Memory Savings Research Articles

Determination of the FIFO-stack size for the algorithm of the image region growing

The aim of the work is to derive an expression that allows determining the size of the FIFO-stack for storing the coordinates of adjacent pixels depending on the image size for the segmentation algorithm based on region growing. The FIFO stack, organized on the principle of a ring multi-bit shift register, is considered. The conditions for maximum loading of the FIFO stack are formulated, for which an expression is obtained that allows one to accurately determine the required size of the FIFO stack, which provides memory savings.

Compressed Neighbour Lists for SPH

AbstractWe propose a novel compression scheme to store neighbour lists for iterative solvers that employ Smoothed Particle Hydrodynamics (SPH). The compression scheme is inspired by Stream VByte, but uses a non‐linear mapping from data to data bytes, yielding memory savings of up to 87%. It is part of a novel variant of the Cell‐Linked‐List (CLL) concept that is inspired by compact hashing with an improved processing of the cell‐particle relations. We show that the resulting neighbour search outperforms compact hashing in terms of speed and memory consumption. Divergence‐Free SPH (DFSPH) scenarios with up to 1.3 billion SPH particles can be processed on a 24‐core PC using 172 GB of memory. Scenes with more than 7 billion SPH particles can be processed in a Message Passing Interface (MPI) environment with 112 cores and 880 GB of RAM. The neighbour search is also useful for interactive applications. A DFSPH simulation step for up to 0.2 million particles can be computed in less than 40 ms on a 12‐core PC.

A fast and memory-efficient implementation of the transfer bootstrap.

MotivationRecently, Lemoine et al. suggested the transfer bootstrap expectation (TBE) branch support metric as an alternative to classical phylogenetic bootstrap support for taxon-rich datasets. However, the original TBE implementation in the booster tool is compute- and memory-intensive.ResultsWe developed a fast and memory-efficient TBE implementation. We improve upon the original algorithm by Lemoine et al. via several algorithmic and technical optimizations. On empirical as well as on random tree sets with varying taxon counts, our implementation is up to 480 times faster than booster. Furthermore, it only requires memory that is linear in the number of taxa, which leads to 10× to 40× memory savings compared with booster.Availability and implementationOur implementation has been partially integrated into pll-modules and RAxML-NG and is available under the GNU Affero General Public License v3.0 at https://github.com/ddarriba/pll-modules and https://github.com/amkozlov/raxml-ng. The parallel version that also computes additional TBE-related statistics is available at: https://github.com/lutteropp/raxml-ng/tree/tbe.Supplementary information Supplementary data are available at Bioinformatics online.

Adaptive spectral solution method for the Landau and Lenard-Balescu equations

We present an adaptive spectral method for solving the Landau/Fokker-Planck equation for electron-ion systems. The heart of the algorithm is an expansion in Laguerre polynomials, which has several advantages, including automatic conservation of both energy and particles without the need for any special discretization or time-stepping schemes. One drawback of such an expansion is the O(N3) memory requirement, where N is the number of polynomials used. This can impose an inconvenient limit in cases of practical interest, such as when two particle species have widely separated temperatures. The algorithm we describe here addresses this problem by periodically re-projecting the solution onto a judicious choice of new basis functions that are still Laguerre polynomials but have arguments adapted to the current physical conditions. This results in a reduction in the number of polynomials needed, at the expense of increased solution time. Because the equations are solved with little difficulty, this added time is not of much concern compared to the savings in memory. To demonstrate the algorithm, we solve several relaxation problems that could not be computed with the spectral method without re-projection. Another major advantage of this method is that it can be used for collision operators more complicated than that of the Landau equation, and we demonstrate this here by using it to solve the non-degenerate quantum Lenard-Balescu (QLB) equation for a hydrogen plasma. We conclude with some comparisons of temperature relaxation problems solved with the latter equation and the Landau equation with a Coulomb logarithm inspired by the properties of the QLB operator. We find that with this choice of Coulomb logarithm, there is little difference between using the two equations for these particular systems.

