The morphology of Fetal Electrocardiogram (FECG) plays an important role in the early diagnosis of fetal health condition. However, it is intractable to extract the clean morphology of FECG signals, which are usually contaminated by Maternal ECG (MECG) and various noises. To extract the clean morphology of FECG signals from noninvasive abdominal ECG records, a high-performance and high-efficient two-stage Slow-Fast Long Short Term Memory (SFLSTM) based architecture is proposed. The MECG elimination and the FECG enhancement are realized by the elaborately designed slow LSTM and fast LSTM to filter out the MECG and the residual noise components, respectively. Qualitative and quantitative experiments are conducted on the records from two public databases. The experimental results show that our proposed MECG elimination and FECG enhancement schemes improve the Signal-to-Noise Ratio (SNR) by 3.09 dB and 1.81 dB, respectively. The proposed fast LSTM reduces the amount of computation by approximately 50%, without any degradation in performance. Our proposed method may leverage the noninvasive FECG monitoring for the early detection of fetal heart diseases.

Since the use of finite element (FE) simulations for the dynamic analysis of railway beams on frictionally damped foundations are (i) very time consuming, and (ii) require advanced know-how and software that go beyond the available resources of typical civil engineering firms, this paper aims to demonstrate the potential of Artificial Neural Networks (ANN) to effectively predict the maximum displacements and the critical velocity in railway beams under moving loads. Four ANN-based models are proposed, one per load velocity range ([50, 175] ∪ [250, 300] m/s; ]175, 250[ m/s) and per displacement type (upward or downward). Each model is function of two independent variables, a frictional parameter and the load velocity. Among all models and the 663 data points used, a maximum error of 5.4 % was obtained when comparing the ANN- and FE-based solutions. Whereas the latter involves an average computing time per data point of thousands of seconds, the former does not even need a millisecond. This study was an important step towards the development of more versatile (i.e., including other types of input variables) ANN-based models for the same type of problem.

Any algorithm that needs to “understand” information to be capable of taking “intelligent” decisions, needs to access a lifetime of memories and experience the world as an embodied consciousness. This paper emphasizes these concepts and proposes a few fundamental constructs that provide algorithms with the capability to understand the human world, build larger sets of cooperative machines and perform causal inferences without requiring human intervention.

This paper is presented at 28th International Joint Conference on Artificial Intelligence (IJCAI-19) 1st Wksp. Federated Machine Learning for User Privacy and Data Confidentiality (FML’19), Macau, August 2019.

Corrugated webs are used to increase the shear stability of steel webs of beam-like members and to eliminate the need of transverse stiffeners. Previously developed formulas for predicting the shear strength of trapezoidal corrugated steel webs, along with the corresponding theory, are summarized. An artificial neural network (ANN)-based model is proposed to estimate the shear strength of steel girders with a trapezoidal corrugated web, and under a concentrated load. 210 test results from previous published research were collected into a database according to relevant test specimen parameters in order to feed the simulated ANNs. Seven (geometrical and material) parameters were identified as input variables and the ultimate shear stress at failure was considered the output variable. The proposed ANN-based analytical model yielded maximum and mean relative errors of 0.0% for the 210 points from the database. Moreover, still based on those points, it was illustrated that the ANN-based model clearly outperforms the other existing analytical models, which yield mean errors larger than 13%.

Headed studs are commonly used as shear connectors to transfer longitudinal shear force at the interface between steel and concrete in composite structures (e.g., bridge decks). Code-based equations for predicting the shear capacity of headed studs are summarized. An artificial neural network (ANN)-based analytical model is proposed to estimate the shear capacity of headed steel studs. 234 push-out test results from previous published research were collected into a database in order to feed the simulated ANNs. Three parameters were identified as input variables for the prediction of the headed stud shear force at failure, namely the steel stud tensile strength and diameter, and the concrete (cylinder) compressive strength. The proposed ANN-based analytical model yielded, for all collected data, maximum and mean relative errors of 3.3 % and 0.6 %, respectively. Moreover, it was illustrated that, for that data, the neural network approach clearly outperforms the existing code-based equations, which yield mean errors greater than 13 %.

