For more information visit: https://cemse.kaust.edu.sa/ee

Recent Submissions

  • Sustained Solar-Powered Electrocatalytic H2 Production by Seawater Splitting Using Two-Dimensional Vanadium Disulfide

    Gnanasekar, Paulraj; Eswaran, Mathan Kumar; Palanichamy, Gayathri; Ng, Tien Khee; Schwingenschlögl, Udo; Ooi, Boon S.; Kulandaivel, Jeganathan (ACS Sustainable Chemistry & Engineering, American Chemical Society (ACS), 2021-06-15) [Article]
    Robust and stable electrodes made from earth-abundant materials have gained widespread interest in large-scale electrocatalytic water splitting toward hydrogen energy technologies. In this study, the vanadium disulfide (VS2)/amorphous carbon (AC) heterostructure was employed as an electrode for direct seawater splitting. Two-dimensional VS2 nanoparticles were deposited on AC with a high degree of uniformity via a well-optimized one-step chemical vapor deposition approach. The VS2/AC heterostructure electrode was found to possess rich active sulfur sites, near-zero Gibbs free energy, a large surface area, and exceptional charge transfer toward the electrolyte, resulting in enhanced hydrogen evolution reaction (HER) performance with a low onset potential and low overpotential of 11 and 61 mV (vs reversible hydrogen electrode (RHE)), respectively. The electrode also sustained robust stability throughout the 50 h of chronoamperometry studies under acidic electrolyte conditions. Interestingly, the VS2/AC electrocatalyst accomplished an exceptional HER performance under natural seawater conditions in the absence of an external electrolyte with an onset potential of 56 mV vs RHE and attained η200 at an overpotential of 0.53 V vs RHE. In spite of this, the heterostructure exhibited superior stability over 21 days at a high current density of 250 mA/cm2 under both indoor and solar-powered outdoor conditions. Overall, this VS2/AC heterostructure may open a new pathway toward direct seawater splitting for long-term, stable, large-scale hydrogen generation.
  • Training Graph Neural Networks with 1000 Layers

    Li, Guohao; Müller, Matthias; Ghanem, Bernard; Koltun, Vladlen (arXiv, 2021-06-14) [Preprint]
    Deep graph neural networks (GNNs) have achieved excellent results on various tasks on increasingly large graph datasets with millions of nodes and edges. However, memory complexity has become a major obstacle when training deep GNNs for practical applications due to the immense number of nodes, edges, and intermediate activations. To improve the scalability of GNNs, prior works propose smart graph sampling or partitioning strategies to train GNNs with a smaller set of nodes or sub-graphs. In this work, we study reversible connections, group convolutions, weight tying, and equilibrium models to advance the memory and parameter efficiency of GNNs. We find that reversible connections in combination with deep network architectures enable the training of overparameterized GNNs that significantly outperform existing methods on multiple datasets. Our models RevGNN-Deep (1001 layers with 80 channels each) and RevGNN-Wide (448 layers with 224 channels each) were both trained on a single commodity GPU and achieve an ROC-AUC of $87.74 \pm 0.13$ and $88.14 \pm 0.15$ on the ogbn-proteins dataset. To the best of our knowledge, RevGNN-Deep is the deepest GNN in the literature by one order of magnitude. Please visit our project website https://www.deepgcns.org/arch/gnn1000 for more information.
  • Integration of Droplet Microfluidic Tools for Single-cell Functional Metagenomics: An Engineering Head Start

