). It uses just two states that converge to the transition state, resulting in a fast, flexible, and memory-efficient method. We also show that it is more robust compared to existing bracketing methods that use only two states. We demonstrate its versatility by applying BITSS to three very different classes of problems: Lennard-Jones clusters, shell buckling, and multiphase phase-field models.

An Accuracy Controllable and Memory Efficient Method for Computing High-Quality Geodesic Distances on Triangle Meshes

This paper presents a new method for computing approximate geodesic distances and paths on triangle meshes. Our method combines two state-of-the-art discrete geodesic methods, which are discrete geodesic graphs (DGG) and vertex-oriented triangle propagation (VTP), so that it allows the user to specify the desired accuracy using a single parameter ɛ . The method, called DGG-VTP, extends the conventional window propagation framework by monitoring the accuracy of the computed distances so that propagation can terminate immediately when the desired accuracy is reached. It is worth noting that for robustness consideration, tiny windows with length less than a threshold (usually, between 1 0 − 7 and 1 0 − 6 ) are discarded in the implementation of the existing exact algorithms, such as the Mitchel–Mount–Papadimitriou (MMP) algorithm, the Chen–Han (CH) algorithm and their many variants. By setting the accuracy parameter ɛ ∈ [ 1 0 − 7 , 1 0 − 6 ] , our method can produce results with comparable accuracy to VTP, while being 3–40 times faster and consuming much less memory. Furthermore, the performance of our method is less sensitive to mesh tessellation than what VTP does. Our method empirically produces O n ɛ 0 . 23 windows and scales well to deal with large-scale models. Though the parameter ɛ in DGG-VTP is not a guaranteed error bound, it acts as an intuitive guide for the user to set the desired accuracy. Extensive evaluations demonstrate the effectiveness of our accuracy control: given a parameter ɛ ∈ [ 1 0 − 7 , 1 0 − 4 ] , 99% of the computed distances have error less than the accuracy parameter. The features of predicable accuracy and computational efficiency distinguish DGG-VTP from the existing approximation methods, and make it an alternative to exact methods in computing accurate geodesic distances on large-scale mesh models. We also develop a parallel version of DGG-VTP on multi-core CPUs, which runs up to 60 × faster than the existing parallel VTP algorithm with comparable accuracy under single floating point precision setting. The source code is available at https://github.com/GeodesicGraph/DGG-VTP . • A new approximation method for computing geodesic distances and paths on triangle meshes. • Combining two SOTA methods, VTP and DGG, our method allows the user to directly specify the desired accuracy using a single parameter ɛ . • By setting the accuracy parameter ɛ = 1 0 − 6 , our method can produce results with comparable accuracy to VTP, while being 3–40 times faster and consuming as low as 1/30 memory.

A Two-Way Split-Step Wavelet Scheme for Tropospheric Long-Range Propagation in Various Environments

In the context of improving the dimensioning of observation and telecommunication, the characterization of the propagation canal is very important. Thus, accurate models of propagation phenomenona in their environment and above a rough surface (maritime or terrestrial) are of major interest for many applications (such as radar, communications, and teledetection). To provide solutions to this problem, in this paper, we propose a fast, memory-efficient, and accurate asymptotic method for 2D tropospheric propagation for a large band of frequency that accounts for relief, as well as ground composition and roughness. This latter is a two-way split-step wavelet scheme with an intrinsic stopping criterion. For overseas propagation, roughness effects are considered through a hybrid method. A complete theoretical comparison with SSF in terms of memory and time efficiency is proposed. Simulations in various environments (ground, sea, and snow), as well as different frequencies (UHF, S, and X-band) are performed to validate the method and highlight its advantages. To highlight the interest of the developed methodology, this latter is applied to different real-life applications, such as the prediction of radar coverage and the optimization of an antenna location.

Memory efficient constrained optimization of scanning-beam lithography.

