Memory Efficiency Research Articles

Marker discovery in the large.

Markers for diagnostic polymerase chain reactions are routinely constructed by taking regions common to the genomes of a target organism and subtracting the regions found in the targets' closest relatives, their neighbors. This approach is implemented in the published package Fur, which originally required memory proportional to the number of nucleotides in the neighborhood. This does not scale well. Here, we describe a new version of Fur that only requires memory proportional to the longest neighbor. In spite of its greater memory efficiency, the new Fur remains fast and is accurate. We demonstrate this by applying it to simulated sequences and comparing it to an efficient alternative. Then we use the new Fur to extract markers from 120 reference bacteria. To make this feasible, we also introduce software for automatically finding target and neighbor genomes and for assessing markers. We pick the best primers from the 10 most sequenced reference bacteria and show their excellent in silico sensitivity and specificity. Fur is available from github.com/evolbioinf/fur, in the Docker image hub.docker.com/r/beatrizvm/mapro, and in the Code Ocean capsule 10.24433/CO.7955947.v1.

Hydrogen assisted cracking using an efficient virtual element scheme

In this paper, a novel virtual element formulation is proposed for phase field modeling of hydrogen assisted cracking (HAC). The multiphysics variational framework is decoupled into three parts which can be solved in a staggered manner, namely, the mechanical, diffusion, and damage sub-problems. The damage and diffusion sub-problems are treated as reaction–diffusion types of equations. The former is subjected to irreversibility constraints while the latter incorporates a transport term to represent the effect of mechanical loading. An efficient super-convergent patch recovery (SPR) scheme is utilized to determine the coefficient of transport term in the hydrogen diffusion equation. Several qualitative and quantitative benchmark problems are considered to validate the proposed framework’s performance. The results show excellent agreement with the corresponding FEM and experimental studies. Meanwhile, without loss of numerical accuracy, the VEM outperforms the FEM with regards to CPU time & memory efficiency and exhibits remarkable versatility in the selection of different element types for simulation. Moreover, this work demonstrates the potential of facilitating the VEM to solve real-scale fracture mechanics problems with multi-physics coupling.

LM-SRPQ: Efficiently Answering Regular Path Query in Streaming Graphs

Regular path query (RPQ) is a basic operation for graph data analysis, and persistent RPQ in streaming graphs is a new-emerging research topic. In this paper, we propose a novel algorithm for persistent RPQ in streaming graphs, named LM-SRPQ. It solves persistent RPQ with a combination of intermediate result materialization and real-time graph traversal. Compared to prior art, it merges redundant storage and computation, achieving higher memory and time efficiency. We carry out extensive experiments with both real-world and synthetic streaming graphs to evaluate its performance. Experiment results confirm its superiority compared to prior art in both memory and time efficiency.

GLULA: Linear attention-based model for efficient human activity recognition from wearable sensors.

Body-worn sensor data is used in monitoring patient activity during rehabilitation and also can be extended to controlling rehabilitation devices based on the activity of the person. The primary focus of research has been on effectively capturing the spatiotemporal dependencies in the data collected by these sensors and efficiently classifying human activities. With the increasing complexity and size of models, there is a growing emphasis on optimizing their efficiency in terms of memory usage and inference time for real-time usage and mobile computers. While hybrid models combining convolutional and recurrent neural networks have shown strong performance compared to traditional approaches, self-attention-based networks have demonstrated even superior results. However, instead of relying on the same transformer architecture, there is an opportunity to develop a novel framework that incorporates recent advancements to enhance speed and memory efficiency, specifically tailored for human activity recognition (HAR) tasks. In line with this approach, we present GLULA, a unique architecture for HAR. GLULA combines gated convolutional networks, branched convolutions, and linear self-attention to achieve efficient and powerful solutions. To enhance the performance of our proposed architecture, we employed manifold mixup as an augmentation variant which proved beneficial in limited data settings. Extensive experiments were conducted on five benchmark datasets: PAMAP2, SKODA, OPPORTUNITY, DAPHNET, and USC-HAD. Our findings demonstrate that GLULA outperforms recent models in the literature on the latter four datasets but also exhibits the lowest parameter count and close to the fastest inference time among state-of-the-art models.

Digit-Serial DA-Based Fixed-Point RNNs: A Unified Approach for Enhancing Architectural Efficiency.

