Maximum Memory Research Articles

RaNAS: Resource-Aware Neural Architecture Search for Edge Computing

Neural architecture search (NAS) for edge devices is often time-consuming because of long-latency deploying and testing on edge devices. The ability to accurately predict the computation cost and memory requirement for convolutional neural networks (CNNs) in advance holds substantial value. Existing work primarily relies on analytical models, which can result in high prediction errors. This paper proposes a resource-aware NAS (RaNAS) model based on various features. Additionally, a new graph neural network is introduced to predict inference latency and maximum memory requirements for CNNs on edge devices. Experimental results show that, within the error bound of ±1%, RaNAS achieves an accuracy improvement of approximately 8% for inference latency prediction and about 25% for maximum memory occupancy prediction over the state-of-the-art approaches.

Enhancing Charge Trapping Performance of Hafnia Thin Films Using Sequential Plasma Atomic Layer Deposition.

We aimed to fabricate reliable memory devices using HfO2, which is gaining attention as a charge-trapping layer material for next-generation NAND flash memory. To this end, a new atomic layer deposition process using sequential remote plasma (RP) and direct plasma (DP) was designed to create charge-trapping memory devices. Subsequently, the operational characteristics of the devices were analyzed based on the thickness ratio of thin films deposited using the sequential RP and DP processes. As the thickness of the initially RP-deposited thin film increased, the memory window and retention also increased, while the interface defect density and leakage current decreased. When the thickness of the RP-deposited thin film was 7 nm, a maximum memory window of 10.1 V was achieved at an operating voltage of ±10 V, and the interface trap density (Dit) reached a minimum value of 1.0 × 1012 eV-1cm-2. Once the RP-deposited thin film reaches a certain thickness, the ion bombardment effect from DP on the substrate is expected to decrease, improving the Si/SiO2/HfO2 interface and thereby enhancing device endurance and reliability. This study confirmed that the proposed sequential RP and DP deposition processes could resolve issues related to unstable interface layers, improve device performance, and enhance process throughput.

Neighbor patches merging reduces spatial redundancy to accelerate vision transformer

Vision Transformers (ViTs) deliver outstanding performance but often require substantial computational resources. Various token pruning methods have been developed to enhance throughput by removing redundant tokens; however, these methods do not address the peak memory consumption, which remains equivalent to that of the unpruned networks. In this study, we introduce Neighbor Patches Merging (NEPAM), a method that significantly reduces the maximum memory footprint of ViTs while pruning tokens. NEPAM targets spatial redundancy within images and prunes redundant patches at the onset of the model, thereby achieving the optimal throughput-accuracy trade-off without fine-tuning. Experimental results demonstrate that NEPAM can accelerate the inference speed of the Vit-Base-Patch16-384 model by 25% with a negligible accuracy loss of 0.07% and a notable 18% reduction in memory usage. When applied to VideoMAE, NEPAM doubles the throughput with a 0.29% accuracy loss and a 48% reduction in memory usage. These findings underscore the efficacy of NEPAM in mitigating computational requirements while maintaining model performance.

DNA sequences alignment method using sparse index on pan-genome graph.

The graph of sequences represents the genetic variations of pan-genome concisely and space-efficiently than multiple linear reference genome. In order to accelerate aligning reads to the graph, an index of graph-based reference genomes is used to obtain candidate locations. However, the potential combinatorial explosion of nodes on the sequence graph leads to increasing the index space and maximum memory usage of alignment process considerably, especially for large-scale datasets. For this, existing methods typically attempt to prune complex regions, or extend the length of seeds, which sacrifices the recall of alignment algorithm despite reducing space usage slightly. We present the Sparse-index of Graph (SIG) and alignment algorithm SIG-Aligner, capable of indexing and aligning at the lower memory cost. SIG builds the non-overlapping minimizers index inside nodes of sequence graph and SIG-Aligner filters out most of the false positive matches by the method based on the pigeonhole principle. Compared to Giraffe, the results of computational experiments show that SIG achieves a significant reduction in index memory space ranging from 50% to 75% for the human pan-genome graphs, while still preserving superior or comparable accuracy of alignment and the faster alignment time.

