Pipeline Scheme Research Articles

The TAC1 Gene in Candida albicans: Structure, Function, and Role in Azole Resistance: A Mini-Review.

Candidiasis is a common fungal infection caused by Candida species, with Candida albicans being the most prevalent. Resistance to azole drugs, commonly used to treat Candida infections, poses a significant challenge. Transcriptional activator candidate 1 (TAC1) gene has emerged as a key player in regulating drug resistance in C. albicans. This review explores the structure and function of the TAC1 gene and its role in azole resistance. This gene encodes a transcription factor that controls the expression of genes involved in drug resistance, such as efflux pump genes (CDR1, CDR2, and MDR1) and ERG11. Mutations in TAC1 can increase these genes' expression and confer resistance to azoles. Various TAC1 gene mutations, mostly gain-of-function mutations, have been identified, which upregulate CDR1 and CDR2 expression, resulting in azole resistance. Understanding the mechanisms of azole resistance mediated by the TAC1 gene is crucial for the strategies in the effective antifungal development pipeline.

Advanced tree-based machine learning methods for predicting the seismic response of regular and irregular RC frames

In this study, predicting the maximum drift ratio (MDR) for two types of structures: 8-story regular (R8) and vertically irregular (IR8) reinforced concrete (RC) frames are focused. To accomplish this, advanced tree-based machine learning (ML) methods, namely Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and Stochastic Gradient Boosting (SGB) are utilized. 2300 ground motion (GM) records, includes twenty-one input parameters and their nonlinear time history analyses results are considered as a dataset. The performance of the RF, XGBoost, and SGB models was examined considering the impact of the data pre-processing issue to test whether the use of data pre-processing approaches in model development could improve the prediction performance of the advanced ML models. Various data pre-processing techniques, including outlier detection using boxplot, missing data imputation using RF, data transformation using log transformation, and feature selection using Hill Climbing are employed as part of ML pipeline. The models' performance is evaluated using six different well-established statistical evaluation metrics. Subsequently, the results are systematically compared to identify the most effective prediction model. Our findings highlight the substantial impact of data pre-processing on the performance of ML models when predicting seismic responses, specifically the MDR of RC frames. Notably, RF consistently outperformed other ML models for both R8 and IR8 structures in the case of the simple ML pipeline strategy and data pre-processing approach. The RF model achieved R2 scores of R2 = 0.87 and R2 = 0.91 on the test data before implementing data pre-processing for R8 and IR8, respectively. After data pre-processing, R2 scores showed an improvement of 12% and 6.22% in the cases of R8 and IR8. In contrast, the performance scores of SGB consistently fell behind those of RF and XGBoost models.

A new hybrid GPU-CPU sparse LDLT factorization algorithm with GPU and CPU factorizing concurrently

This paper proposes a new task assignment scheme for sparse LDLT factorization on a hybrid GPU-CPU platform, and correspondingly an efficient supernodal algorithm, GPU Node and CPU Pipelining (GNCP). GNCP assigns a large number of weakly coupled tasks to GPU and CPU respectively so that GPU and CPU can execute different factorization tasks concurrently without explicit synchronization, which has not been achieved by previous researches. More precisely, based on the number of CPU threads, the global memory of GPU as well the size of the matrix, we introduce the concept of the truncation level in the partition tree built by the multi-level graph partitioning. The sum of FLOP associated with decoupled nodes in the truncation level accounts for about 65% of the whole computations in tested numerical examples. GNCP assigns the factorization tasks of each node in the truncation level to GPU. The data of each node in the truncation level are copied to the global memory of GPU and factorized entirely and independently by GPU one-by-one. Once factorization of any node in the truncation level has been completed, the cross-factorizations of this node to its ancestors and the following related operations are decoupled from other GPU tasks. These tasks are assigned to CPU threads with a reentrant pipelining strategy, accounting for about 35% of the whole calculations. Using this novel design, CPU and GPU are working concurrently and a large part of factorization is performed in overlapped time. Numerical tests for matrices from SuiteSparse matrix collection show higher performance than the well-known hybrid GPU-CPU solver CHOLMOD.

High-Matching and Low-Cost Realization of the FHN Neuron Model on Reconfigurable FPGA Board.

