ACE-GCN: A Fast Data-driven FPGA Accelerator for GCN Embedding

Mapping Intimacies ◽

10.31219/osf.io/qcd45 ◽

2021 ◽

Author(s):

Jose Romero Hung

Keyword(s):

Computational Complexity ◽

Low Power ◽

Power Law ◽

Real World ◽

Energy Efficient ◽

Storage Capacity ◽

Structural Similarity ◽

Optimization Techniques ◽

Data Driven ◽

Implicit Processing

ACE-GCN is a fast, resource conservative and energy-efficient, FPGA accelerator for graph convolutional embedding with data-drivenqualities, intended for low-power in-place deployment. Our accelerator exploits the inherent qualities of power law distributionexhibited by real-world graphs, such as structural similarity, replication, and features exchangeability. Contrary to other hardwareimplementations of GCN, on which dataset sparsity becomes an issue and is bypassed with multiple optimization techniques, ourarchitecture is designed to take advantage of this very same situation. We implement an innovative hardware architecture, supportedby our “implicit-processing-by-association” concept. The computational relief and consequential acceleration effect come from thepossibility of replacing rather complex convolutional operations for faster LUT-based comparators and automatic convolutionalresult estimations. We are able to transfer computational complexity into storing capacity, under controllable design parameters.ACE-GCN accelerator core operation consists of orderly parading a set of vector-based, sub-graph structures named “types”, linked topre-calculated embeddings, to incoming "sub-graphs-in-observance", denominated SIO in our work, for either their graph embeddingassumption or their unavoidable convolutional processing, decision depending on the level of similarity obtained from a Jaccardfeature-based coefficient. Results demonstrate that our accelerator has a competitive amount of acceleration; depending on datasetand resource target; between 100× to 1600× PyG baseline, coming close to AWB-GCN by 40% to 70% on smaller datasets and evensurpassing AWB-GCN for larger with controllable accuracy loss levels. We further demonstrate the parallelism potentiality of ourapproach by analyzing the effect of storage capacity on the gradual reliving

Download Full-text

ACE-GCN: A Fast Data-driven FPGA Accelerator for GCN Embedding

ACM Transactions on Reconfigurable Technology and Systems ◽

10.1145/3470536 ◽

2021 ◽

Vol 14 (4) ◽

pp. 1-23

Author(s):

José Romero Hung ◽

Chao Li ◽

Pengyu Wang ◽

Chuanming Shao ◽

Jinyang Guo ◽

...

Keyword(s):

Real World ◽

Energy Efficient ◽

Optimization Techniques ◽

Data Driven ◽

Processing Conditions ◽

Power Law Distribution ◽

Implicit Processing ◽

Convolution Operation ◽

Hardware Implementations ◽

Similarity Calculation

ACE-GCN is a fast and resource/energy-efficient FPGA accelerator for graph convolutional embedding under data-driven and in-place processing conditions. Our accelerator exploits the inherent power law distribution and high sparsity commonly exhibited by real-world graphs datasets. Contrary to other hardware implementations of GCN, on which traditional optimization techniques are employed to bypass the problem of dataset sparsity, our architecture is designed to take advantage of this very same situation. We propose and implement an innovative acceleration approach supported by our “implicit-processing-by-association” concept, in conjunction with a dataset-customized convolutional operator. The computational relief and consequential acceleration effect arise from the possibility of replacing rather complex convolutional operations for a faster embedding result estimation. Based on a computationally inexpensive and super-expedited similarity calculation, our accelerator is able to decide from the automatic embedding estimation or the unavoidable direct convolution operation. Evaluations demonstrate that our approach presents excellent applicability and competitive acceleration value. Depending on the dataset and efficiency level at the target, between 23× and 4,930× PyG baseline, coming close to AWB-GCN by 46% to 81% on smaller datasets and noticeable surpassing AWB-GCN for larger datasets and with controllable accuracy loss levels. We further demonstrate the unique hardware optimization characteristics of our approach and discuss its multi-processing potentiality.

Download Full-text

A quantum-enhanced precision medicine application to support data-driven clinical decisions for the personalized treatment of advanced knee osteoarthritis: development and preliminary validation of precisionKNEE QNN

10.1101/2021.12.13.21267704 ◽

2021 ◽

Author(s):

Stefano Olgiati ◽

Nima Heidari ◽

Davide Meloni ◽

Federico Pirovano ◽

Ali Noorani ◽

...