The next crucial step in artificial intelligence involves integrating neural network models into embedded and mobile systems. This requires designing compact and energy-efficient neural network models in silicon for optimized performance. This article introduces a unified approach for enhancing the architectural efficiency of long short-term memory (LSTM) recurrent neural networks (RNNs). Precisely, two new structures (I and II) based on the two's complement (TC) digit-serial distributed arithmetic (DSDA) technique are presented. The block-circulant matrix-vector multiplications (MVMs) and element-wise multiplications (EWMs) are formulated using TC DSDA. In addition, a fixed-point (FxP) training procedure for quantized LSTM RNNs is considered and validated for speech recognition tasks. Both structures leverage the circular rotation of weights and generate partial products with input digit slices. A new partial-product generator (PPG) and partial-product selector (PPS) designed to work with both unsigned and signed digits is introduced. In Structure I, a nonpipelined MVM is realized with a few PPGs and PPSs, followed by a shift-accumulate unit (SAU). Conversely, in Structure II, a suitably chosen depth-pipelined MVM is achieved with multiple PPGs and PPSs, followed by a shift-to-add tree (SAT). A critical path delay (CPD) analysis for both the proposed structures is also presented. Compared with previous works, post-synthesis results on 28 -nm fully depleted silicon-on-insulator (FDSOI) technology reveal that for a model size of 128 × 128 , Structures I and II provide 39.87% , 95.63% , and 30.95% , 91.18% more area and energy efficiencies, respectively.

A Non-parametric Bootstrap Method for Kinetic Monte Carlo Variance Reduction

A new variance reduction technique for Monte Carlo transport methods is investigated. This approach is based on non-parametric bootstrapping, a statistical inference and resampling method which is used to generate simulated samples of Monte Carlo scores to estimate the statistical properties of underlying integral response distributions. It is also used to quantify the uncertainty of the corresponding estimates as well as any bias introduced by bootstrapping, which then can be corrected. In this context, non-parametric bootstrapping is used to estimate the first and second-order moments of estimations of Monte Carlo responses for time-dependent neutron transport problems, which are particularly time-consuming. This variance reduction technique was implemented in SCONE - Stochastic Calculator Of Neutron transport Equation, developed at the University of Cambridge. Results show that the non-parametric bootstrap method can improve the trade-off between simulation time and variance of integral response estimates, especially when the number of responses is kept to a moderate level. As the number of responses increases, however, challenges with bias, memory and computational efficiency become more prominent, which is addressed.

Dynamic Recognition of Speakers for Consent Management by Contrastive Embedding Replay.

Voice assistants overhear conversations, and a consent management mechanism is required. Consent management can be implemented using speaker recognition. Users that do not give consent enroll their voice, and all their further recordings are discarded. Building speaker recognition-based consent management is challenging as dynamic registration, removal, and reregistration of speakers must be efficiently handled. This work proposes a consent management system addressing the aforementioned challenges. A contrastive-based training is applied to learn the underlying speaker equivariance inductive bias. The contrastive features for buckets of speakers are trained a few steps into each iteration and act as replay buffers. These features are progressively selected using a multi-strided random sampler for classification. Moreover, new methods for dynamic registration using a portion of old utterances, removal, and reregistration of speakers are proposed. The results verify memory efficiency and dynamic capabilities of the proposed methods and outperform the existing approaches from the literature in terms of convergence rate and number of required parameters.

Validation of Smagorinsky LES turbulence model in FluidX3D LBM: In-place vs central difference

Lattice Boltzmann method (LBM) has been rapidly developing as a CFD approach throughout the last three decades. The rise of new computational architectures like GPU in the last decade pawed the way to emergence of high-performance CFD software based on this method. The most outstanding of such is FluidX3D, utilizing locality of algorithm operations in connection with massive parallelism of GPU. High performance and memory efficiency of this software make it a promising idea to apply FluidX3D to challenging CFD problems, like turbulence simulation. In this work an effort is put into validation of Smagorinsky LES turbulence model implemented in this software. In addition, direct central difference implementation of this model is added to verify existing approach. The validation is done on the turbulent plane channel flow at Reτ = 180. Reasonable agreement between two LES model results was obtained. The convergence of LES profiles to DNS profiles with sufficient resolution proves correctness of implemented model.

Innovative Dual-Decoupling CNN with Layer-wise Temporal-Spatial Attention for Sensor-Based Human Activity Recognition.