    Conchouso Gonzalez, David; Alma’abadi, Amani D.; Behzad, Hayedeh; Alarawi, Mohammed; Hosokawa, Masahito; Nishikawa, Yohei; Takeyama, Haruko; Mineta, Katsuhiko; Gojobori, Takashi (Institute of Electrical and Electronics Engineers (IEEE), 2021-06-11) [Preprint]
    <p>Droplet microfluidics techniques have shown promising results to study single-cells at high throughput. However, their adoption in laboratories studying “-omics” sciences is still irrelevant because of the field’s complex and multidisciplinary nature. To facilitate their use, here we provide engineering details and organized protocols for integrating three droplet-based microfluidic technologies into the metagenomic pipeline to enable functional screening of bioproducts at high throughput. First, a device encapsulating single-cells in droplets at a rate of ~ 250 Hz is described considering droplet size and cell growth. Then, we expand on previously reported fluorescent activated droplet sorting (FADS) systems to integrate the use of 4 independent fluorescence-exciting lasers (e.g., 405, 488, 561, 637 nm) in a single platform to make it compatible with different fluorescence-emitting biosensors. For this sorter, both hardware and software are provided and optimized for effortlessly sorting droplets at 60 Hz. Then, a passive droplet merger was also integrated into our method to enable adding new reagents to already made droplets at a rate of 200 Hz. Finally, we provide an optimized recipe for manufacturing these chips using silicon dry-etching tools. Because of the overall integration and the technical details presented here, our approach allows biologists to quickly use microfluidic technologies and achieve both single-cell resolution and high-throughput (> 50,000 cells/day) capabilities to mining and bioprospecting metagenomic data.</p>
  • Snapshot Space–Time Holographic 3D Particle Tracking Velocimetry

    Chen, Ni; Wang, Congli; Heidrich, Wolfgang (Laser & Photonics Reviews, Wiley, 2021-06-10) [Article]
    Digital inline holography is an amazingly simple and effective approach for 3D imaging, to which particle tracking velocimetry is of particular interest. Conventional digital holographic particle tracking velocimetry techniques are computationally separated in particle and flow reconstruction, plus the expensive computations. Usually, the particle volumes are recovered first, from which fluid flows are computed. Without iterative reconstructions, This sequential space–time process lacks accuracy. This paper presents a joint optimization framework for digital holographic particle tracking velocimetry: particle volumes and fluid flows are reconstructed jointly in a higher space–time dimension, enabling faster convergence and better reconstruction quality of both fluid flow and particle volumes within a few minutes on modern GPUs. Synthetic and experimental results are presented to show the efficiency of the proposed technique.
  • Magnetic sensors – A review and recent technologies

    Khan, Mohammed Asadullah; Sun, Jian; Li, Bodong; Przybysz, Alexander; Kosel, Jürgen (Engineering Research Express, IOP Publishing, 2021-06-04) [Article]
    Magnetic field sensors are an integral part of many industrial and biomedical applications, and their utilization continues to grow at a high rate. The development is driven both by new use cases and demand like internet of things as well as by new technologies and capabilities like flexible and stretchable devices. Magnetic field sensors exploit different physical principles for their operation, resulting in different specifications with respect to sensitivity, linearity, field range, power consumption, costs etc. In this review, we will focus on solid state magnetic field sensors that enable miniaturization and are suitable for integrated approaches to satisfy the needs of growing application areas like biosensors, ubiquitous sensor networks, wearables, smart things etc. Such applications require a high sensitivity, low power consumption, flexible substrates and miniaturization. Hence, the sensor types covered in this review are Hall Effect, Giant Magnetoresistance, Tunnel Magnetoresistance, Anisotropic Magnetoresistance and Giant Magnetoimpedance.
  • Rapid Point-of-Care COVID-19 Diagnosis with a Gold-Nanoarchitecture-Assisted Laser-Scribed Graphene Biosensor