Keyword(s):

Machine Learning ◽

Computational Complexity ◽

Knee Osteoarthritis ◽

Precision Medicine ◽

Real World ◽

Data Driven ◽

Personalized Treatment ◽

Validation Dataset ◽

Clinical Decisions ◽

Quantum Machine Learning

Background Quantum computing (QC) and quantum machine learning (QML) are promising experimental technologies which can improve precision medicine applications by reducing the computational complexity of algorithms driven by big, unstructured, real-world data. The clinical problem of knee osteoarthritis is that, although some novel therapies are safe and effective, the response is variable, and defining the characteristics of an individual who will respond remains a challenge. In this paper we tested a quantum neural network (QNN) application to support precision data-driven clinical decisions to select personalized treatments for advanced knee osteoarthritis. Methods Following patients consent and Research Ethics Committee approval, we collected clinico-demographic data before and after the treatment from 170 patients eligible for knee arthroplasty (Kellgren-Lawrence grade ≥ 3, OKS ≤ 27, Age ≥ 64 and idiopathic aetiology of arthritis) treated over a 2 year period with a single injection of microfragmented fat. Gender classes were balanced (76 M, 94 F) to mitigate gender bias. A patient with an improvement ≥ 7 OKS has been considered a Responder. We trained our QNN Classifier on a randomly selected training subset of 113 patients to classify responders from non-responders (73 R, 40 NR) in pain and function at 1 year. Outliers were hidden from the training dataset but not from the validation set. Results We tested our QNN Classifier on a randomly selected test subset of 57 patients (34 R, 23 NR) including outliers. The No Information Rate was equal to 0.59. Our application correctly classified 28 Responders out of 34 and 6 non-Responders out of 23 (Sensitivity = 0.82, Specificity = 0.26, F1 Statistic= 0.71). The Positive (LR+) and Negative (LR-) Likelihood Ratios were respectively 1.11 and 0.68. The Diagnostic Odds Ratio (DOR) was equal to 2. Conclusions Preliminary results on a small validation dataset show that quantum machine learning applied to data-driven clinical decisions for the personalized treatment of advanced knee osteoarthritis is a promising technology to reduce computational complexity and improve prognostic performance. Our results need further research validation with larger, real-world unstructured datasets, and clinical validation with an AI Clinical Trial to test model efficacy, safety, clinical significance and relevance at a public health level.

Download Full-text

A Data-Driven Method for Energy Consumption Prediction and Energy-Efficient Routing of Electric Vehicles in Real-World Conditions

Energies ◽

10.3390/en10050608 ◽

2017 ◽

Vol 10 (5) ◽

pp. 608 ◽

Cited By ~ 20

Author(s):

Cedric De Cauwer ◽

Wouter Verbeke ◽

Thierry Coosemans ◽

Saphir Faid ◽

Joeri Van Mierlo

Keyword(s):

Energy Consumption ◽

Electric Vehicles ◽

Real World ◽

Energy Efficient ◽

Data Driven ◽

Energy Efficient Routing ◽

Energy Consumption Prediction ◽

Consumption Prediction

Download Full-text

Sublinear domination and core–periphery networks

Scientific Reports ◽

10.1038/s41598-021-94105-8 ◽

2021 ◽

Vol 11 (1) ◽

Author(s):

Marios Papachristou

Keyword(s):

Network Model ◽

Power Law ◽

Real World ◽

Random Network ◽

Small World ◽

The Core ◽

Entire Network

AbstractIn this paper we devise a generative random network model with core–periphery properties whose core nodes act as sublinear dominators, that is, if the network has n nodes, the core has size o(n) and dominates the entire network. We show that instances generated by this model exhibit power law degree distributions, and incorporates small-world phenomena. We also fit our model in a variety of real-world networks.

Download Full-text

Coordination of low-power nonlinear multi-agent systems using cloud computing and a data-driven hybrid predictive control method

Control Engineering Practice ◽

10.1016/j.conengprac.2020.104722 ◽

2021 ◽

Vol 108 ◽

pp. 104722

Author(s):

Haoran Tan ◽

Yaonan Wang ◽

Hang Zhong ◽

Min Wu ◽

Yiming Jiang

Keyword(s):

Cloud Computing ◽

Low Power ◽

Predictive Control ◽

Control Method ◽

Data Driven ◽

Multi Agent Systems ◽

Agent Systems ◽

Multi Agent

Download Full-text

Maximal modularity and the optimal size of parliaments

Scientific Reports ◽

10.1038/s41598-021-93639-1 ◽

2021 ◽

Vol 11 (1) ◽

Author(s):

Luca Gamberi ◽

Yanik-Pascal Förster ◽

Evan Tzanis ◽

Alessia Annibale ◽

Pierpaolo Vivo

Keyword(s):