The main objectives of neuromorphic engineering are the research, modeling, and implementation of neural functioning in the human brain. We provide a hardware solution that can replicate such a nature-inspired system by merging multiple scientific domains and is based on neural cell processes. This work provides a modified version of the original Fitz-Hugh Nagumo (FHN) neuron using a simple 2V term called Hybrid Piece-Wised Base-2 Model (HPWBM), which accurately reproduces numerous patterns of the original neuron model. With reduced terms, we suggest modifying the original nonlinear term to achieve high matching accuracy and little computing error. Time domain and phase portraits are used to validate the proposed model, which shows that it can reproduce all of the FHN model's properties with high accuracy and little mistake. We provide an effective digital hardware approach for large-scale neuron implementations based on resource-sharing and pipelining strategies. The Hardware Description Language (HDL) is used to construct the hardware on an FPGA as a proof of concept. The recommended model hardly uses 0.48 percent of the resources on a Virtex 4 FPGA board, according to the results of the hardware implementation. The circuit can run at a maximum frequency of 448.236 MHz, according to the static timing study.

A multi-modal vision-language pipeline strategy for contour quality assurance and adaptive optimization

Objective. Accurate delineation of organs-at-risk (OARs) is a critical step in radiotherapy. The deep learning generated segmentations usually need to be reviewed and corrected by oncologists manually, which is time-consuming and operator-dependent. Therefore, an automated quality assurance (QA) and adaptive optimization correction strategy was proposed to identify and optimize ‘incorrect’ auto-segmentations. Approach. A total of 586 CT images and labels from nine institutions were used. The OARs included the brainstem, parotid, and mandible. The deep learning generated contours were compared with the manual ground truth delineations. In this study, we proposed a novel contour quality assurance and adaptive optimization (CQA-AO) strategy, which consists of the following three main components: (1) the contour QA module classified the deep learning generated contours as either accepted or unaccepted; (2) the unacceptable contour categories analysis module provided the potential error reasons (five unacceptable category) and locations (attention heatmaps); (3) the adaptive correction of unacceptable contours module integrate vision-language representations and utilize convex optimization algorithms to achieve adaptive correction of ‘incorrect’ contours. Main results. In the contour QA tasks, the sensitivity (accuracy, precision) of CQA-AO strategy reached 0.940 (0.945, 0.948), 0.962 (0.937, 0.913), and 0.967 (0.962, 0.957) for brainstem, parotid and mandible, respectively. The unacceptable contour category analysis, the (FI,AccI,Fmicro,Fmacro) of CQA-AO strategy reached (0.901, 0.763, 0.862, 0.822), (0.855, 0.737, 0.837, 0.784), and (0.907, 0.762, 0.858, 0.821) for brainstem, parotid and mandible, respectively. After adaptive optimization correction, the DSC values of brainstem, parotid and mandible have been improved by 9.4%, 25.9%, and 13.5%, and Hausdorff distance values decreased by 62%, 70.6%, and 81.6%, respectively. Significance. The proposed CQA-AO strategy, which combines QA of contour and adaptive optimization correction for OARs contouring, demonstrated superior performance compare to conventional methods. This method can be implemented in the clinical contouring procedures and improve the efficiency of delineating and reviewing workflow.

Adaptability and efficiency of the CMS Level-1 Global Trigger firmware implementation for Phase-2

We present details on the new Level-1 Global Trigger at CMS for the upcoming high-luminosity operation of the LHC. Our focus is on the newly developed firmware, which employs a bottom-up generic approach to enhance menu adaptability and accommodate the increase in upstream data. We also highlight our efficient pipelining strategy that ensures excellent routability at 480 MHz. Furthermore, we discuss the firmware implementation for three prototypes targeting Serenity boards, together with their current and future testing and validation endeavours.

Montgomery Multiplication Scalable Systolic Designs Optimized for DSP48E2

This article describes an extensive study of the use of DSP48E2 Slices in Ultrascale FPGAs to design hardware versions of the Montgomery Multiplication algorithm for the hardware acceleration of modular multiplications. Our fully scalable systolic architectures result in parallelized, DSP48E2-optimized scheduling of operations analogous to the FIOS block variant of the Montgomery Multiplication. We explore the impacts of different pipelining strategies within DSP blocks, scheduling of operations, processing element configurations, global design structures and their tradeoffs in terms of performance and resource costs. We discuss the application of our methodology to multiple types of DSP primitives. We provide ready-to-use fast, efficient, and fully parametrizable designs, which can adapt to a wide range of requirements and applications. Implementations are scalable to any operand width. Our most efficient designs can perform 128, 256, 512, 1024, 2048, and 4096 bits Montgomery modular multiplications in 0.0992 μs, 0.2032 μs, 0.3952 μs, 0.7792μs, 1.550 μs, and 3.099 μs using 4, 6, 11, 21, 41, and 82 DSP blocks, respectively.