Human Activity Recognition (HAR) is essential for monitoring and analyzing human behavior, particularly in health applications such as fall detection and chronic disease management. Traditional methods, even those incorporating attention mechanisms, often oversimplify the complex temporal and spatial dependencies in sensor data by processing features uniformly, leading to inadequate modeling of high-dimensional interactions. To address these limitations, we propose a novel framework: the Temporal-Spatial Feature Decoupling Unit with Layer-wise Training Convolutional Neural Network (CNN-TSFDU-LW). Our model enhances HAR accuracy by decoupling temporal and spatial dependencies, facilitating more precise feature extraction and reducing computational overhead. The TSFDU mechanism enables parallel processing of temporal and spatial features, thereby enriching the learned representations. Furthermore, layer-wise training with a local error function allows for independent updates of each CNN layer, reducing the number of parameters and improving memory efficiency without compromising performance. Experiments on four benchmark datasets (UCI-HAR, PAMAP2, UNIMIB-SHAR, and USC-HAD) demonstrate accuracy improvements ranging from 0.9% to 4.19% over state-of-the-art methods while simultaneously reducing computational complexity. Specifically, our framework achieves accuracy rates of 97.90% on UCI-HAR, 94.34% on PAMAP2, 78.90% on UNIMIB-SHAR, and 94.71% on USC-HAD, underscoring its effectiveness in complex HAR tasks. In conclusion, the CNN-TSFDU-LW framework represents a significant advancement in sensor-based HAR, delivering both improved accuracy and computational efficiency, with promising potential for enhancing health monitoring applications.

Sleep Disruption, Fatigue, and Altered Neurobehavioral Performance Among Flight Controllers During Spaceflight Operations

In spaceflight operations, flight controllers manage technical aspects of spaceflight control and spacecraft systems. The flight control team is a group subjected to shift work. Acute and chronic exposure to shift work has been associated with circadian misalignment, sleep impairment, and a negative impact on cognitive performance. This study aims to review the effects of shift work on sleep, circadian rhythms, mood, and cognitive performance of flight controllers during real and simulated spaceflight operations. Shift work during low-Earth orbit spaceflight missions is associated with a reduction in alertness, motivation, processing speed, and working memory performance efficiency. Flight controllers also report excessive insomnia and insufficient total sleep time. The development of shift work sleep disorder may be present in up to 40% of workers, especially among night and evening shift workers. Mars operations and Mars-simulated missions are associated with an impairment of visual-motor performance, working memory efficiency, and reaction time. Between 50%–87% of controllers can synchronize their circadian rhythm to a Mars day (24.65 hours), although this adaptation is not reflected in improved neurobehavioral performance.

A 3D finite element spectral integral (FESI) method for acoustics

We present a novel hybrid method that integrates the spectral integral method (SIM) with finite element for pressure acoustics in an inhomogeneous medium surrounded by spherical or incomplete spherical surfaces. Our approach, called the finite element spectral integral (FESI) method, enables efficient calculations of the pressure field in near-field and far-field zones. It is more efficient than the conventional finite element method hybridized with the boundary element method (i.e., the FEBI method). Furthermore, we construct a symmetric coefficient matrix using the proposed symmetric coupling to speed up the FESI method. We demonstrate through several examples that our FESI method offers significantly improved efficiency and lower memory consumption compared to the FEM combined with the perfect matching layer method while achieving higher accuracy.

An efficient weighted partial MaxSAT encoding for scheduling in overloaded real-time systems

Scheduling tasks in overloaded real-time systems is a challenging problem that has received a significant amount of attention in recent years. The processor is overloaded with more tasks than its capacity, resulting in missed deadlines and degraded system performance. Therefore, scheduling algorithms play a critical role in ensuring that high-priority tasks are completed on time while minimizing the impact of incomplete lower-priority tasks. This paper proposes an efficient Weighted Partial MaxSAT(WPMS) encoding that returns an optimal solution in which the total weight of incomplete tasks is minimized in a single machine environment. To assess the efficiency of our proposed formulation, a comparative analysis is conducted alongside the state-of-the-art encoding. By examining the solving 75 distinct problems, it becomes evident that the WPMS encoding proposed herein exhibits a considerable advantage in terms of both time and memory efficiency.