This article describes a memory efficient method for solving large-scale optimization problems that arise when planning scanning-beam lithography processes. These processes require the identification of an exposure pattern that minimizes the difference between a desired and predicted output image, subject to constraints. The number of free variables is equal to the number of pixels, which can be on the order of millions or billions in practical applications. The proposed method splits the problem domain into a number of smaller overlapping subdomains with constrained boundary conditions, which are then solved sequentially using a constrained gradient search method (L-BFGS-B). Computational time is reduced by exploiting natural sparsity in the problem and employing the fast Fourier transform for efficient gradient calculation. When it comes to the trade-off between memory usage and computational time we can make a different trade-off compared to previous methods, where the required memory is reduced by approximately the number of subdomains at the cost of more computations. In an example problem with 30 million variables, the proposed method reduces memory requirements by 67% but increases computation time by 27%. Variations of the proposed method are expected to find applications in the planning of processes such as scanning laser lithography, scanning electron beam lithography, and focused ion beam deposition, for example.

Bank Locker Security System Using Machine Learning with Face and Liveness Detection

To increase reliability of face recognition system, the system must be able to distinguish real face from a copy of face such as a photograph. This system introduces a distinguished approach towards Machine learning and Neural Network, in which we propose a fast and memory efficient method of live face detection for embedded face recognition system, based on the analysis of the movement of the eyes. We detect eyes in sequential input images and calculate variation of each eye region to determine whether the input face is a real face or not.

A compound deep learning model for long range forecasting in electricity sale.

AbstractAccurate prediction of electricity sale has a positive effect on power companies in rationally arranging power supply plans, scientifically optimizing power resource allocation, improving power management efficiency, saving energy and reducing consumption. Predicting future electricity sale based on historical electricity sale data can essentially be summarized as a time series forecasting problem. This paper proposes a fast and memory-efficient method, which adopts the expressive power of deep neural networks and the time characteristics of sequence-to-sequence structure (parallel convolution and recurrent neural network) for long range forecasting in electricity sale. Through a large number of experiments and evaluation of real-world datasets, the effectiveness of the proposed method is proved and verified in terms of prediction accuracy, time consuming and training speed.

Memory-Efficient Modeling and Slicing of Large-Scale Adaptive Lattice Structures

Abstract Lattice structures have been widely used in various applications of additive manufacturing due to its superior physical properties. If modeled by triangular meshes, a lattice structure with huge number of struts would consume massive memory. This hinders the use of lattice structures in large-scale applications (e.g., to design the interior structure of a solid with spatially graded material properties). To solve this issue, we propose a memory-efficient method for the modeling and slicing of adaptive lattice structures. A lattice structure is represented by a weighted graph where the edge weights store the struts' radii. When slicing the structure, its solid model is locally evaluated through convolution surfaces and in a streaming manner. As such, only limited memory is needed to generate the toolpaths of fabrication. Also, the use of convolution surfaces leads to natural blending at intersections of struts, which can avoid the stress concentration at these regions. We also present a computational framework for optimizing supporting structures and adapting lattice structures with prescribed density distributions. The presented methods have been validated by a series of case studies with large number (up to 100M) of struts to demonstrate its applicability to large-scale lattice structures.

Deep learning compound trend prediction model for hydraulic turbine time series

AbstractAs a clean energy with mature technology, hydropower has been widely applied in industry. The hydraulic turbine unit plays an important role in hydropower station. Since the fault of turbine unit will affect the normal operation of the whole hydropower station, this paper proposes a universal, fast and memory-efficient method trend for time-series prediction of hydraulic turbines. The proposed method adopts the expressive power of deep neural networks and the time characteristics of sequence-to-sequence structure (parallel convolution and recurrent neural network) to make time-series prediction. It also uses convolutional quantile loss and memory network to predict future extreme events. The experimental results show that the proposed method is fast, robust and accurate. It can reduce at least 34% in mean square error and 33% in convergence speed compared with the existing methods.

Train Sparsely, Generate Densely: Memory-Efficient Unsupervised Training of High-Resolution Temporal GAN

Training of Generative Adversarial Network (GAN) on a video dataset is a challenge because of the sheer size of the dataset and the complexity of each observation. In general, the computational cost of training GAN scales exponentially with the resolution. In this study, we present a novel memory efficient method of unsupervised learning of high-resolution video dataset whose computational cost scales only linearly with the resolution. We achieve this by designing the generator model as a stack of small sub-generators and training the model in a specific way. We train each sub-generator with its own specific discriminator. At the time of the training, we introduce between each pair of consecutive sub-generators an auxiliary subsampling layer that reduces the frame-rate by a certain ratio. This procedure can allow each sub-generator to learn the distribution of the video at different levels of resolution. We also need only a few GPUs to train a highly complex generator that far outperforms the predecessor in terms of inception scores.