    Beduk, Tutku; Beduk, Duygu; De Oliveira Filho, Jose; Zihnioglu, Figen; Cicek, Candan; Sertoz, Ruchan; Arda, Bilgin; Goksel, Tuncay; Turhan, Kutsal; Salama, Khaled N.; Timur, Suna (Analytical Chemistry, American Chemical Society (ACS), 2021-06-03) [Article]
    The global pandemic caused by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) virus has revealed the urgent need for accurate, rapid, and affordable diagnostic tests for epidemic understanding and management by monitoring the population worldwide. Though current diagnostic methods including real-time polymerase chain reaction (RT-PCR) provide sensitive detection of SARS-CoV-2, they require relatively long processing time, equipped laboratory facilities, and highly skilled personnel. Laser-scribed graphene (LSG)-based biosensing platforms have gained enormous attention as miniaturized electrochemical systems, holding an enormous potential as point-of-care (POC) diagnostic tools. We describe here a miniaturized LSG-based electrochemical sensing scheme for coronavirus disease 2019 (COVID-19) diagnosis combined with three-dimensional (3D) gold nanostructures. This electrode was modified with the SARS-CoV-2 spike protein antibody following the proper surface modifications proved by X-ray photoelectron spectroscopy (XPS) and scanning electron microscopy (SEM) characterizations as well as electrochemical techniques. The system was integrated into a handheld POC detection system operated using a custom smartphone application, providing a user-friendly diagnostic platform due to its ease of operation, accessibility, and systematic data management. The analytical features of the electrochemical immunoassay were evaluated using the standard solution of S-protein in the range of 5.0-500 ng/mL with a detection limit of 2.9 ng/mL. A clinical study was carried out on 23 patient blood serum samples with successful COVID-19 diagnosis, compared to the commercial RT-PCR, antibody blood test, and enzyme-linked immunosorbent assay (ELISA) IgG and IgA test results. Our test provides faster results compared to commercial diagnostic tools and offers a promising alternative solution for next-generation POC applications.
  • APES: Audiovisual Person Search in Untrimmed Video

    Alcazar, Juan Leon; Mai, Long; Perazzi, Federico; Lee, Joon-Young; Arbelaez, Pablo; Ghanem, Bernard; Heilbron, Fabian Caba (arXiv, 2021-06-03) [Preprint]
    Humans are arguably one of the most important subjects in video streams, many real-world applications such as video summarization or video editing workflows often require the automatic search and retrieval of a person of interest. Despite tremendous efforts in the person reidentification and retrieval domains, few works have developed audiovisual search strategies. In this paper, we present the Audiovisual Person Search dataset (APES), a new dataset composed of untrimmed videos whose audio (voices) and visual (faces) streams are densely annotated. APES contains over 1.9K identities labeled along 36 hours of video, making it the largest dataset available for untrimmed audiovisual person search. A key property of APES is that it includes dense temporal annotations that link faces to speech segments of the same identity. To showcase the potential of our new dataset, we propose an audiovisual baseline and benchmark for person retrieval. Our study shows that modeling audiovisual cues benefits the recognition of people's identities. To enable reproducibility and promote future research, the dataset annotations and baseline code are available at: https://github.com/fuankarion/audiovisual-person-search
  • Cloud-Enabled High-Altitude Platform Systems: Challenges and Opportunities

    Mershad, Khaleel; Dahrouj, Hayssam; Sarieddeen, Hadi; Shihada, Basem; Al-Naffouri, Tareq Y.; Alouini, Mohamed-Slim (arXiv, 2021-06-03) [Preprint]
    Augmenting ground-level communications with flying networks, such as the high-altitude platform system (HAPS), is among the major innovative initiatives of the next generation of wireless systems (6G). Given HAPS quasi-static positioning at the stratosphere, HAPS-to-ground and HAPS-to-air connectivity frameworks are expected to be prolific in terms of data acquisition and computing, especially given the mild weather and quasi-constant wind speed characteristics of the stratospheric layer. This paper explores the opportunities stemming from the realization of cloud-enabled HAPS in the context of telecommunications applications and services. The paper first advocates for the potential physical advantages of deploying HAPS as flying data-centers, also known as super-macro base stations. The paper then presents the merits that can be achieved by integrating various cloud services within the HAPS, and the corresponding cloud-type applications that would utilize the HAPS for enhancing the quality, range, and types of offered services. The paper further sheds light on the challenges that need to be addressed for realizing practical cloud-enabled HAPS, mainly, those related to the high energy, processing power, quality of service (QoS), and security considerations. Finally, the paper discusses some open issues on the topic, namely, HAPS mobility and message routing, HAPS security via blockchain and machine learning, artificial intelligence-based resource allocation in cloud-enabled HAPS, and integration with vertical heterogeneous networks.
  • A Survey on Integrated Access and Backhaul Networks