Power Law ◽

Real World ◽

Empirical Data ◽

Random Network ◽

Functional Relation ◽

Optimal Size ◽

Real World Data ◽

Universal Power Law ◽

The Empirical Analysis ◽

Democratic Country

AbstractAn important question in representative democracies is how to determine the optimal parliament size of a given country. According to an old conjecture, known as the cubic root law, there is a fairly universal power-law relation, with an exponent equal to 1/3, between the size of an elected parliament and the country’s population. Empirical data in modern European countries support such universality but are consistent with a larger exponent. In this work, we analyse this intriguing regularity using tools from complex networks theory. We model the population of a democratic country as a random network, drawn from a growth model, where each node is assigned a constituency membership sampled from an available set of size D. We calculate analytically the modularity of the population and find that its functional relation with the number of constituencies is strongly non-monotonic, exhibiting a maximum that depends on the population size. The criterion of maximal modularity allows us to predict that the number of representatives should scale as a power-law in the size of the population, a finding that is qualitatively confirmed by the empirical analysis of real-world data.

Download Full-text

Investigating the Potential of Network Optimization for a Constrained Object Detection Problem

Journal of Imaging ◽

10.3390/jimaging7040064 ◽

2021 ◽

Vol 7 (4) ◽

pp. 64

Author(s):

Tanguy Ophoff ◽

Cédric Gullentops ◽

Kristof Van Beeck ◽

Toon Goedemé

Keyword(s):

Computational Complexity ◽

Object Detection ◽

Network Optimization ◽

Real Life ◽

Optimization Techniques ◽

Training Data ◽

Single Shot ◽

Standard Object ◽

Number Of Classes

Object detection models are usually trained and evaluated on highly complicated, challenging academic datasets, which results in deep networks requiring lots of computations. However, a lot of operational use-cases consist of more constrained situations: they have a limited number of classes to be detected, less intra-class variance, less lighting and background variance, constrained or even fixed camera viewpoints, etc. In these cases, we hypothesize that smaller networks could be used without deteriorating the accuracy. However, there are multiple reasons why this does not happen in practice. Firstly, overparameterized networks tend to learn better, and secondly, transfer learning is usually used to reduce the necessary amount of training data. In this paper, we investigate how much we can reduce the computational complexity of a standard object detection network in such constrained object detection problems. As a case study, we focus on a well-known single-shot object detector, YoloV2, and combine three different techniques to reduce the computational complexity of the model without reducing its accuracy on our target dataset. To investigate the influence of the problem complexity, we compare two datasets: a prototypical academic (Pascal VOC) and a real-life operational (LWIR person detection) dataset. The three optimization steps we exploited are: swapping all the convolutions for depth-wise separable convolutions, perform pruning and use weight quantization. The results of our case study indeed substantiate our hypothesis that the more constrained a problem is, the more the network can be optimized. On the constrained operational dataset, combining these optimization techniques allowed us to reduce the computational complexity with a factor of 349, as compared to only a factor 9.8 on the academic dataset. When running a benchmark on an Nvidia Jetson AGX Xavier, our fastest model runs more than 15 times faster than the original YoloV2 model, whilst increasing the accuracy by 5% Average Precision (AP).

Download Full-text

Research on a novel data-driven aging estimation method for battery systems in real-world electric vehicles

Advances in Mechanical Engineering ◽

10.1177/16878140211027735 ◽

2021 ◽

Vol 13 (7) ◽

pp. 168781402110277

Author(s):

Yankai Hou ◽

Zhaosheng Zhang ◽

Peng Liu ◽

Chunbao Song ◽

Zhenpo Wang

Keyword(s):

Electric Vehicles ◽

Real World ◽

Regression Models ◽

Estimation Method ◽

Recursive Least Squares ◽

Data Driven ◽

Accurate Estimation ◽

Support Vector ◽

Battery Degradation ◽

Operational Data

Accurate estimation of the degree of battery aging is essential to ensure safe operation of electric vehicles. In this paper, using real-world vehicles and their operational data, a battery aging estimation method is proposed based on a dual-polarization equivalent circuit (DPEC) model and multiple data-driven models. The DPEC model and the forgetting factor recursive least-squares method are used to determine the battery system’s ohmic internal resistance, with outliers being filtered using boxplots. Furthermore, eight common data-driven models are used to describe the relationship between battery degradation and the factors influencing this degradation, and these models are analyzed and compared in terms of both estimation accuracy and computational requirements. The results show that the gradient descent tree regression, XGBoost regression, and light GBM regression models are more accurate than the other methods, with root mean square errors of less than 6.9 mΩ. The AdaBoost and random forest regression models are regarded as alternative groups because of their relative instability. The linear regression, support vector machine regression, and k-nearest neighbor regression models are not recommended because of poor accuracy or excessively high computational requirements. This work can serve as a reference for subsequent battery degradation studies based on real-time operational data.

Download Full-text