Strategic Employee Performance Analysis in the USA: Deploying Machine Learning Algorithms Intelligently

Strategic employee performance assessment assists organizations in steering productivity, affirming employee satisfaction, and accomplishing strategic organizational goals. Machine learning algorithms provide several benefits over mainstream techniques in assessing employee performance. This research paper aimed to explore the deployment of machine learning algorithms in assessing employee performance. The prime objective of employee performance analysis is to assess an employee's achievement during a specific time frame. The dataset for this research revolved around the leadership team of a global retailer's specific store level in the USA, extending over 18 months. The dataset for this study was subjected to Python programming software for intensive and comprehensive data analysis as well as for visualization purposes. From the experiment design, it was evident that XG-Boost seems to be the best-performing model overall. In particular, it had the greatest AUC for both holdout and training data (0.86 and 0.88, respectively), and it has a relatively low runtime (16 minutes) and maximum memory utilization (12%). By contrast, Random Forest displayed an average AUC for training data (0.79) but a lesser AUC for holdout data (0.51), which indicates that it may be overfitting the training data; besides, it had a longer runtime than XG-Boost.

A numerical method for two-dimensional transient nonlinear convection-diffusion equations

In this research, we have developed the half-boundary method (HBM) for nonlinear convection-diffusion equations (CDEs), which hold significant importance in nuclear power engineering. The HBM employs a variable relationship between the nodes within the computational domain and the nodes located on half of the boundaries. This approach offers notable benefits, includingthe reduction ofthe maximum matrix order and the optimization the maximum memory storage for calculations. Moreover, the HBM is an efficient and streamlined approach to directly handle Neumann boundary conditions, thanks to the utilization of mixed variables. We primarily investigate the memory usage and accuracy of the proposed algorithm in the unsteady-state CDEs, in context of material nonlinear CDEs, the Burgers’ equation and the system of Burgers' equations. The numerical results obtained demonstrate the method’s potential in simulating flow and heat transfer phenomena, particularly in situations where convection is dominant.

PQSDC: a parallel lossless compressor for quality scores data via sequences partition and run-length prediction mapping.

The quality scores data (QSD) account for 70% in compressed FastQ files obtained from the short and long reads sequencing technologies. Designing effective compressors for QSD that counterbalance compression ratio, time cost, and memory consumption is essential in scenarios such as large-scale genomics data sharing and long-term data backup. This study presents a novel parallel lossless QSD-dedicated compression algorithm named PQSDC, which fulfills the above requirements well. PQSDC is based on two core components: a parallel sequences-partition model designed to reduce peak memory consumption and time cost during compression and decompression processes, as well as a parallel four-level run-length prediction mapping model to enhance compression ratio. Besides, the PQSDC algorithm is also designed to be highly concurrent using multicore CPU clusters. We evaluate PQSDC and four state-of-the-art compression algorithms on 27 real-world datasets, including 61.857 billion QSD characters and 632.908 million QSD sequences. (1) For short reads, compared to baselines, the maximum improvement of PQSDC reaches 7.06% in average compression ratio, and 8.01% in weighted average compression ratio. During compression and decompression, the maximum total time savings of PQSDC are 79.96% and 84.56%, respectively; the maximum average memory savings are 68.34% and 77.63%, respectively. (2) For long reads, the maximum improvement of PQSDC reaches 12.51% and 13.42% in average and weighted average compression ratio, respectively. The maximum total time savings during compression and decompression are 53.51% and 72.53%, respectively; the maximum average memory savings are 19.44% and 17.42%, respectively. (3) Furthermore, PQSDC ranks second in compression robustness among the tested algorithms, indicating that it is less affected by the probability distribution of the QSD collections. Overall, our work provides a promising solution for QSD parallel compression, which balances storage cost, time consumption, and memory occupation primely. The proposed PQSDC compressor can be downloaded from https://github.com/fahaihi/PQSDC.

Impact of channel thickness on device scaling in vertical InGaZnO channel charge-trap memory transistors with ALD Al2O3 tunneling layer