Optimization of pipeline scheme for measuring rheological parameters of drilling fluid by pipe flow method

Drilling fluid is known as the blood of drilling engineering, which plays an important role in drilling engineering. The real-time measurement of drilling fluid rheological parameters is crucial to the optimization and adjustment of drilling scheme. Pipe flow method is a common method of measuring drilling fluid rheology parameters, pipe flow method monitoring equipment is an important reason of the pipe flow method monitoring principle, the analysis of the reason, put forward the spiral development section optimization scheme, using F LUENT numerical simulation software for simulation, the optimization platform and verify the reliability of the optimization of numerical simulation results, greatly reduce the pipe flow, method monitoring equipment volume. It also provides a theoretical basis for other pipeline measurement fields.

Obstetric Racism, Education, and Racial Concordance.

The United States holds the distinction of being the developed country with the worst perinatal outcomes despite spending the most per capita on health care. Black women are three to four times more likely than White women to experience adverse birth outcomes. These outcomes persist despite access to prenatal care, insurance, and college education. A long overdue racial reckoning has arrived, beginning with acknowledging the fallacy of race-based medicine and the role of enduring systemic racism as foundational to obstetric racism in the reproductive lives of Black women. Centering voices of Black women and honoring their lived experiences are essential to providing respectful maternity care. Naming micro- and macroaggressions experienced by Black women allows for dismantling of systemic barriers which perpetuate inequitable outcomes and enable substandard care. Racial concordance (shared racial identity) is one tool to consider in creating safe health care spaces for Black women. Improving diversity of the nursing workforce specifically, and the health care workforce in general, is necessary to affect racial concordance. Application of skills in cultural humility can mitigate challenges associated with adverse patient encounters for Black women as diversity pipeline strategies are explored. Histories of foundational fallacy, their impact on care and outcomes, and patient-driven indicators for improving pregnancy care experiences for women of color are explored through the lens of a Black physician and the collective reproductive health workforce.

How APIs Create Growth by Inverting the Firm

Traditional asset management strategy has emphasized building barriers to entry or closely guarding unique assets to maintain a firm’s comparative advantage. A new “inverted firm” paradigm, however, has emerged. Under this strategy, firms share data seeking to become platforms by opening digital services to third parties and capturing part of their external surplus. This contrasts with a “pipeline” strategy where the firm itself creates value. This paper quantitatively estimates the effect of adopting an inverted firm strategy through the lens of application programming interfaces (APIs), a key enabling technology. Using both public data and those of a private API development firm, we document rapid growth of the API network and connecting apps since 2005. We then perform difference-in-difference and synthetic control analyses and find that public firms adopting public APIs grew an additional 38.7% over 16 years relative to similar nonadopters. We find no significant effect from the use of APIs purely for internal productivity: the pipeline strategy. Within the subset of firms that adopt public APIs, those that attract more third-party complementors and those that become more central to the network see faster growth. Using variation in network centrality caused by API degradation, an instrumental variables analysis confirms a causal role for APIs in firm market value. Finally, we document an important downside of public APIs: increased risk of data breach. Overall, these facts lead us to conclude that APIs have a large and positive impact on economic growth and do so primarily by enabling an inverted firm strategy. This paper was accepted by D. J. Wu, information systems. Funding: This work was partially supported by Mulesoft and Apigee Corporation. Supplemental Material: The data files and online appendix are available at https://doi.org/10.1287/mnsc.2023.4968 .

Simulation‐based decision‐making and implementation of tools in hybrid crop breeding pipelines

AbstractNew technologies have been developed over the last few years aiming to support breeding pipeline optimization for long‐term genetic gains. However, the implementation of these new tools and their impact on any breeding program's budget are not well studied. Here, we compare multiple breeding pipeline strategies accounting for genomic selection and high‐throughput phenotyping (HTP) by means of hybrid gain and cost‐effectiveness. We simulated a hybrid crop breeding program through coalescent theory. We compared two strategies for parental updates and four breeding pipelines: conventional breeding pipeline; conventional breeding pipeline with HTP; conventional breeding pipeline with genomic selection; conventional breeding pipeline with genomic selection and HTP. All analyses were implemented under three different levels of genotype‐by‐environment interaction (G×E) and two trait heritabilities (0.3 and 0.7). Overall, the results show that scenarios with early parental selection perform better than the others. In addition, the implementation of HTP delivered the highest hybrid gain in the long‐term, whereas the implementation of genomic selection seems to be more cost‐effective. We suggest, considering breeding programs with complex trait inheritance and accounting for higher levels of G×E, investing in breeding pipelines accounting for genomic selection as a strategy to create and maintain long‐term hybrid gain. Moreover, considering an unconstrained budget, the investment in both, genomic selection and HTP, represents the best strategy. Hence, these results provide strategies that may aid breeders in optimizing self‐pollination breeding programs.