    Zhang, Yongqiang; Kishk, Mustafa Abdelsalam; Alouini, Mohamed-Slim (Frontiers in Communications and Networks, Frontiers Media SA, 2021-06-01) [Article]
    Benefiting from the usage of the high-frequency band, utilizing part of the large available bandwidth for wireless backhauling is feasible without considerable performance sacrifice. In this context, integrated access and backhaul (IAB) has been proposed by the Third Generation Partnership Project (3GPP) to reduce the expenses related to the deployment of fiber optics for 5G and beyond networks. In this paper, first, a brief introduction of IAB based on the 3GPP release is presented. Then, the existing research on IAB networks based on 3GPP specifications and possible non-3GPP research extensions are surveyed. The research on non-3GPP extensions includes the integration of IAB networks with other advanced techniques beyond the currently defined protocol stacks, such as the integration of IAB to cache-enabled, optical communication transport, and non-terrestrial networks. Finally, the challenges and opportunities related to the development and commercialization of the IAB networks are discussed.
  • Computational Wavefront Sensing: Theory, Practice, and Applications

    Wang, Congli (2021-06) [Dissertation]
    Advisor: Heidrich, Wolfgang
    Committee members: Heidrich, Wolfgang; Ghanem, Bernard; Wonka, Peter; Waller, Laura
    Wavefront sensing is a fundamental problem in applied optics. Wavefront sensors that work in a deterministic manner are of particular interest. Initialized with a unified theory for classical wavefront sensors, this dissertation discusses relevant properties of wavefront sensor designs. Based on which, a new wavefront sensor, termed Coded Wavefront Sen- sor, is proposed to leverage the advantages of the analysis, especially the lateral wavefront resolution. A prototype was built to demonstrate this new wavefront sensor. Given that, two specific applications are demonstrated: megapixel adaptive optics and simultaneous intensity and phase imaging. Combined with a spatial light modulator, a hard- ware deconvolution approach is demonstrated for computational cameras via a high resolu- tion adaptive optics system. By simply switching the normal image sensor with the proposed one, as well as slight change of illumination, a bright field microscope can be configured to a simultaneous intensity and phase microscope. These show the broad application range of the proposed computational wavefront sensing approach. Lastly, this dissertation proposes the idea of differentiable optics for wavefront engineer- ing and lens metrology. By making use of automatic differentiation, a physically-correct differentiable ray tracing engine is built, with its potentials being illustrated via several chal- lenging applications in optical design and metrology.
  • Green Tethered UAVs for EMF-Aware Cellular Networks

    Lou, Zhengying; Elzanaty, Ahmed; Alouini, Mohamed-Slim (arXiv, 2021-05-29) [Preprint]
    A prevalent theory circulating among the non-scientific community is that the intensive deployment of base stations over the territory significantly increases the level of electromagnetic field (EMF) exposure and affects population health. To alleviate this concern, in this work, we propose a network architecture that introduces tethered unmanned aerial vehicles (TUAVs) carrying green antennas to minimize the EMF exposure while guaranteeing a high data rate for users. In particular, each TUAV can attach itself to one of the possible ground stations at the top of some buildings. The location of the TUAVs, transmit power of user equipment and association policy are optimized to minimize the EMF exposure. Unfortunately, the problem turns out to be mixed-integer non-linear programming (MINLP), which is non-deterministic polynomial-time (NP) hard. We propose an efficient low-complexity algorithm composed of three submodules. Firstly, we propose an algorithm based on the greedy principle to determine the optimal association matrix between the users and base stations. Then, we offer two approaches, a modified K-mean and shrink and realign (SR) process, to associate each TUAV with a ground station. Finally, we put forward two algorithms based on the golden search and SR process to adjust the TUAV's position within the hovering area over the building. After that, we consider the dual problem that maximizes the sum rate while keeping the exposure below a predefined value, such as the level enforced by the regulation. Next, we perform extensive simulations to show the effectiveness of the proposed TUAVs to reduce the exposure compared to various architectures. Eventually, we show that TUAVs with green antennas can effectively mitigate the EMF exposure by more than 20% compared to fixed green small cells while achieving a higher data rate.
  • Reply to “Comment on ‘Scattering Cancellation-Based Cloaking for the Maxwell-Cattaneo Heat Waves’”