This study investigates the impact of channel thickness (TCH) variation on memory performance and its physical origins in vertical channel charge trap memory (V-CTM) using InGaZnO (IGZO) channels for X–Y scaling. When the TCH decreased, the subthreshold slope (SS) increased from 0.21 to 0.30 V/dec, while the minimum program voltage decreased from 12 to 8 V, indicating a performance trade-off. The analysis of bulk trap density extracted by SS, density of states distribution, and x-ray photoelectron spectroscopy confirmed an increase in the number of trap states and hydrogen-related defects in the IGZO channel when a thin TCH is employed. As a result, it was expected that the barrier height at the active layer/TL would be lowered during program operations at relatively thin channel. However, the SS would also be degraded due to an increase in sub-gap states, such as OH- and VOH, induced by hydrogen incorporation. The V-CTM with an optimal TCH of 7 nm had a maximum memory window of 5.7 V. The program/erase states were maintained for 106 s before experiencing a 50 % charge loss, and 5-level memory states were verified to be available within a memory window of 3.2 V for 108 s with little degradation. It was noteworthy that the long-term reliability and multilevel implementation could be secured for the V-CTM with a TCH as thin as 7 nm, demonstrating the sound feasibility of its X–Y scaling. This work presents new key parameters and understanding of device integration issues for future 3D structured memory transistors, and aims to contribute to the advancement of memory technologies using oxide semiconductor channels.

Stress-induced detwinning/reorientation of hierarchically twinned martensite and deformation micro-mechanism in Ni52Mn27Ga17Co4 shape memory alloys: Experimental and phase-field studies

Mechanical training via cyclic loading and unloading was widely used in NiMnGa shape memory alloys (SMAs) to create single or double variant states for reducing the resistance to variant reorientation and thus achieving a maximum shape memory strain. In this work, an effort was made to reveal the deformation mechanism of a polycrystalline Ni52Mn27Ga17Co4 high-temperature SMAs consisting of non-modulated (NM) martensite with uniform orientation and surrounding γ precipitates. The NM martensite microstructure was characterized as a self-accommodated hierarchical twinned-structure prior to deformation. Based on the interrupted in-situ EBSD measurements and phase-field simulations, it was demonstrated that the compressive loading resulted in the thickening of the stress-favored variants with the <001>NM direction perpendicular to compression axis at different dimensional scales. The path consisted of two subsequent processes–first detwinning of the nano-lamellae within one micro-variant through inter-lamellae boundary motion, and then reorientation of micro-variants through inter-plate boundary motion. The Schmid factor and preferred orientation of nano-lamellae (or strain accommodation) dominated the entire detwinning/reorientation process of NM martensite. As a result, the hierarchically twinned microstructure almost evolved into a single-variant state with the disappearance of packet boundaries. The ductile γ phase with network structure possessed more excellent strain-accommodated ability to the macroscopic deformation, hence significantly enhancing the mechanical properties including the ultimate compressive strength and elongation of NiMnGa SMAs.

Edge of Stability Echo State Network.

Echo state networks (ESNs) are time series processing models working under the echo state property (ESP) principle. The ESP is a notion of stability that imposes an asymptotic fading of the memory of the input. On the other hand, the resulting inherent architectural bias of ESNs may lead to an excessive loss of information, which in turn harms the performance in certain tasks with long short-term memory requirements. To bring together the fading memory property and the ability to retain as much memory as possible, in this article, we introduce a new ESN architecture called the Edge of Stability ESN (). The introduced model is based on defining the reservoir layer as a convex combination of a nonlinear reservoir (as in the standard ESN), and a linear reservoir that implements an orthogonal transformation. In virtue of a thorough mathematical analysis, we prove that the whole eigenspectrum of the Jacobian of the map can be contained in an annular neighborhood of a complex circle of controllable radius. This property is exploited to tune the 's dynamics close to the edge-of-chaos regime by design. Remarkably, our experimental analysis shows that model can reach the theoretical maximum short-term memory capacity (MC). At the same time, in comparison to conventional reservoir approaches, is shown to offer an excellent trade-off between memory and nonlinearity, as well as a significant improvement of performance in autoregressive nonlinear modeling and real-world time series modeling.

Hydrometeorological Factors Affecting the Carbon Exchange of the Himalayan Pine-dominated Ecosystem

The net carbon uptake of the natural and dense Himalayan ecosystems is not well explored at a sub-daily scale due to limited data availability. Here, we present the first-ever analysis of the hydrometeorological controls of the sub-daily scale Net Ecosystem Exchange (NEE) in a Western Himalayan pine-dominated ecosystem (Pinus roxburghii). We used observed half-hourly flux tower data for 104 weeks from December 2014 to November 2016. We found that the sub-daily (6 h) scale variability of NEE contributes 20–40% to its within-week variability. We generated transfer entropy (TE)-based weekly process networks with a maximum memory of 6 h to understand sub-daily scale processes. Using weekly networks, we present consistent causal links on a seasonal scale. The hydrometeorological variables affecting the NEE and Ecosystem Respiration (RE) at a sub-daily scale are Net Solar Radiation (NSR), Relative Humidity (RH), Surface Air Temperature (TA), and Sensible Heat Flux (SH). These variables are consistently active in the networks with seasonally varying magnitudes of TE. We also found that for both monsoon and post-monsoon, NEE receives more incoming links from meteorological variables during dry weeks than wet weeks. Precipitation (P) did not send a direct causal link to NEE or RE within a sub-daily scale but strongly influenced the causal associations between hydrometeorological and carbon exchange variables. The network does not show any immediate (within 6 h) impact of P on NEE, which probably implies a delayed response of vegetation to P through soil moisture. At higher memory, P may act as an influential background variable.