There has been much discussion about the impact and future of artificial intelligence (AI) in our lives and future generations. Many experts even believe that AI will “rule” the world. Artificial Neural Networks (ANN) have provided a convenient and often extremely accurate solution to problems within all fields, and can be seen as advanced general-purpose regression models that try to mimic the behavior of the human brain. The adoption and use of ANN-based methods in the Mechanistic-Empirical Pavement Design Guide is a clear sign of the successful use of neural nets in geomechanical and pavement systems. This work aims to provide an extensive and detailed state-of-the-art of the application of ANN models to pavement management, materials and design problems. Unlike former review articles published before 2014, this work is more descriptive and makes the review much more appealing to the reader by highlighting numerically and/or graphically the effectiveness and possible drawbacks of each ANN application.

The amount of randomness in a signal generated by physical or non-physical process can reveal important information about that process. In this direction, the present work proposes to use the disentropy of the autocorrelation function as a measure of randomness. Examples using noisy and chaotic signals, economic and biological data are shown.

Hybrid beamforming systems represent an efficient architectural solution to realize massive multiple-input multiple-output (MIMO) communication systems in the centimeter wave (cmW) and millimeter wave (mmW) region. These hybrid beamforming systems separate the beamforming process into a digital and analog beamforming network. The analog beamforming networks can be realized by different architectural solutions, which demand dedicated algorithms to determine the complex weighting factors in the digital and analog domain. To date, novel hybrid beamforming architectures and algorithms are solely compared in numerical simulations based on statistical channel models. These abstract channel models simplify the complicated electromagnetic propagation process, thereby not exactly reconstructing the wireless channel. Within this work, we present a measurement-based evaluation of hybrid beamforming algorithms and compare them with numerical results gained from a statistical path-based MIMO channel model. The results show that by adjustment of the channel model parameter the simulation achieves a good match with the measured maximum achievable spectral efficiencies.

Despite the fact that artificial intelligence boosted with data-driven methods (e.g., deep neural networks) has surpassed human-level performance in various tasks, its application to autonomous systems still faces fundamental challenges such as lack of interpretability, intensive need for data and lack of verifiability. In this overview paper, I overview some attempts to address these fundamental challenges by explaining, guiding and verifying autonomous systems, taking into account limited availability of simulated and real data, the expressivity of high-level knowledge representations and the uncertainties of the underlying model. Specifically, this paper covers learning high-level knowledge from data for interpretable autonomous systems, guiding autonomous systems with high-level knowledge, and verifying and controlling autonomous systems against high-level specifications.

In this paper we use TDA mapper alongside with deep convolutional neural networks in the classification of 7 major skin diseases. First we apply kepler mapper with neural network as one of its filter steps to classify the dataset HAM10000. Mapper visualizes the classification result by a simplicial complex, where neural network can not do this alone, but as a filter step neural network helps to classify data better. Furthermore we apply TDA mapper and persistent homology to understand the weights of layers of mobilenet network in different training epochs of HAM10000. Also we use persistent diagrams to visualize the results of analysis of layers of mobilenet network.