End-to-End Models for Chemical-Protein Interaction Extraction: Better Tokenization and Span-Based Pipeline Strategies.

End-to-end relation extraction (E2ERE) is an important task in information extraction, more so for biomedicine as scientific literature continues to grow exponentially. E2ERE typically involves identifying entities (or named entity recognition (NER)) and associated relations, while most RE tasks simply assume that the entities are provided upfront and end up performing relation classification. E2ERE is inherently more difficult than RE alone given the potential snowball effect of errors from NER leading to more errors in RE. A complex dataset in biomedical E2ERE is the ChemProt dataset (BioCreative VI, 2017) that identifies relations between chemical compounds and genes/proteins in scientific literature. ChemProt is included in all recent biomedical natural language processing benchmarks including BLUE, BLURB, and BigBio. However, its treatment in these benchmarks and in other separate efforts is typically not end-to-end, with few exceptions. In this effort, we employ a span-based pipeline approach to produce a new state-of-the-art E2ERE performance on the ChemProt dataset, resulting in > 4% improvement in F1-score over the prior best effort. Our results indicate that a straightforward fine-grained tokenization scheme helps span-based approaches excel in E2ERE, especially with regards to handling complex named entities. Our error analysis also identifies a few key failure modes in E2ERE for ChemProt.

NCache: A Machine-Learning Cache Management Scheme for Computational SSDs

Inside a solid-state disk (SSD), cache stores frequently accessed data to shorten user-I/O response time and reduce the number of read/write operations in flash memory, thereby improving SSD performance and lifetime. Most existing cache schemes anchor in the spatiotemporal locality of I/O requests in workloads. In the face of a long-time workload, high performance and hit rate often get lost in these caching schemes. Flash memory-aware caching schemes trade hit ratio to prolong SSD lifetime. In this paper, we advocate for a machine learNing-based caching scheme named NCache to optimize both hit ratio and SSD performance. In NCache, we construct a machine learning (i.e., ML) model to predict whether data are re-accessed before being evicted from the cache. The cache replacement scheme preferentially evicts data that would not be accessed in the cache. The cache space is conserved for valid data that are likely to be repeatedly accessed. A pipelined scheme is implemented to accelerate the ML model, alleviating the time-cost of NCache. A double-linked list boosts the data addressing and cache replacement process. NCache is orthogonal to the existing caching schemes within the flash translation layer. The results validate NCache under a handful of real-world enterprise traces. Taking prn_0 as an example, NCache reduces the response time of LRU, CFLRU, GCaR_LRU, GCaR_CFLRU, and LCR by up to 15% with an average of 6.4%. The erase count is slashed by 16% at the maximum. Importantly, NCache is adroit at optimizing write amplification by up to 15.9%.

Auditory perception architecture with spiking neural network and implementation on FPGA

Spike-based perception brings up a new research idea in the field of neuromorphic engineering. A high-performance biologically inspired flexible spiking neural network (SNN) architecture provides a novel method for the exploration of perception mechanisms and the development of neuromorphic computing systems . In this article, we present a biological-inspired spike-based SNN perception digital system that can realize robust perception. The system employs a fully paralleled pipeline scheme to improve the performance and accelerate the processing of feature extraction. An auditory perception system prototype is realized on ten Intel Cyclone field-programmable gate arrays, which can reach the maximum frequency of 107.28 MHz and the maximum throughput of 5364 Mbps. Our design also achieves the power of 5. 148 W/system and energy efficiency of 845.85 μJ. Our auditory perception implementation is also proved to have superior robustness compared with other SNN systems. We use TIMIT digit speech in noise in accuracy testing. Result shows that it achieves up to 85.75% speech recognition accuracy under obvious noise conditions (signal-to-noise ratio of 20 dB) and maintain small accuracy attenuation with the decline of the signal-to-noise ratio. The overall performance of our proposed system outperforms the state-of-the-art perception system on SNN.

