The paper built on First Impression Challenge from Chalearn V2 Workshop on Explainable Computer Vision Multimedia and Job Candidate Screening Competition CVPR17 by focusing solely on Textual Input in contrast to other Challenge’s participants who considered video or audio modalities. Therefore, the paper aims to develop a new deep learning architecture capable of predicting human personality traits and job interview from the video transcripts. Several feature representations that involve statistical and deep learning have contrasted. Our approach achieved the best score when text modality alone were employed, yielding an average of 89% score in human personality traits and 89.10% value for job interview. The research results will help companies and other organization studying human personality to assess a human personality using a minimum textual resources from the job candidates

There are demographic biases in current models used for facial recognition (FR). Our Balanced Faces In the Wild (BFW) dataset serves as a proxy to measure bias across ethnicity and gender subgroups, allowing one to characterize FR performances per subgroup. We show performances are non-optimal when a single score threshold is used to determine whether sample pairs are genuine or imposter. Across subgroups, performance ratings vary from the reported across the entire dataset. Thus, claims of specific error rates only hold true for populations matching that of the validation data. We mitigate the imbalanced performances using a novel domain adaptation learning scheme on the facial features extracted using state-of-the-art. Not only does this technique balance performance, but it also boosts the overall performance. A benefit of the proposed is to preserve identity information in facial features while removing demographic knowledge in the lower dimensional features. The removal of demographic knowledge prevents future potential biases from being injected into decision-making. This removal satisfies privacy concerns. We explore why this works qualitatively; we also show quantitatively that subgroup classifiers can no longer learn from the features mapped by the proposed.

A document by Eiman Kanjo . Click on the document to view its contents.

In complex application wherein the signal propagating through free space is subject to multipath interference due to scatter by line-of-sight and non-line-of-sight objects in the propagation channel. The aims is to identify scatter centers in the propagation channel and characterize them based on their subjective characteristics, interpreted based on machine learning algorithm operations. Data-driven models are employed, replacing the traditional analytical approaches, in order to profile the scatter centers as either of absorbing or reflecting types based on the manner in which the signals are affected. A typical multistatic detection scenario is reconstructed under controlled laboratory conditions in order to create spatially independent data sets, while operating in the C-band frequency. The outcomes of this study are then applied to identify the scatter centers based on the distinct signatures they register in the experimental data set. As a converse argument, the process of antenna pattern estimation can now be performed free of an anechoic chamber setup, which is time and cost insensitive. A greater relevance shall be in the context of mid-band 5G-NR cellular communication systems that need to optimize the distributed antenna location attributes on time and cost constrained scales before attempting a large-scale deployment.

Machine learning techniques lend themselves as promising decision-making and analytic tools in a wide range of applications. Different ML algorithms have various hyper-parameters. In order to tailor an ML model towards a specific application working at its best, its hyper-parameters should be tuned. Tuning the hyper-parameters directly affects the performance. However, for large-scale search spaces, efficiently exploring the ample number of combinations of hyper-parameters is computationally expensive. Many of the automated hyper-parameter tuning techniques suffer from low convergence rates and high experimental time complexities. In this paper, we propose HyP-ABC, an automatic innovative hybrid hyper-parameter optimization algorithm using the modified artificial bee colony approach, to measure the classification accuracy of three ML algorithms: random forest, extreme gradient boosting, and support vector machine. In order to ensure the robustness of the proposed method, the algorithm takes a wide range of feasible hyper-parameter values and is tested using a real-world educational dataset. Experimental results show that HyP-ABC is competitive with state-of-the-art techniques. Also, it has fewer hyper-parameters to be tuned than other population-based algorithms, making it worthwhile for real-world HPO problems.

In classical binary statistical pattern recognition optimality in Neyman-Pearson sense, achieved by a (log) likelihood ratio based classifier, is often desirable. A drawback of a Neyman-Pearson optimal classifier is that it requires full knowledge of the (quotient of the) class-conditional probability densities of the input data, which is often unrealistic. The design of neural net classifiers is data driven, meaning that no explicit use is made of the class-conditional probability densities of the input data. In this paper a proof is presented that a neural net can also be trained to approximate a log-likelihood ratio and be used as a Neyman-Pearson optimal, prior-independent classifier. Properties of the approximation of the log-likelihood ratio are discussed. Examples of neural nets trained on synthetic data with known log-likelihood ratios as ground truth illustrate the results.