Wideband spectrum sensing in cognitive radio using discrete wavelet packet transform and principal component analysis

The cognitive radio is a wireless technology which offers so much promise in the mitigation of the problem of spectrum scarcity. To achieve this noble objective, a cognitive radio must be characterized by speed and accuracy. In this paper, we present a spectrum sensing technique using discrete wavelet packet transform (DWPT) and principal component analysis (PCA) with convex optimization. An input signal was decomposed using DWPT to generate the required data matrix whose dimension was reduced using PCA and optimized using convex optimization. This technique was extensively compared with spectrum sensing using discrete Fourier transform (DFT), which is a popular wideband scheme. It was found that the DWPT scheme significantly outperformed DFT in spectrum detection accuracy. This was attributed to the fact that the time–frequency nature of wavelet transform makes it less lossy, i.e., it preserves information content much more than the DFT. The effect of PCA on the scheme was also evaluated and it was found that data dimensionality reduction was considerable, which impacts positively on the speed of spectrum detection. Data dimensionality reduction due to PCA also translates to memory and energy savings for the CR, since less energy is spent in processing signal data. It can then be said that this technique is also energy efficient, a desirable feature for mobile devices.

Architect: Arbitrary-Precision Hardware With Digit Elision for Efficient Iterative Compute

Many algorithms feature an iterative loop that converges to the result of interest. The numerical operations in such algorithms are generally implemented using finite-precision arithmetic, either fixed- or floating-point, most of which operate least-significant digit first. This results in a fundamental problem: if, after some time, the result has not converged, is this because we have not run the algorithm for enough iterations or because the arithmetic in some iterations was insufficiently precise? There is no easy way to answer this question, so users will often over-budget precision in the hope that the answer will always be to run for a few more iterations. We propose a fundamentally new approach: with the appropriate arithmetic able to generate results from most-significant digit first, we show that fixed compute-area hardware can be used to calculate an arbitrary number of algorithmic iterations to arbitrary precision, with both precision and approximant index increasing in lockstep. Consequently, datapaths constructed following our principles demonstrate efficiency over their traditional arithmetic equivalents where the latter’s precisions are either under- or over-budgeted for the computation of a result to a particular accuracy. Use of most-significant digit-first arithmetic additionally allows us to declare certain digits to be stable at runtime, avoiding their recalculation in subsequent iterations and thereby increasing performance and decreasing memory footprints. Versus arbitrary-precision iterative solvers without the optimizations we detail herein, we achieve up-to $16\times $ performance speedups and $1.9\times $ memory savings for the evaluated benchmarks.

Challenging the point neuron dogma: FS basket cells as 2-stage nonlinear integrators

Interneurons are critical for the proper functioning of neural circuits. While often morphologically complex, their dendrites have been ignored for decades, treating them as linear point neurons. Exciting new findings reveal complex, non-linear dendritic computations that call for a new theory of interneuron arithmetic. Using detailed biophysical models, we predict that dendrites of FS basket cells in both hippocampus and prefrontal cortex come in two flavors: supralinear, supporting local sodium spikes within large-volume branches and sublinear, in small-volume branches. Synaptic activation of varying sets of these dendrites leads to somatic firing variability that cannot be fully explained by the point neuron reduction. Instead, a 2-stage artificial neural network (ANN), with sub- and supralinear hidden nodes, captures most of the variance. Reduced neuronal circuit modeling suggest that this bi-modal, 2-stage integration in FS basket cells confers substantial resource savings in memory encoding as well as the linking of memories across time.

The choice of the sampling frequency and optimal method of signal digital processing in the problems considering random loading process for assessing durability

Theoretical and practical issues of processing digital information in the problems related to durability assessment in conditions of loading are considered. Due to the specifics of the problem, in which the precise dedermination of the extreme values and their sequence is of primary importance, a number of generally accepted recommendations appeared invalid. For example, the Kotelnikov theorem, which was originally proposed in relation to the problems of estimating the frequency composition of the process, can lead to significant errors. It is shown that the requirements to the analysis of random loading processes with a goal of further assessment of the durability, due to their specificity, contradict with the specified requirements, namely, when choosing a frequency according to this rule, an error can be made, and not to the margin of the strength assessment. We considered the issue regarding digital filtering of hardware overshoots. Alternative approaches to the selection of extrema of the random process are analyzed: 1) direct hardware selection of extrema and 2) discretization by the method of level crossings. The latter approach has an optimal algorithm for isolating extremes and makes it possible to isolate extremes of a random process with lower costs and greater accuracy. The natural transition to integer arithmetic provides further optimization of the algorithm. The model and real examples demonstrate gains in terms of speed and memory, which ultimately promote an increase in the reliability of the information required for assessing the durability. The savings in memory and performance will allow processing of long-term implementations and result in a more accurate estimating of the remaining life and durability at the stage of production.