FERFM: An Enhanced Facial Emotion Recognition System Using Fine-tuned MobileNetV2 Architecture

Deep learning techniques have recently drawn a lot of attention towards automatic facial emotion recognition (FER) applications such as diagnosis of psychiatric disorders, mental health monitoring, and other human–computer interaction applications. But the models that have been developed, have still not been generalized due to the absence of substantial emotion datasets and high operational costs. To address these issues, we propose a novel technique called facial emotion recognition using fine-tuned MobileNetV2 architecture (FERFM). It uses the concept of transfer learning to improve the performance of FER systems for mobile devices. It transfers the knowledge from pre-trained ImageNet dataset to profile view RGB-KDEF dataset for diversity among learnt features. Initially, data descriptive analysis is performed on input KDEF images to automate the necessary knowledge. Then the pipeline strategy is introduced, where the pre-trained MobileNetV2 architecture is fine tuned by eliminating the last six layers and adding dropout regularization, max pooling and dense layer. To test the performance of the proposed work, we ran several tests using the KDEF-RGB dataset, FER13 dataset and real-time facial expression images to classify the results into seven discrete categories of emotions. It was also verified on VGG-16 pretrained model and demonstrated remarkable performance comparatively. Proposed FERFM achieves 85.7% accuracy and requires 1,510,599 number of trainable parameters, 43 ms running time per image that proves the enhancement of FER performance and outperforms the existing state of art methods. Also, the ability to achieve low operational costs while maintaining high accuracy demonstrates proficiency in mobile applications.

Simulation-based fault-tolerant multiprocessors system

System reliability is an important issue in designing modern multiprocessor systems. This paper proposes a fault-tolerant, scalable, multiprocessor system architecture that adopts a pipeline scheme. To verify the performance of the proposed system, the SimEvent/Stateflow tool of the MATLAB program was used to simulate the system. The proposed system uses twelve processors (P), connected in a linear array, to build a ten-stage system with two backup processors (BP). However, the system can be expanded by adding more processors to increase pipeline stages and performance, and more backup processors to increase system reliability. The system can automatically reorganize itself in the event of a failure of one or two processors and execution continues without interruption. Each processor communicates with its neighboring processors through input/output (I/O) ports which are used as bypass links between the processors. In the event of a processor failure, the function of the faulty processor is assigned to the next processor that is free from faults. The fast Fourier transform (FFT) algorithm is implemented on the simulated circuit to evaluate the performance of the proposed system. The results showed that the system can continue to execute even if one or two processors fail without a noticeable decrease in performance.

Real-time data processing for ultrafast X-ray computed tomography using modular CUDA based pipelines

In this article, a new version of the Real-time Image Stream Algorithms (RISA) data processing suite is introduced. It now features online detector data acquisition, high-throughput data dumping and enhanced real-time data processing capabilities. The achieved low-latency real-time data processing extends the application of ultrafast electron beam X-ray computed tomography (UFXCT) scanners to real-time scanner control and process control. We implemented high performance data packet reception based on data plane development kit (DPDK) and high-throughput data storing using both hierarchical data format version 5 (HDF5) as well as the adaptable input/output system version 2 (ADIOS2). Furthermore, we extended RISA's underlying pipelining framework to support the fork-join paradigm. This allows for more complex workflows as it is necessary, e.g. for online data processing. Also, the pipeline configuration is moved from compile-time to runtime, i.e. processing stages and their interconnections can now be configured using a configuration file. In several benchmarks, RISA is profiled regarding data acquisition performance, data storage throughput and overall processing latency. We found that using direct IO mode significantly improves data writing performance on the local data storage. We could further prove that RISA is now capable of concurrently receiving, processing and storing data from up to 768 detector channels (3072 MB/s) at 8000 fps on a single-GPU computer in real-time. Program summaryProgram Title: GLADOS/RISACPC Library link to program files:https://doi.org/10.17632/65sx747rvm.2Developer's repository link:https://codebase.helmholtz.cloud/risaLicensing provisions: Apache-2.0Programming language: C++Journal reference of previous version: Comput. Phys. Commun. 219 (2017) 353-360 [1]Does the new version supersede the previous version?: Yes.Reasons for the new version: Extended capabilities for real-time operation with latest UFXCT hardware.Summary of revisions: (i) Add forking and joining of processing pipeline branches(ii) Add runtime (re-)configuration of pipeline stages and connections(iii) Add UDP receiver stage to acquire detector data in real-time(iv) Add high-throughput data dumpingNature of problem: Ultrafast electron beam X-ray computed tomography scanners stream multiple Gigabytes of raw data per second via Ethernet to a control computer. Receiving the data with low latency, real-time image-based control would become possible. For this, data need to be captured from the network, stored on disk, reconstructed and post-processed concurrently. The current total data rate of up to 3072 MB/s requires high-throughput solutions for each of these tasks.Solution method: Using a pipeline scheme, RISA processes incoming raw data in distinct stages (sources, processors, sinks). These are implemented in GPU kernels and are executed concurrently to exploit data parallelism as well as task parallelism. To capture detector data, we implemented a UDP packet capturing stage based on DPDK [2] which acts as a source stage. By allowing the pipeline to fork up into multiple branches, we concurrently acquire, store and process the data. For storing these data, we use the HDF5 format [3]. We achieve the required data rates by writing in direct IO mode onto an SSD array in RAID 0 configuration.Additional comments including restrictions and unusual features: RISA provides a set of general-purpose processing stages which are suitable for generic image stream processing.