Big data-based digital management system for the whole process of financial reporting in SMEs

Abstract Based on big data technology, this paper designs a digital management system for the whole process of financial reimbursement and gives the automatic transformation process of digitalization of financial reimbursement claims and accounting documents. In the K-Means algorithm, the AHC algorithm idea is introduced to construct the DH-K-Means clustering algorithm, which in turn enables the financial reimbursement behavioral portrait of enterprise employees to be constructed. To confirm the effectiveness of the system application in this paper, performance testing and portrait analysis were carried out. The results show that the maximum memory utilization of the database server is 50.42% when the number of concurrent users is 100, and the average response time of the system reaches a maximum of 825.63ms when the number of concurrent users is 200. Based on big data, the whole process of financial reporting digital management system has good stability, which can make enterprises more aware of their financial situation and then develop appropriate financial reimbursement strategies.

149 Marker Effect P-Value for Large Genotype Populations with the Algorithm for Proven and Young

Abstract The single-step method (ssGBLUP) allows for simultaneous evaluations of populations composed of genotyped and non-genotyped animals. As a GBLUP-based method, the output of ssGBLUP is breeding values. However, interest might rely on the estimation of SNP effects and on how genome segments are associated with the phenotype of interest. When that is the case, SNP effects can be obtained from a linear transformation of breeding values, and P-values can be used as a measure of estimation certainty. P-values are derived from the prediction error variance of SNP effects, which relies on the prediction error (co)variance matrix of breeding values obtained from the inverse of the left-hand side of the mixed model equations (LHS). However, inverting the LHS becomes unfeasible with large, genotyped populations. Therefore, alternative strategies to overcome this problem and obtain p-values with increasing genotyped populations might be evaluated. In this study, we aimed to estimate marker effect p-values when the algorithm for proven and young (APY) was used to approximate the predictor error (co)variance for animals in a large, genotyped population. We estimate P-values of SNP marker effects for PWG in an Angus cattle population. The dataset was composed of around 844K phenotyped animals, 450K genotypes, and 1.8M pedigree records. Analyses were split into two sets. In the first set, a reduced genotype data of 50K was used, so p-values obtained from a direct-inverse of LHS using the regular G-1 (Exact_G) or APY G-1 (Exact_GAPY) were compared with APY G-1 and an approximation of the prediction error variance that did not require the inversion of the LHS (Approx_GAPY). In the second set of analyses, P-values with Approx_GAPY were estimated with the full genomic set of 450K (Approx_GAPY_450K), and computational requirements were recorded. All analyses were performed in three replicates. In the first set of analyses, genome-wide association with Exact_G uncovered two significant regions in chromosomes 7 and 20. Along all replicates, the same regions were also identified with Exact_GAPY and Approx_GAPY, indicating that similar resolutions are obtained with exact and approximated p-values. Results from Approx_GAPY_450K showed that, with a complete genome set, besides the two regions in chromosomes 7 and 20, two new regions in chromosomes 6 and 14 were identified. On average, the entire procedure for obtaining P-values with 450K genotyped animals had an elapsed wall clock time of 24h with a maximum memory usage of 87.6 GB. Altogether, our results suggest that with APY and the approximation of the prediction error variance, current computational boundaries for obtaining marker effect P-values for large genotyped populations should be lifted.