Ultra high-throughput satellite systems are expected to play an essential role in future beyond 5G and 6G networks. These systems must remain as flexible as possible to adapt to heterogeneous traffic demands, while also delivering the highest possible rate for dedicated services. Satellites flexible payloads are increasingly employing wideband output multiplexers. In this context, it is now more important than ever to evaluate frequency-dependent degradations on multicarrier signals. In particular, it is critical to characterize the distortions entailed by the output multiplexers filters. In this paper, models are presented and novel formulas are derived to determine the carrier-to-interference ratio resulting from these distortions. Derivations are oriented towards the applicability of either high-accuracy (e.g., for link budget) or low-complexity calculations (e.g., for real-time carrier allocation). The influence of key parameters such as the optimal decision instant, symbol rate and roll-off factor is thoroughly analyzed. Furthermore, formulas are evaluated in a practical scenario: the dynamic carrier allocation optimization. They are combined with efficient optimization algorithms to obtain the best performance based on user fairness. Relevant metrics such as accuracy, complexity and allocation gain are also investigated. In the end, the application of the proposed formulas and algorithms leads to a significant allocation gain that is increasing with the number of carriers. The feasibility of real-time dynamic carrier allocation to further increase the capacity of the next generation of satellite systems is emphasized.

A document by Yasmina Zaky . Click on the document to view its contents.

Hamidreza. Amiri

Geometry-based point cloud compression (G-PCC) has been rapidly evolving in the context of international standards. Despite the inherent scalability of octree-based geometry description, current G-PCC attribute compression techniques prevent full scalability for compressed point clouds. In this paper, we present a solution to add scalability to attributes compressed using the region-adaptive hierarchical transform (RAHT), enabling the reconstruction of the point cloud using only a portion of the original bitstream. Without the full geometry information, one cannot compute the weights in which the RAHT relies on to calculate its coefficients for further levels of detail. In order to overcome this problem, we propose a linear relationship approximation relating the downsampled point cloud to the truncated inverse RAHT coefficients at that same level. The linear relationship parameters are sent as side information. After truncating the bitstream at a point corresponding to a given octree level, we can, then, recreate the attributes at that level. Tests were carried out and results attest the good approximation quality of the proposed technique.

Currently, remote monitoring systems have evolvedto respond to the peculiar needs in the healthcare industry, suchas Cardiovascular diseases, which is an essential foundation in themodern concept of the smart city. We propose a system to monitorpatient current health conditions as a healthcare system basedon the widely spread available low-cost technologies, namely,GSM and Arduino platform. Statistics show that hypertensiveheart disease and blood pressure are risk factors for the highdeath rate. To decrease it, preventive measures should be appliedproviding a health monitoring system to save the patient life atan acceptable time. The objective of this paper is to developa prototype of a low-cost system, that is capable of performingremote monitoring of a patient’s heart rate and body temperaturein real-time and stores the data on the memory card using low-cost hardware such as Arduino, and also, using the servicesprovided by the GSM network, to enable communication witha health specialists cell phone upon request or if there areanomalies in the values measured by the sensors. The studyfocuses on heartbeat rate using the Difference Operation Method(DOM), described by Yeh & Wang, and body temperature, thus,in case of emergency, a short call is done, and then an SMS issent to the health specialist’s mobile containing measured values.Using this algorithm and after some tests using data from theMIT BIH Database distribution, we manage to find the bestsample frequency and eliminate the distortion of the signal. sample frequency and eliminate the distortion of the signal.

The proposed paper addresses how Support Vector Data Description (SVDD) can be used to detect safety regions with zero statistical error. It provides a detailed methodology for the applicability of SVDD in real-life applications, such as Vehicle Platooning, by addressing common machine learning problems such as parameter tuning and handling large data sets. Also, intelligible analytics for knowledge extraction with rules is presented: it is targeted to understand safety regions of system parameters. Results are shown by feeding data through simulation to the train of different rule extraction mechanisms.