    Farhat, Mohamed; Guenneau, Sebastien; Chen, P.-Y.; Alù, A.; Salama, Khaled N. (Physical Review Applied, American Physical Society (APS), 2021-05-28) [Article]
    Comment [1] points out possible inconsistencies in the notations of our paper [2] and, based on these remarks, it questions the validity of our conclusions. In this Reply, we demonstrate the general validity of all conclusions in Ref. [2], and we take the opportunity to clarify our notation and our results and to discuss their domain of validity.
  • A Microneedles Balloon Catheter for Endovascular Drug Delivery

    Moussi, Khalil; Haneef, Ali A.; Alsiary, Rawiah A.; Diallo, Elhadj; Boone, Marijn Antoine; Abu-Araki, Huda; Al-Radi, Osman O.; Kosel, Jürgen (Advanced Materials Technologies, Wiley, 2021-05-28) [Article]
    Disorders of the inner parts of blood vessels have been significant triggers of cardiovascular diseases (CVDs). Different interventional methods have been employed, from complex surgeries to balloon angioplasty techniques to open the narrowed blood vessels. However, CVDs continue to be the lead cause of death in the world. Delivering a therapeutic agent directly to the inner wall of affected blood vessels can be a transformative step toward a better treatment option. To open the door for such an approach, a catheter delivery system is developed based on a conventional balloon catheter where a fluidic channel and microneedles (MNs) are integrated on top of it. This enables precise and localized delivery of therapeutics directly into vessel walls. Customizable MNs are fabricated using a high-resolution 3D printing technique where MN's height ranges from 100 to 350 µm. The MNs penetration into a synthetic vascular model is investigated with a computerized tomography scan. Ex vivo tests on rabbit aorta confirm the MN-upgraded balloon catheter's performance on real tissue. Delivery of fluorescent dye is accomplished by injecting it through the fluidic channel and MNs into the aortic tissue. The dye is observed at up to 180 µm of depth, confirming site-specific endovascular delivery.
  • Exploitation of Field Drilling Data with an Innovative Drilling Simulator: Highly Effective Simulation of Rotating and Sliding Mode

    Koulidis, Alexis; Kelessidis, Vassilios; Ahmed, Shehab (SPE, 2021-05-25) [Conference Paper]
    Drilling challenging wells requires a combination of drilling analytics and comprehensive simulation to prevent poor drilling performance and avoid drilling issues for the upcoming drilling campaign. This work focuses on the capabilities of a drilling simulator that can simulate the directional drilling process with the use of actual field data for the training of students and professionals. This paper presents the results of simulating both rotating and sliding modes and successfully matching the rate of penetration and the trajectory of an S-type well. Monitored drilling data from the well were used to simulate the drilling process. These included weight on bit, revolutions per minute, flow rate, bit type, inclination and drilling fluid properties. The well was an S-type well with maximum inclination of 16 degrees. There were continuous variations from rotating to sliding mode, and the challenge was approached by classifying drilling data into intervals of 20 feet to obtain an appropriate resolution and efficient simulation. The simulator requires formation strength, pore and fracture pressures, and details of well lithology, thus simulating the actual drilling environment. The uniaxial compressive strength of the rock layer is calculated from p–wave velocity data from an offset field. Rock drillability is finally estimated as a function of the rock properties of the drilled layer, bit type and the values of the drilling parameters. It is then converted to rate of penetration and matched to actual data. Changes in the drilling parameters were followed as per the field data. The simulator reproduces the drilling process in real-time and allows the driller to make instantaneous changes to all drilling parameters. The simulator provides the rate of penetration, torque, standpipe pressure, and trajectory as output. This enables the user to have on-the-fly interference with the drilling process and allows him/her to modify any of the important drilling parameters. Thus, the user can determine the effect of such changes on the effectiveness of drilling, which can lead to effective drilling optimization. Certain intervals were investigated independently to give a more detailed analysis of the simulation outcome. Additional drilling data such as hook load and standpipe pressure were analyzed to determine and evaluate the drilling performance of a particular interval and to consider them in the optimization process. The resulting rate of penetration and well trajectory simulation results show an excellent match with field data. The simulation illustrates the continuous change between rotating and sliding mode as well as the accurate synchronous matching of the rate of penetration and trajectory. The results prove that the simulator is an excellent tool for students and professionals to simulate the drilling process prior to actual drilling of the next inclined well.
  • Aortic blood pressure estimation: A hybrid machine-learning and cross-relation approach