Randomized Admission Policy for Efficient Top-k, Frequency, and Volume Estimation

Network management protocols often require timely and meaningful insight about per flow network traffic. This paper introduces Randomized Admission Policy (RAP) –a novel algorithm for the frequency , top-k , and byte volume estimation problems, which are fundamental in network monitoring. We demonstrate space reductions compared to the alternatives, for the frequency estimation problem, by a factor of up to 32 on real packet traces and up to 128 on heavy-tailed workloads. For top- $k$ identification, RAP exhibits memory savings by a factor of between 4 and 64 depending on the workloads’ skewness. These empirical results are backed by formal analysis, indicating the asymptotic space improvement of our probabilistic admission approach. In Addition, we present d-way RAP , a hardware friendly variant of RAP that empirically maintains its space and accuracy benefits.

A generalized massively parallel ultra-high order FFT-based Maxwell solver

Dispersion-free ultra-high order FFT-based Maxwell solvers have recently proven to be paramount to a large range of applications, including the high-fidelity modeling of high-intensity laser–matter interactions with Particle-In-Cell (PIC) codes. To enable a massively parallel scaling of these solvers, a novel parallelization technique was recently proposed, which consists in splitting the simulation domain into several processor sub-domains, with guard regions appended at each sub-domain boundary. Maxwell’s equations are advanced independently on each sub-domain using local shared-memory FFTs (instead of a single distributed global FFT). This implies small truncation errors at sub-domain boundaries, the amplitude of which depends on guard regions sizes and order of the Maxwell solver. For moderate guard region sizes, this ’local’ technique proved to be highly scalable on up to a million cores and notably enabled the 3D modelingof so-called plasma mirrors, for which 8 guard cells only were enough to prevent truncation error growth. Yet, for other applications, the required number of guard cells might be much higher, which would severely limit the parallel efficiency of this technique due to the large volume of guard cells to be exchanged between sub-domains. In this context, we propose a novel parallelization technique that ensures very good scaling of FFT-based solvers with an arbitrarily high number of guard cells. Our ’hybrid’ technique consists in performing distributed FFTs on local groups of processors with guard regions now appended to boundaries of each group of processors. It uses a dual domain decomposition method for the Maxwell solver and other parts of the PIC cycle to keep the simulation load-balanced. This ’hybrid’ technique was implemented in the open source exascale library PICSAR. Benchmarks show that for a large number of guard cells (>16), the ’hybrid’ technique offers up to ×3 speed-up and ×8 memory savings compared to the ’local’ one.

Deep Liquid State Machines With Neural Plasticity for Video Activity Recognition.

Real-world applications such as first-person video activity recognition require intelligent edge devices. However, size, weight, and power constraints of the embedded platforms cannot support resource intensive state-of-the-art algorithms. Machine learning lite algorithms, such as reservoir computing, with shallow 3-layer networks are computationally frugal as only the output layer is trained. By reducing network depth and plasticity, reservoir computing minimizes computational power and complexity, making the algorithms optimal for edge devices. However, as a trade-off for their frugal nature, reservoir computing sacrifices computational power compared to state-of-the-art methods. A good compromise between reservoir computing and fully supervised networks are the proposed deep-LSM networks. The deep-LSM is a deep spiking neural network which captures dynamic information over multiple time-scales with a combination of randomly connected layers and unsupervised layers. The deep-LSM processes the captured dynamic information through an attention modulated readout layer to perform classification. We demonstrate that the deep-LSM achieves an average of 84.78% accuracy on the DogCentric video activity recognition task, beating state-of-the-art. The deep-LSM also shows up to 91.13% memory savings and up to 91.55% reduction in synaptic operations when compared to similar recurrent neural network models. Based on these results we claim that the deep-LSM is capable of overcoming limitations of traditional reservoir computing, while maintaining the low computational cost associated with reservoir computing.

Abridged Symbolic Representation of Time Series for Clustering

In recent years a couple of methods aimed at time series symbolic representation have been introduced or developed. This activity is mainly justified by practical considerations such memory savings or fast data base searching. However, some results suggest that in the subject of time series clustering symbolic representation can even upgrade the results of clustering. The article contains a proposal of a new algorithm directed at the task of time series abridged symbolic representation with the emphasis on efficient time series clustering. The idea of the proposal is based on the PAA (piecewise aggregate approximation) technique followed by segmentwise correlation analysis. The primary goal of the article is to upgrade the quality of the PAA technique with respect to possible time series clustering (its speed and quality). We also tried to answer the following questions. Is the task of time series clustering in their original form reasonable? How much memory can we save using the new algorithm? The efficiency of the new algorithm was investigated on empirical time series data sets. The results prove that the new proposal is quite effective with a very limited amount of parametric user interference needed.