Quantum entanglement is a primary feature of quantum mechanics and probably results from the quantum state's conservation and the quantum state's transition with the matter wave's phase velocity at the fast-than-light speed. The quantum state transition of entangled particles proceeds with the phase velocity, while the observer measures the process with the electromagnetic or the light speed. This speed difference makes the causality law no longer fully valid except in certain areas

Cross-chain exchange swaps assets among different blockchains, which facilitates cooperation among blockchains. A cross-chain exchange contains several transactions from different blockchains. It requires to synchronize asset transfers in those associated blockchains to avoid partial transfers. Current cross-chain methods are divided into two main types. One locks the assets first and transfers the frozen asset to receivers later. The second one is the two-phase or three-phase transaction commit protocol. Those methods require at least two steps, which makes those blockchains coherent to synchronize at different time(steps). Serialization is required as the second step has to wait for the completion of the first step, and even some steps in the same stage are required to be serialized. Meanwhile, their implements are either by smart contracts or special blockchain structures/roles. Smart contract based methods require to pre-deploy a smart contract and cannot change dynamically. Special structure or role-based methods force associated blockchains to have those special requirements. In this paper, we propose a new cross-chain exchange model based on the dependence of associated transactions, which is expressed inside a transaction and can change when sending transactions. It saves the step to lock the asset or perform a pre-commit and has no special requirements for blockchains or pre-deployment of smart contracts. The simulation results show the proposed model exchanges asset effectively among different blockchains.

Identification of static nonlinear elements (i.e., nonlinear elements whose outputs depend only on the present value of inputs) is crucial for the success of system identification tasks. Identification of static nonlinear elements though can pose several challenges. Two of the main challenges are: (1) mathematical models describing the elements being unknown and thus requiring black-box identification; and (2) collection of sufficiently informative measurements. With the aim of addressing the two challenges, we propose in this paper a method of predetermining informative measurement points offline (i.e., prior to conducting experiments or seeing any measured data), and using those measurements for online model calibration. Since we deal with an unknown model structure scenario, a high order polynomial model is assumed. Over fit and under fit avoidance are achieved via checking model convergence via an iterative means. Model dependent information maximization is done via a D-optimal design of experiments strategy. Due to experiments being designed offline and being designed prior to conducting measurements, this method eases off the computation burden at the point of conducting measurements. The need for in-the-loop information maximization while conducting measurements is avoided. We conclude by comparing the proposed D-optimal design method with a method of in-the-loop information maximization and point out the pros and cons. The method is demonstrated for the single-input-single-output (SISO) static nonlinear element case. The method can be extended to MISO systems as well.

The paper presents a mixed method approach for the analysis of power systems in augmented uncertainty scenarios, related to the increasing penetration of variable renewable energy and country specific constraints to be found in fragile states. In the formulated methodology, both deterministic and probabilistic load flow have their own specific, necessary and interactive role. To establish the soundness of the methodology, the analysis is conducted for a real case study, along with wind speed measurements (eleven-month duration), visual model validations, statistical and load flow analysis. The probabilistic simulations are based on Monte Carlo (MC) analysis. Synthetic data are created from probabilistic distribution functions (PDF) calculated on original measured samples, operational constraints, and load uncertainties. These data are processed by load flow simulations and the results consolidated and analyzed. To facilitate the implementation of the proposed methods, scripts developed in Python programming language have been created for the analysis of statistical data, sample generation, post processing, data visualization and the interaction with conventional software for load flow analysis. The scripts are made public and available for download. The proposed methodology of analysis, conceptualized for developing and fragile states, may also be used as a basis for all power system planning where the number of uncertainties is no longer negligible, and the use of deterministic methods alone would provide inadequate results.

The ability to perceive prosthetic grasping may enable amputees to better interact with external objects. This may require customized coding of multiple sensory feedback for each amputee. This study developed a protocol to determine optimal modulation ranges of sensations elicited by transcutaneous electrical nerve stimulation (TENS). These sensations that were referred to the lost fingers provided the possibility for restoring multi-modalities of sensory feedback for amputees with evoked tactile sensation (ETS) non-invasively. To match the restricted projected finger map area, smaller electrodes must be used to deliver electrical stimulation for multi-channel sensory information, which resulted in fewer types of sensations. Our protocol provided comprehensive information for optimal selection of amplitude and frequency in a personalized, pulse-width encoding paradigm. The good sensitivity for vibration and buzz in both able-bodied and amputee subjects suggested that perceptual intensity can be effectively modulated to convey sensory information via either of the sensations. The efficacy of this protocol in sensory coding for forearm amputees was demonstrated in finger-specific identification experiment. This protocol may allow customization of ETS-based sensory feedback with an optimal encoding strategy for individual amputees.

we characterize the network structure of data such that federated mulit-task learning is possible.