An Online Support Vector Machine Algorithm for Dynamic Social Network Monitoring

Online monitoring of social networks offers exciting features for platforms, enabling both technical and behavioral analysis. Numerous studies have explored the adaptation of traditional quality control methods for detecting change points within social networks. However, the current research studies face limitations such as an overreliance on case-based attributes, high computational costs, poor scalability with large networks, and low sensitivity in fast change point detection. This paper proposes a novel algorithm for social network monitoring using One-Class Support Vector Machines (OC-SVMs) to address these limitations. Additionally, using both nodal and network-level attributes makes it versatile for diverse social network applications and effectively detecting network disturbances. The algorithm utilizes a well-defined training data dictionary with an updating procedure for evolutionary networks, enhancing memory and time efficiency by reducing the processing of input data. Extensive numerical experiments are conducted using an EpiCNet model to simulate interactions in an online social network, covering six change scenarios to evaluate the proposed methodology. The results show lower Average Run Length (ARL) and Expected Delay Detection (EDD), demonstrating the superior accuracy and effectiveness of the OC-SVM algorithm compared to alternative methods. Applying OC-SVM to the Enron Email network indicates its capability to identify change points, reflecting the tumultuous timeline that led to Enron's downfall. This further validates the substantial advancement of OC-SVM in social network monitoring and opens doors to broader real-world applications.

Optimal Configuration of Multi-Task Learning for Autonomous Driving.

For autonomous driving, it is imperative to perform various high-computation image recognition tasks with high accuracy, utilizing diverse sensors to perceive the surrounding environment. Specifically, cameras are used to perform lane detection, object detection, and segmentation, and, in the absence of lidar, tasks extend to inferring 3D information through depth estimation, 3D object detection, 3D reconstruction, and SLAM. However, accurately processing all these image recognition operations in real-time for autonomous driving under constrained hardware conditions is practically unfeasible. In this study, considering the characteristics of image recognition tasks performed by these sensors and the given hardware conditions, we investigated MTL (multi-task learning), which enables parallel execution of various image recognition tasks to maximize their processing speed, accuracy, and memory efficiency. Particularly, this study analyzes the combinations of image recognition tasks for autonomous driving and proposes the MDO (multi-task decision and optimization) algorithm, consisting of three steps, as a means for optimization. In the initial step, a MTS (multi-task set) is selected to minimize overall latency while meeting minimum accuracy requirements. Subsequently, additional training of the shared backbone and individual subnets is conducted to enhance accuracy with the predefined MTS. Finally, both the shared backbone and each subnet undergo compression while maintaining the already secured accuracy and latency performance. The experimental results indicate that integrated accuracy performance is critically important in the configuration and optimization of MTL, and this integrated accuracy is determined by the ITC (inter-task correlation). The MDO algorithm was designed to consider these characteristics and construct multi-task sets with tasks that exhibit high ITC. Furthermore, the implementation of the proposed MDO algorithm, coupled with additional SSL (semi-supervised learning) based training, resulted in a significant performance enhancement. This advancement manifested as approximately a 12% increase in object detection mAP performance, a 15% improvement in lane detection accuracy, and a 27% reduction in latency, surpassing the results of previous three-task learning techniques like YOLOP and HybridNet.

Fast and Accurate Parasitic Extraction in Multichip Power Module Design Automation Considering Eddy-Current Losses

Recent studies in power electronic design automation have introduced various models for parasitic extraction of multichip power module layouts. However, none of these studies consider the eddy current effect in the direct-bonded-copper substrate, accounting for 40-50% error in the extraction result. This work introduces a methodology for eddy-current consideration through numerical simulation and regression modeling. The regression model utilized in the characterization process in this work is fast and memory-efficient compared to the finite element approach. This characterization process can improve the accuracy of any partial element model without sacrificing performance. Combining this characterization process and partial element approach achieves less than 10% extraction error compared to Ansys Q3D while showing a maximum speed-up of 35× and 17× more memory efficiency. This method also significantly reduces the number of elements in the extracted netlist and the complexity of loop evaluation. This method is attractive for use with optimization routines and therefore has been used successfully in a layout optimization tool.

Enhancing Efficiency of the Fast Quantum Memory on Single-Atom in Cavity

Enhancing Efficiency of the Fast Quantum Memory on Single-Atom in Cavity

Advanced Pattern-Mining System for Fake News Analysis

Decomposition MapReduce mining for fake news analysis (DMRM-FNA), a novel generic parallel pattern-mining framework, is developed in this article to solve difficulties in social network analysis using big data exploration. The first difficulty faced by existing techniques is the inability to retrieve actionable insights into the structure of fake news data. This can be solved by extracting patterns from fake news and matching them with real news data. The second difficulty is the computational time of existing pattern-mining solutions. This might be solved by combining both decomposition and MapReduce mining techniques to extract relevant patterns from fake news data. The multiobjective <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">k</i> -means algorithm is used to first aggregate fake news data. To generate useful patterns, a parallel pattern-mining method based on MapReduce structures is applied. To evaluate the created DMRM-FNA framework (DFAST) and demonstrate the high performance of sequential pattern-mining challenges on massive social network data, several tests were conducted. Our results show that the proposed DMRM-FNA performs well in terms of memory usage and efficiency. Moreover, DMRM-FNA outperforms existing models in terms of accuracy and feasibility.