Quantization and Training of Low Bit-Width Convolutional Neural Networks for Object Detection

We present LBW-Net, an efficient optimization based method for quantization and training of the low bit-width convolutional neural networks (CNNs). Specifically, we quantize the weights to zero or powers of two by minimizing the Euclidean distance between full-precision weights and quantized weights during backpropagation. We characterize the combinatorial nature of the low bit-width quantization problem. For 2-bit (ternary) CNNs, the quantization of $N$ weights can be done by an exact formula in $O(N\log N)$ complexity. When the bit-width is three and above, we further propose a semi-analytical thresholding scheme with a single free parameter for quantization that is computationally inexpensive. The free parameter is further determined by network retraining and object detection tests. LBW-Net has several desirable advantages over full-precision CNNs, including considerable memory savings, energy efficiency, and faster deployment. Our experiments on PASCAL VOC dataset show that compared with its 32-bit floating-point counterpart, the performance of the 6-bit LBW-Net is nearly lossless in the object detection tasks, and can even do better in some real world visual scenes, while empirically enjoying more than 4$\times$ faster deployment.

GC-Wise: A Self-adaptive approach for memory-performance efficiency in Java VMs

Abstract High-level language runtimes are ubiquitous in every cloud deployment. From the geo-distributed heavy resources cloud provider to the new Fog and Edge deployment paradigms, all rely on these runtimes for portability, isolation and resource management. Across these clouds, an efficient resource management of several managed runtimes involves limiting the heap size of some VMs so that extra memory can be assigned to higher priority workloads. The challenges in this approach rely on the potential scale of systems and the need to make decisions in an application-driven way, because performance degradation can be severe, and therefore it should be minimized. Also, each tenant tends to repeat the execution of applications with similar memory-usage patterns, giving opportunity to reuse parameters known to work well for a given workload. This paper presents GC-Wise, a system to determine, at run-time, the best values for critical heap management parameters of the OpenJDK JVM, aiming to maximize memory-performance efficiency. GC-Wise comprises two main phases: 1) a training phase where it collects, with different heap resizing policies, representative execution metrics during the lifespan of a workload; and 2) an execution phase where an oracle matches the execution parameters of new workloads against those of already seen workloads, and enforces the best heap resizing policy. Distinctly from other works, the oracle can also decide upon unknown workloads. Using representative applications and different hardware setting (a resourceful server and a fog-like device), we show that our approach can lead to significant memory savings with low-impact on the throughput of applications. Furthermore, we show that we can predict with high accuracy the best heap resizing configuration in a relatively short period of time.

Phase chaotic encryption and efficiency evaluation for an image multiplexing method

We propose a new form of chaotic encryption of information contained in a multiplexed package, in which it has been possible to compress an amount of 1521 gray level images of human faces in a complex function, by using theta modulation and dynamic range optimization techniques. The phase is encrypted by using the confusion technique, through a chaotic mask (CM) that modify the phase values of the package, and then performing a linear transformation (LT) to store the encrypted phase in a commercial image format. The encryption key is composed of the generated seed values of the CM besides the scale factor of the LT. The cryptanalysis performed consists of a brute-force attack with a million iterations, besides evaluating the sensitivity of the system to slight variations of the encryption key, evidencing the strength of the method against this type of attacks. This work also shows advances in the efficiency achieved by the multiplexing method, regarding N number of multiplexed images; percentage of memory savings achieved, and the quality factors obtained when comparing the recovered information with the original one. All these parameters allow to compare our results with recent publications in this area.

Efficient Ab Initio Auxiliary-Field Quantum Monte Carlo Calculations in Gaussian Bases via Low-Rank Tensor Decomposition.

We describe an algorithm to reduce the cost of auxiliary-field quantum Monte Carlo (AFQMC) calculations for the electronic structure problem. The technique uses a nested low-rank factorization of the electron repulsion integral (ERI). While the cost of conventional AFQMC calculations in Gaussian bases scales as , where N is the size of the basis, we show that ground-state energies can be computed through tensor decomposition with reduced memory requirements and subquartic scaling. The algorithm is applied to hydrogen chains and square grids, water clusters, and hexagonal BN. In all cases, we observe significant memory savings and, for larger systems, reduced, subquartic simulation time.