Performance analysis of optimized versatile video coding software decoders on embedded platforms

In recent years, the global demand for high-resolution videos and the emergence of new multimedia applications have created the need for a new video coding standard. Therefore, in July 2020, the versatile video coding (VVC) standard was released, providing up to 50% bit-rate savings for the same video quality compared to its predecessor high-efficiency video coding (HEVC). However, these bit-rate savings come at the cost of high computational complexity, particularly for live applications and on resource-constrained embedded devices. This paper evaluates two optimized VVC software decoders, named OpenVVC and Versatile Video deCoder (VVdeC), designed for low resources platforms. These decoders exploit optimization techniques such as data-level parallelism using single instruction multiple data (SIMD) instructions and functional-level parallelism using frame, tile, and slice-based parallelisms. Furthermore, a comparison of decoding runtime, energy, and memory consumption between the two decoders is presented while targeting two different resource-constraint embedded devices. The results showed that both decoders achieve real-time decoding of full high-definition (FHD) resolution on the first platform using 8 cores and high-definition (HD) real-time decoding for the second platform using only 4 cores with comparable results in terms of the average energy consumed: around 26 J and 15 J for the 8 cores and 4 cores platforms, respectively. Furthermore, OpenVVC showed better results regarding memory usage with a lower average maximum memory consumed during runtime than VVdeC.

Effects of the BalanCI on Working Memory and Balance in Children and Young Adults With Cochleovestibular Dysfunction.

This study aimed to: (1) determine the interaction between cognitive load and balance in children and young adults with bilateral cochleovestibular dysfunction who use bilateral cochlear implants (CIs) and (2) determine the effect of an auditory balance prosthesis (the BalanCI) on this interaction. Many (20 to 70%) children with sensorineural hearing loss experience some degree of vestibular loss, leading to poorer balance. Poor balance could have effects on cognitive resource allocation which might be alleviated by the BalanCI as it translates head-referenced cues into electrical pulses delivered through the CI. It is hypothesized that children and young adults with cochleovestibular dysfunction will demonstrate greater dual-task costs than typically-developing children during dual balance-cognition tasks, and that BalanCI use will improve performance on these tasks. Study participants were 15 typically-developing children (control group: mean age ± SD = 13.6 ± 2.75 years, 6 females) and 10 children and young adults who use bilateral CIs and have vestibular dysfunction (CI-V group: mean age ± SD=20.6 ± 5.36 years, 7 females). Participants completed two working memory tasks (backward auditory verbal digit span task and backward visuospatial dot matrix task) during three balance conditions: seated, standing in tandem stance with the BalanCI off, and standing in tandem stance with the BalanCI on. Working memory performance was quantified as total number of correct trials achieved. Postural stability was quantified as translational and rotational path length of motion capture markers worn on the head, upper body, pelvis, and feet, normalized by trial time. Relative to the control group, children and young adults in the CI-V group exhibited poorer overall working memory across all balance conditions ( p = 0.03), poorer translational postural stability (larger translational path length) during both verbal and visuospatial working memory tasks ( p < 0.001), and poorer rotational stability (larger rotational path length) during the verbal working memory task ( p = 0.026). The CI-V group also exhibited poorer translational ( p = 0.004) and rotational ( p < 0.001) postural stability during the backward verbal digit span task than backward visuospatial dot matrix task; BalanCI use reduced this stability difference between verbal and visuospatial working memory tasks for translational stability overall ( p > 0.9), as well as for rotational stability during the maximum working memory span (highest load) participants achieved in each task ( p = 0.91). Balance and working memory were impaired in the CI-V group compared with the control group. The BalanCI offered subtle improvements in stability in the CI-V group during a backward verbal working memory task, without producing a negative effect on working memory outcomes. This study supports the feasibility of the BalanCI as a balance prosthesis for individuals with cochleovestibular impairments.

Damage-coupled unified constitutive modeling of 316LN stainless steel including dynamic strain aging under various tension dwell time: A macroscopic phenomenological study

A damage-coupled unified constitutive model is developed for 316LN stainless steel based on the framework of the Abdel-Karim and Ohno model. In the modified model, a kinematic hardening coefficient related to accumulated plastic strain is introduced in the linear hardening term, and a damage coefficient is incorporated in the static recovery term. Meanwhile, the parameters of isotropic hardening and kinematic hardening are associated with the maximum plastic strain rate and plastic strain memory to describe the effects of dynamic strain aging and plastic strain memory. Additionally, the kinematic hardening coefficients and static recovery coefficients correlate with dwell time to simulate stress relaxation throughout the whole-life time. The comparison between the simulation and experimental results indicates the validity of the modified model under the conditions of low-cycle fatigue and creep-fatigue interaction. After identifying the material parameters using a combination of classic parameter determination methods and optimization algorithms, the cyclic stress response and hysteresis loops can be accurately simulated throughout the whole-life time.