Optimizing a Three-Channel Sensor Spectral Sensitivity Using A Genetic AlgorithmDorukalp Durmus*Pennsylvania State University, University Park, PA, USA 16802*[email protected]: Previous spectral error estimation studies are focused only on daylight. Spectral sensitivity of three sensors are optimized for electric light sources using genetic algorithm, which resulted in reduced errors between actual and estimated spectra.OCIS codes: 120.0280, 280.4788.1. IntroductionA multi-channel spectrum imaging system enables accurate spectral measurement across changes in illumination and ensures color matches for all observer types [1]. Multi-channel imaging is important for areas that require high-end color reproduction and spectral data collection, such as artwork reproduction and conservation [2], archeology [3], telemedicine [4], agriculture [5], study of minerals and gems [6], and integrative lighting systems [7]. Research on multi-channel imaging systems also impacts filter design [8] and target analysis [9], where spectral mismatches are considered detrimental for the optical systems. However, previous studies comparing and evaluating the mismatches in spectral power distributions (SPDs) are daylight oriented [10–13]. Optical imaging systems that are aimed to detect spectra during night time (i.e., sky glow, ecological impacts of lighting) require spectral analysis of electric light sources [14]. Here, the optimal spectral sensitivity of a three-channel sensing system is described using electric and natural light sources (i.e., one standard illuminant and ten commercially available electric light sources).2. MethodsThe spectral properties of three theoretical sensors were optimized using a genetic algorithm (GA) to minimize the error between reconstructed (estimated) and actual light source spectra. A GA is a computational tool inspired by the natural selection [15], and it is widely used in engineering and lighting research to find optimal solutions for a given problem [16,17]. The spectral sensitivity of each sensor was generated using a Gaussian distribution and characterized by their peak wavelengths and bandwidths (i.e., the full width at half maximum (FWHM)).The differences between reconstructed (estimated) and measured spectrum were analyzed using spectral curve difference metrics. Root mean square error (RMSE) is a simple, but widely used, metric for spectral estimation evaluation [18,19]. In addition to RMSE, two other metrics (integrated irradiance error (IIE) [10] and goodness-of-fit (GFC) coefficient [11]) were also considered for spectral analysis. While RMSE and IIE range between 0 and 1 (a smaller value denotes smaller error), a spectrally accurate estimation requires a GFC > 0.995 (“acceptable” fit), a “good” spectral fit requires a GFC > 0.999, and GFC > 0.9999 is needed for an “excellent” fit [11, 12]. Instead of a mean absolute average, the root-mean-square of three metrics was used, which is found to be more sensitive to distance differences and more appropriate when the error distribution is expected to be Gaussian [20].3. Results and discussionThe optimal peak wavelength and bandwidth of the three sensors are λsens1 = 380 nm, FWHMsens1 = 160 nm, λsens2 = 563 nm, FWHMsens2 = 194 nm, λsens3 = 750 nm, FWHMsens3 = 166 nm. The resulting error for each light source and error measures are summarized in Table 1. The highest RMSE was found for daylight illuminant and the smallest error was recorded for low-pressure sodium. There was one “excellent” fit for GFC (LPS), eight “good” fits, and two “acceptable” fits. None of the light sources were below the “acceptable” level for GFC. The reconstructed spectra for tri-phosphor fluorescent and phosphor-coated LED with additional red peak performed the best according to IIE.The results obtained here are comparable to other spectral mismatch studies, where values for daylight ranged between IIE = 0.032 [10], GFC = 0.9900 [11], RMSE = 0.3715, GFC = 0.9997, IIE = 0.0133 [14], and GFC = 0.9985, IIE = 0.70 [13]. Although some of the GFC values in these previous studies are marginally better than results presented here, the RMSE and IIE scores found in previous studies are lower compared to data gained through GA.Table 1. Spectral properties of the reference light sources and the error between the estimated and measured spectra according to three spectral mismatch metrics.

In this paper, we present a novel method to expand a graph over a finite field into a larger one over one of the original field's subfields. This allows a number of different graph expansions for any given graph. These expansions can be used in various applications, and we focus specifically on the case of decoding NB-LDPC codes. Using the novel expanded graphs, it is possible to reduce decoding complexity of NB- LDPC codes significantly with minimal performance losses.

Modern mobile devices come equipped with heterogeneous multi-processor system-on-chip (MPSoC) consisting of different types of cores such as CPU and GPU to cater for performance requirement of different types of applications. These MPSoCs also supports dynamic voltage and frequency scaling (DVFS) to reduce dynamic power consumption, which retrospectively affects the peak temperature of the multi-processors on the chip as well. In this poster presentation, we introduce three novel approaches to DVFS in mobile MPSoCs such that the executing applications can meet the performance requirement while consuming less power, reducing peak temperature and improving thermal side-channel security and reliability in terms of lifespan of the device.