SwPHoToNs: Toward trillion‐body‐scale cosmologicalN‐body simulations on SunwayTaihuLightsupercomputer

AbstractCosmologicalN‐body simulations have been essential for astronomers to study the formation of nonlinear structures and hypotheses of dark matter, dark energy, etc. The scale of the problem naturally leads to extreme scenarios with billions or even trillions of particles, thus demanding massive computational power and highly efficient algorithms. In this paper, we present swPHoToNs, a Particle‐Mesh (PM) and Fast Multipole Method‐ (FMM) based code that can perform cosmological simulations with trillions of particles efficiently on the Sunway TaihuLight supercomputer. Our design includes three novel optimizations: (1) a multilevel domain decomposition and dynamic load‐balancing scheme; (2) a pipeline strategy for tree traversal and gravity calculation; (3) optimizations for both computation and MPI kernels with consideration of the hardware features. We manage to conduct cosmological simulations which contain up to 1.6 trillion particles, obtaining a sustained performance of 56.3 PFlops with a weak‐scaling parallel efficiency of 80.9% and a computational efficiency of 44.9%.

An Integrated Real-Time FMCW Radar Baseband Processor in 40-nm CMOS

In this paper, a pipelined frequency-modulated continuous-wave (FMCW) radar baseband processor applied to real-time applications is proposed and implemented in 40-nm CMOS technology. The FMCW radar signal processing time is analyzed according to the system specifications. On the basis of the theoretical analysis and systematic considerations, a pipelined baseband architecture with internal single-port static random access memory (SRAM) is employed. The baseband processor is mainly composed of two-dimensional fast Fourier transform (2D-FFT), two-dimensional constant false alarm rate (2D-CFAR), digital beam-forming (DBF), and memory control modules. The 2D-FFT module is structured with a pipelined scheme and avoids the waste of data transferring time between modules. The 2D-CFAR module is programmable for different applications. The designed address control is proposed to depose the edge cells. The processor occupies a core chip area of 3.353 mm ×3.353 mm and has been tested on the personal computer (PC) and field programmable gate array (FPGA) platform. The power consumption and processing time are also analyzed and compared with other works. The processor consumes 55.65 mW, including SRAMs. The processing time is 12.67 ms with the maximum window size and 256 targets when operating at 125 MHz. This time is estimated based on the assumption that each chirp lasts for 0.04096 ms, and data input takes 10.48 ms. Within this period, the range FFT is completed. The Doppler FFT, 2D-CFAR with the maximum window size, and DBF with 256 targets require 0.80 ms, 1.16 ms, and 0.23 ms respectively.

A Pipelining-Based Heterogeneous Scheduling and Energy-Throughput Optimization Scheme for CNNs Leveraging Apache TVM

Extracting information of interest from continuous video streams is a strongly demanded computer vision task. For the realization of this task at the edge using the current de-facto standard approach, i.e., deep learning, it is critical to optimize key performance metrics such as throughput and energy consumption according to prescribed application requirements. This allows achieving timely decision-making while extending the battery lifetime as much as possible. In this context, we propose a method to boost neural-network performance based on a co-execution strategy that exploits hardware heterogeneity on edge platforms. The enabling tool is Apache TVM, a highly efficient machine-learning compiler compatible with a diversity of hardware back-ends. The proposed approach solves the problem of network partitioning and distributes the workloads to make concurrent use of all the processors available on the board following a pipeline scheme. We conducted experiments on various popular CNNs compiled with TVM on the Jetson TX2 platform. The experimental results based on measurements show a significant improvement in throughput with respect to a single-processor execution, ranging from 14% to 150% over all tested networks. Power-efficient configurations were also identified, accomplishing energy reductions above 10%.