    Magbool, Ahmed; Bahloul, Mohamed; Ballal, Tarig; Al-Naffouri, Tareq Y.; Laleg-Kirati, Taous-Meriem (Biomedical Signal Processing and Control, Elsevier BV, 2021-05-23) [Article]
    Aortic blood pressure is a vital signal that provides valuable medical information about cardiovascular health condition. Noninvasive measurement of this signal is very challenging, which motivates several researchers to develop mathematical approaches over the years to estimate the aortic pressure from peripheral measurements. Most of these approaches are limited in their performance as they fail to recover important features of the blood pressure signal. To overcome this issue, we investigate the application of machine-learning methods to estimate the aortic blood pressure from peripheral signals. In the absence of reasonably large datasets, we rely on pre-validated virtual databases to train our machine-learning models. To avoid model bias due to the lack of diversity and variability in these databases, we propose a hybrid approach that combines machine-learning models with the cross-relation blind estimation approach. On top of that, a sparse representation, coupled with a dictionary-learning approach, is employed to emphasize the characteristics of the aortic pressure signals and generate more meaningful outputs. Our results show that the proposed hybrid approach offers a reduction in the root-mean-squared error compared to pure machine-learning models and improvement compared to the cross-relation method. The proposed approach also shows a noticeable potency in capturing fine features of the aortic blood pressure signal.
  • 606-nm InGaN amber micro-light-emitting diodes with an on-wafer external quantum efficiency of 0.56%

    Zhuang, Zhe; Iida, Daisuke; Velazquez-Rizo, Martin; Ohkawa, Kazuhiro (IEEE Electron Device Letters, Institute of Electrical and Electronics Engineers (IEEE), 2021-05-17) [Article]
    We demonstrated amber InGaN 47 × 47 μ2 micro-light-emitting diodes (μLEDs) with the peak wavelength of 606 nm and full-width at maximum (FWHM) of 50 nm at 20 A/cm2. The amber μLEDs exhibited a 33-nm blue-shift of the peak wavelength and obtain broader FWHMs to approximately 56 nm at 5 to 100 A/cm2. The peak on-wafer external quantum efficiency was 0.56% at 20 A/cm2. The characteristic temperature was 50.80 K at 20 to 60 A/cm2 but increased to 120.140 K at 80 to 100 A/cm2. The strong increase in the characteristic temperature from 60 to 80 A/cm2 could mainly be attributed to the saturation of the Shockley-Read-Hall non-radiative recombination at high current densities.
  • Retrofitting FSO Systems in Existing RF Infrastructure: A Non-Zero Sum Game Technology

    Trichili, Abderrahmen; Ragheb, Amr; Briantcev, Dmitrii; Esmail, Maged A.; Altamimi, Majid; Ashry, Islam; Ooi, Boon S.; Alshebeili, Saleh; Alouini, Mohamed-Slim (arXiv, 2021-05-16) [Preprint]
    Progress in optical wireless communication (OWC) has unleashed the potential to transmit data in an ultra-fast manner without incurring large investments and bulk infrastructure. OWC includes wireless data transmissions in three optical sub-bands; ultraviolet, visible, and infrared. This paper discusses installing infrared OWC, known as free space optics (FSO), systems on top of installed radio frequency (RF) networks for outdoor applications to benefit from the reliability of RF links and the unlicensed broad optical spectrum, and the large data rates carried by laser beams propagating in free space. We equally review commercially available solutions and the hardware requirements for RF and FSO technology co-existence. The potential of hybrid RF/FSO for space communication is further discussed. Finally, open problems and future research directions are presented.
  • An Adaptive Regularization Approach to Portfolio Optimization