Subarray-based hybrid beamforming (SBHB) systems represent an efficient architectural solution to realize massive multiple-input multiple-output (MIMO) systems. Thereby, the beamforming is no longer realized exclusively in the digital domain, but in coordination with an analog beamforming network by connecting each digital channel to a dedicated number of antenna elements via phase shifters. Most channel estimation and beamforming algorithms for SBHB systems are based on beam steering approaches, demanding a precise calibration of the analog beamforming network. In this letter, the SBHB system is calibrated by over-the-air (OTA) scattering parameter measurements and two criteria to optimize the beam pattern are formulated exploiting the ambiguities of the phase shifters. Both optimization criteria are evaluated using a designed SBHB receiver system operating at 27.8 GHz.

Hyper spectral imaging (HSI) is a technique that is used to obtain the spectrum for each pixel in the image. It helps in finding objects and identifying materials etc. Such an identification is very difficult using other imaging techniques. It allows the researchers to investigate the documents without any physical contact. Nowadays detection of unequal Ink mismatch based on HSI has shown vast improvement in distinguishing the inks. Detection of unequal Ink mismatch is an unbalanced clustering problem. This paper used K-means Clustering for ink mismatch detection. K-means Clustering find same subgroups in the data based on Euclidean distance. This paper demonstrates performance in unequal Ink mismatch based on HSI.

The sudden development of the COVID-19 pandemic exposed the limitations in modern healthcare systems to handle public health emergencies. It is evident that adopting innovative technologies such as blockchain can help in effective planning operations and resource deployments. Blockchain technology can play an important role in the healthcare sector such as improved clinical trial data management by reducing delays in regulatory approvals, streamline the communication between diverse stakeholders of the supply chain etc. Moreover, the spread of misinformation has intensely increased during the outbreak and existing platforms lack the ability to validate the authenticity of data, causing people to panic and act irrationally. Thus, developing a blockchain-based tracking system is important to ensure that the information received by the public and government agencies are reliable and trustworthy. In this paper, we focus on blockchain abilities to track the COVID-19 data collected from various sources including news, healthcare professionals, researchers etc, verify and append them in a secure and trusted distributed ledger. Thus, we propose a generic framework using Ethereum smart contracts and oracles to track real-time data related to the number of new cases, deaths and recovered cases obtained from trusted sources. We present detailed algorithms that capture the interactions between stakeholders in the network. The smart contract code was developed and tested in Remix environment. We present the cost and security analysis incurred by the stakeholders and highlight the challenges and future directions of our work. Our work demonstrates that the proposed solution is economically feasible and ensures data integrity, security, transparency, data traceability among stakeholders.

For a ground moving target indication (GMTI) radar, the presence ofDoppler blind zone (DBZ) results in many short tracks with frequent label switching, which seriously deteriorates the tracking performance. When the DBZ masking is coupled with targets maneuvering, tracking multiple maneuvering targets hidden in the DBZ becomes very challenging, which is reflected in the fact that there is no public research on this issue. To overcome this complicated problem, we propose a practical and fully functional GMTI multi-maneuvering-target tracker based on the multiple model probability hypothesis density (MM-PHD) filter. Unlike the standard MM-PHD filter, the proposed tracker utilizes the Doppler information and incorporates the minimum detectable velocity (MDV) to suppress the DBZ masking. Furthermore, to cope with the problems of the fixed initiation and no label output of the standard MM-PHD filter, the resulting MM-PHD filter with the Doppler and MDV information is augmented with measurement-driven adaptive track initiation and track label propagation, which are necessary for a practical tracker and also required for evaluating the overall GMTI tracking performance. Finally, numerical examples show that the proposed tracker outperforms significantly the existing ones, thus verifying its effectiveness.