Quasiparticle Levels at Large Interface Systems from Many-Body Perturbation Theory: The XAF-GW Method

We present a fully ab initio approach based on many-body perturbation theory in the GW approximation to compute the quasiparticle levels of large interface systems without significant covalent interactions between the different components of the interface (meaning that the different components can be separated without the creation of dangling bonds). The only assumption in our approach is that the polarizability matrix (chi) of the interface can be given by the sum of the polarizability matrices of individual components of the interface. We show analytically, using a two-state hybridized model, that this assumption is valid even in the presence of interface hybridization to form bonding and antibonding states up to first order in the overlap matrix elements involved in the hybridization. We validate our approach by showing that the band structure obtained in our method is almost identical to that obtained using a regular GW calculation for bilayer black phosphorus, where interlayer hybridization is significant. Significant savings in computational time and memory are obtained by computing chi only for the smallest subunit cell of each component and expanding (unfolding) the chi matrix to that in the unit cell of the interface. To treat interface hybridization, the full wave functions of the interface are used in computing the self-energy. We thus call the method XAF-GW (X, eXpand-chi; A, Add-chi; F, Full wave functions). Compared to GW-embedding type approaches in the literature, the XAF-GW approach is not limited to specific screening environments or to nonhybridized interface systems. XAF-GW can also be applied to systems with different dimensionalities, as well as to Moire superlattices such as in twisted bilayers. We illustrate the generality and usefulness of our approach by applying it to self-assembled PTCDA monolayers on Au(111) and Ag(111) and PTCDA monolayers on graphite-supported monolayer WSe2. In all cases, the predicted HOMO and LUMO levels agree well with experimental measurements.

EPCI: A new tool for predicting absolute permeability from computed tomography images

We have developed a new and fast MATLAB algorithm for predicting absolute permeability. The developed tool relies on measuring the connectivity of pores in a given 3D microcomputed tomography rock image. An index of pore connectivity is evaluated. After a calibration step, the developed index is used to estimate permeability in a variety of rocks with challenging pore structures (e.g., complex carbonate formations). The developed algorithm was tested on sandstone and carbonate rock samples. It offers large computational and memory savings when compared with algorithms based on the lattice Boltzmann method. Permeability estimates were, in general, in good agreement with laboratory measurements and numerical simulation results.

A Clustering Approach to Path Planning for Big Groups

The article introduces a new method of planning paths for big groups in dynamic environments represented by a graph of vertices and edges, where the edge weight as well as the graph topology may change, but the method is also applicable to environments with a different representation. The utilization of clustering enables the use of a computed path for a group of agents. In this way, a speed-up and memory savings are achieved at a cost of some path sub-optimality. Examples of proper applications of the suggested approach are crowd simulation in urban environments or path-planning-based tasks in molecular biology. The experiments showed good behaviour of the method to speed-up, relative error and online computation.

Performance Analysis of Compression Techniques for Chronic Wound Image Transmission Under Smartphone-Enabled Tele-Wound Network

The healing status of chronic wounds is important for monitoring the condition of the wounds. This article designs and discusses the implementation of smartphone-enabled compression technique under a tele-wound network (TWN) system. Nowadays, there is a huge demand for memory and bandwidth savings for clinical data processing. Wound images are captured using a smartphone through a metadata application page. Then, they are compressed and sent to the telemedical hub with a set partitioning in hierarchical tree (SPIHT) compression algorithm. The transmitted image can then be reduced, followed by an improvement in the segmentation accuracy and sensitivity. Better wound healing treatment depends on segmentation and classification accuracy. The proposed framework is evaluated in terms of rates (bits per pixel), compression ratio, peak signal to noise ratio, transmission time, mean square error and diagnostic quality under telemedicine framework. A SPIHT compression technique assisted YDbDr-Fuzzy c-means clustering considerably reduces the execution time (105s), is simple to implement, saves memory (18 KB), improves segmentation accuracy (98.39%), and yields better results than the same without using SPIHT. The results favor the possibility of developing a practical smartphone-enabled telemedicine system and show the potential for being implemented in the field of clinical evaluation and the management of chronic wounds in the future.