An empirical study of schema-associated mnemonic method for classical Chinese poetry in primary and secondary education based on cognitive schema migration theory

Classical Chinese poetry, as an indispensable part of China′s national culture, provides valuable resources for improving literary literacy and fostering patriotism among primary and secondary school students. This study, based on cognitive schema migration theory, investigated the effect of a "schema-associated mnemonic method" on the memory efficiency and retention of classical Chinese poetry memorization among the students. Through a controlled experiment, the results revealed that the experimental group (n = 63) outperformed the control one in memorizing classical poetry with higher memory efficiency and retention, which indicates the sustainable application prospects of the method in primary and secondary education. The findings of this research can provide beneficial insights for the effective integration of classical Chinese poetry teaching and educational technologies.

On the Caching Schemes to Speed Up Program Reduction

Program reduction is a highly practical, widely demanded technique to help debug language tools, such as compilers, interpreters and debuggers. Given a program P that exhibits a property ψ, conceptually, program reduction iteratively applies various program transformations to generate a vast number of variants from P by deleting certain tokens and returns the minimal variant preserving ψ as the result. A program reduction process inevitably generates duplicate variants, and the number of them can be significant. Our study reveals that on average 61.8% and 24.3% of the generated variants in two representative program reducers HDD and Perses, respectively, are duplicates. Checking them against ψ is thus redundant and unnecessary, which wastes time and computation resources. Although it seems that simply caching the generated variants can avoid redundant property tests, such a trivial method is impractical in the real world due to the significant memory footprint. Therefore, a memory-efficient caching scheme for program reduction is in great demand. This study is the first effort to conduct a systematic, extensive analysis of memory-efficient caching schemes for program reduction. We first propose to use two well-known compression methods, ZIP and SHA , to compress the generated variants before they are stored in the cache. Furthermore, our keen understanding on the program reduction process motivates us to propose a novel, domain-specific, both memory and computation-efficient caching scheme, R efreshable C ompact C aching ( RCC ). Our key insight is two-fold: ① by leveraging the correlation between variants and the original program P , we losslessly encode each variant into an equivalent , compact , canonical representation; ② periodically, stale cache entries, which will never be accessed, are timely removed to minimize the memory footprint over time. Our extensive evaluation on 31 real-world C compiler bugs demonstrates that caching schemes help avoid issuing redundant queries by 61.8% and 24.3% in HDD and Perses, respectively; correspondingly, the runtime performance is notably boosted by 22.8% and 18.2%. With regard to the memory efficiency, all three methods use less memory than the state-of-the-art string-based scheme STR . Specifically, ZIP and SHA cut down the memory footprint by more than 80% and 90% in both Perses and HDD compared to STR ; moreover, the highly-scalable, domain-specific RCC dominates peer schemes, and outperforms the SHA by 96.4% and 91.74% in HDD and Perses, respectively.

An Efficient and Light Transformer-Based Segmentation Network for Remote Sensing Images of Landscapes

High-resolution image segmentation for landscape applications has garnered significant attention, particularly in the context of ultra-high-resolution (UHR) imagery. Current segmentation methodologies partition UHR images into standard patches for multiscale local segmentation and hierarchical reasoning. This creates a pressing dilemma, where the trade-off between memory efficiency and segmentation quality becomes increasingly evident. This paper introduces the Multilevel Contexts Weighted Coupling Transformer (WCTNet) for UHR segmentation. This framework comprises the Mult-level Feature Weighting (MFW) module and Token-based Transformer (TT) designed to weigh and couple multilevel semantic contexts. First, we analyze the multilevel semantics within a local patch without image-level contextual reasoning. It avoids complex image-level contextual associations and eliminates the misleading information carried. Second, MFW is developed to weigh shallow and deep features for enhancing object-related attention at different grain sizes from multilevel semantics. Third, the TT module is introduced to couple multilevel semantic contexts and transform them into semantic tokens using spatial attention. Then, we can capture token interactions and obtain clearer local representations. The suggested contextual weighting and coupling of single-scale patches empower WCTNet to maintain a well-balanced relationship between accuracy and computational overhead. Experimental results show that WCTNet achieves state-of-the-art performance on two UHR datasets of DeepGlobe and Inria Aerial.