    Ballal, Tarig; Abdelrahman, Abdelrahman S.; Muqaibel, Ali H.; Al-Naffouri, Tareq Y. (IEEE, 2021-05-13) [Conference Paper]
    We address the portfolio optimization problem using the global minimum variance portfolio (GMVP). The GMVP gives the weights as a function of the inverse of the covariance matrix (CM) of the stock net returns in a closed-form. The matrix inversion operation usually intensifies the impact of noise when the matrix is ill-conditioned, which often happens when the sample covariance matrix (SCM) is used. A regularized sample covariance matrix (RSCM) is usually used to alleviate the problem. In this work, we address the regularization issue from a different perspective. We manipulate the expression of the GMVP weights to convert it to an inner product of two vectors; then, we focus on obtaining accurate estimations of these vectors. We show that this approach results in a formula similar to those of the RSCM based methods, yet with a different interpretation of the regularization parameter’s role. In the proposed approach, the regularization parameter is adjusted adaptively based on the current stock returns, which results in improved performance and enhanced robustness to noise. Our results demonstrate that, with proper regularization parameter tuning, the proposed adaptively regularized GMVP outperforms state-of-the-art RSCM methods in different test scenarios.
  • Hardware Acceleration of High Sensitivity Power-Aware Epileptic Seizure Detection System Using Dynamic Partial Reconfiguration

    Elhosary, Heba; Zakhari, Michael H.; Elgammal, Mohamed A.; Kelany, Khaled A. Helal; Ghany, Mohamed A. Abd El; Salama, Khaled N.; Mostafa, Hassan (IEEE Access, IEEE, 2021-05-11) [Article]
    In this paper, a high-sensitivity low-cost power-aware Support Vector Machine (SVM) training and classification based system, is hardware implemented for a neural seizure detection application. The training accelerator algorithm, adopted in this work, is the sequential minimal optimization (SMO). System blocks are implemented to achieve the best trade-off between sensitivity and the consumption of area and power. The proposed seizure detection system achieves 98.38% sensitivity when tested with the implemented linear kernel classifier. The system is implemented on different platforms: such as Field Programmable Gate Array (FPGA) Xilinx Virtex-7 board and Application Specific Integrated Circuit (ASIC) using hardware-calibrated UMC 65nm CMOS technology. A power consumption evaluation is performed on both the ASIC and FPGA platforms showing that the ASIC power consumption is lower by at least 65% when compared with the FPGA counterpart. A power-aware system is implemented with FPGAs by the adoption of the Dynamic Partial Reconfiguration (DPR) technique that allows the dynamic operation of the system based on power level available to the system at the expense of degradation of the system accuracy. The proposed system exploits the advantages of DPR technology in FPGAs to switch between two proposed designs providing a decrease of 64% in power consumption.
  • SCTN: Sparse Convolution-Transformer Network for Scene Flow Estimation

    Li, Bing; Zheng, Cheng; Giancola, Silvio; Ghanem, Bernard (arXiv, 2021-05-10) [Preprint]
    We propose a novel scene flow estimation approach to capture and infer 3D motions from point clouds. Estimating 3D motions for point clouds is challenging, since a point cloud is unordered and its density is significantly non-uniform. Such unstructured data poses difficulties in matching corresponding points between point clouds, leading to inaccurate flow estimation. We propose a novel architecture named Sparse Convolution-Transformer Network (SCTN) that equips the sparse convolution with the transformer. Specifically, by leveraging the sparse convolution, SCTN transfers irregular point cloud into locally consistent flow features for estimating continuous and consistent motions within an object/local object part. We further propose to explicitly learn point relations using a point transformer module, different from exiting methods. We show that the learned relation-based contextual information is rich and helpful for matching corresponding points, benefiting scene flow estimation. In addition, a novel loss function is proposed to adaptively encourage flow consistency according to feature similarity. Extensive experiments demonstrate that our proposed approach achieves a new state of the art in scene flow estimation. Our approach achieves an error of 0.038 and 0.037 (EPE3D) on FlyingThings3D and KITTI Scene Flow respectively, which significantly outperforms previous methods by large margins.

View more