Quantum-Teleportation-Inspired Algorithm for Sampling Large Random Quantum Circuits

Quantum teleportation transfers and processes quantum information through quantum entanglement channels. It is one of the most versatile protocols in quantum information science and leads to many remarkable applications, particularly the one-way quantum computing. Here, we show, for the first time, that the concept of teleportation can also be used to facilitate an important classical computing task, sampling random quantum circuits, which is highly relevant to prove the near-term demonstration of quantum computational supremacy. In our method, the classical computation in the physical-qubit state space is converted to simulate teleportation in logical-qubit state space, resulting in a much smaller number of qubits involved in classical computing. We tested this new method on 1D and 2D lattices up to 1000 qubits. This Letter presents a new quantum-inspired classical computing technology and is helpful to design and optimize classically hard quantum sampling experiments.

Combustible Gas Classification Modeling using Support Vector Machine and Pairing Plot Scheme.

Combustible gases, such as CH4 and CO, directly or indirectly affect the human body. Thus, leakage detection of combustible gases is essential for various industrial sites and daily life. Many types of gas sensors are used to identify these combustible gases, but since gas sensors generally have low selectivity among gases, coupling issues often arise which adversely affect gas detection accuracy. To solve this problem, we built a decoupling algorithm with different gas sensors using a machine learning algorithm. Commercially available semiconductor sensors were employed to detect CH4 and CO, and then support vector machine (SVM) applied as a supervised learning algorithm for gas classification. We also introduced a pairing plot scheme to more effectively classify gas type. The proposed model classified CH4 and CO gases 100% correctly at all levels above the minimum concentration the gas sensors could detect. Consequently, SVM with pairing plot is a memory efficient and promising method for more accurate gas classification.

A memory efficient maximal clique enumeration method for sparse graphs with a parallel implementation

Maximal clique enumeration (MCE) is a widely studied problem that plays a crucial role in structure mining of undirected graphs. The increasing scale of real-world graphs has brought the challenges of high memory cost and high CPU workload to the problem. In this paper, we propose a memory efficient method named CMC-bit for MCE on sparse graphs. It reduces the memory cost via minimizing the candidate cliques and representing them by the data structure bitset. It generates an appropriate order for the vertex set according to two optimized principles to reduce the CPU cost. We further design a partition-based CMC-bit algorithm with a one-side extending strategy to solve the memory-limited problem. We parallelize the CMC-bit algorithm based on MapReduce with a range-based partition strategy to make an optimal trade-off between the shuffling workload of graph decomposition and load balance in the Reduce phase. We conduct extensive experiments on 30 real-world datasets. The results demonstrate that both the CMC-bit algorithm and its parallel implementation significantly outperform the respective state-of-the-art algorithms in speed. We also show that the parallel CMC-bit algorithm achieves good performance on the scalability with respect to both the reducer number and the CPU number.

Variable Resolution Occupancy Mapping Using Gaussian Mixture Models

Occupancy mapping is fundamental for active perception systems to enable reasoning about known and unknown regions of the environment. The majority of occupancy mapping approaches enforce an a priori discretization on the environment, resulting in a fixed resolution map that limits the expressiveness of the representation. The proposed approach removes this a priori discretization, learns continuous representations for the evidence of occupied and free space to derive the probability of occupancy, and enables occupancy grid maps to be generated at arbitrary resolution. Efficient methods are also presented that accurately evaluate the probability of occupancy in individual cells and enable multi-resolution mapping and local occupancy evaluation. The efficacy of the approach is demonstrated by comparison to state-of-the-art discrete and continuous mapping techniques in both two dimensions and three dimensions. The core contribution of this work is a memory-efficient method for deriving occupancy that is amenable to small or large corrections in pose without the need to regenerate the entire map. The applications under considerations are low-bandwidth scenarios (e.g., multi-robot exploration) and operations in expansive environments, where storing an occupancy grid map of the entire environment would be prohibitive.