CluMP: Clustered Markov Chain for Storage I/O Prefetch

Due to advancements in CPU and storage technologies, the processing speed of tasks has been increasing. However, there has been a relative slowdown in the data transfer speeds between disks and memory. Consequently, the issue of I/O processing speed has become a significant concern in I/O-intensive tasks. This research paper proposes CluMP, which predicts the next block to be requested within a process using a clustered Markov chain. Compared to the simple read-ahead approach commonly used in Linux systems, CluMP can predict prefetching more accurately and requires less memory for the prediction process. CluMP demonstrated a maximum memory hit ratio improvement of 191.41% in the KVM workload compared to read-ahead, as well as a maximum improvement of 130.81% in the Linux kernel build workload. Additionally, CluMP provides the advantage of adaptability to user objectives and utilized workloads by incorporating several parameters, thereby allowing for optimal performance across various workload patterns.

Formula omitted]S: Adaptive anomaly detection on sporadic data streams

With the widespread use of Internet applications, ensuring the quality and reliability of online services has become increasingly important. Therefore, anomaly detection methods play a critical role in identifying potential anomalies in the data streams of infrastructure systems and service applications. However, most of known detection methods have an underlying assumption that the data streams are continuous. In practice, we learn that many real-world data streams can be sporadic. It incurs particular challenges for the task of anomaly detection, for which the common preprocessing of downsampling on sporadic data can omit potential anomalies and delay alarms. In this paper, we propose an ensemble learning-based anomaly detection method on sporadic data streams named AD2S. It consists of two modules: a monitor module to continuously and adaptively determine the measure windows for observations, and a detection module that utilizes an isolation partition strategy to estimate the anomaly degree of each incoming observation. Based on experimental results on eight synthetic and public real-world datasets, our method outperforms other state-of-the-art methods with an average AUC score of 0.923. Additionally, our analysis demonstrates that the proposed method has constant amortized time and space complexity, enabling once detection within an average of 9.9 ms and maximum memory usage of 26.14 KB. The code of AD2S is open-sourced for further research.

The improvement of the shape memory effect of Cu-13.5Al–4Ni high-temperature shape memory alloys through Cr-, Mo-, or V-alloying

This study presents the effect of small additions of Cr, Mo, or V on the microstructure, crystallographic features of martensitic transformation, and shape memory effect of Cu-13.5Al-4Ni master alloy. The results show that the studied alloys consist of dominant β'1 (18R) martensitic structure and fine bcc precipitates which are riched in the alloying elements but poor in the Cu content. The martensitic transformation start temperatures of the alloys are 414K, 385K, and 394K when X = 0.5Cr, 0.5Mo, or 0.5V, respectively. More importantly, a drop of grain size was achieved in the studied alloys, leading to the improvement of the shape memory effect in Cu-13.5Al-4Ni alloys. As the result, the studied alloys exhibit a steady increase of shape memory effect as the pre-strain changed from 6% to 10%, which the maximum shape memory effect reaching 6.2%, 5.8%, and 5.5% for X = 0.5Cr, 0.5Mo, and 0.5V, respectively.

Data frame aware optimized Octomap-based dynamic object detection and removal in Mobile Laser Scanning data

The Mobile Laser Scanning (MLS) data inevitably includes dynamic objects because there are always other vehicles (e.g., other cars, motorbikes, bikes, etc.) moving in the area near the MLS data collection vehicle on the road. These dynamic objects need to be removed in advance for many point cloud applications. This paper designs an efficient and memory-friendly data frame aware optimized Octomap-based dynamic object detection and removal method for MLS data. Firstly, the input MLS data is split into multiple data frames based on the timestamp. Each data frame is inserted into a separate Octomap with part of its neighbouring data frames. A statistics-based method is applied to each data frame to find the passable voxel cell space (free space) in Octomap and all points in the free space are extracted as free points. Second, the region of interest (ROI) related to the dynamic object is delineated to retain free points related to dynamic objects. Then the free-point rate and the multi-return rate are calculated to further remove noise and vegetation points from free points. Finally, the fixed radius search is used to extract dynamic objects from the filtered free points. The proposed method is tested in four case sites in Delft, the Netherlands. Results show that 84.98% of dynamic objects are detected and extracted correctly. The proposed method is 18.27% more efficient on average than the original Octomap method, can be further accelerated by parallel computing, and only needs 39.40% of the maximum memory consumption.