NCTA 2019 Abstracts


Full Papers
Paper Nr: 4
Title:

Neural Models for Benchmarking of Truck Driver Fuel Economy Performance

Authors:

Alwyn J. Hoffman

Abstract: The transport industry is a primary contributor towards emissions that impact climate change. Fuel economy is also of critical importance to the profitability of road freight transport operators. Empirical evidence identified a variety of factors impacting fuel consumption, including route inclination, payload and truck driver behaviour. This creates the need for accurate fuel usage models and objective methods to distinguish the impact of drivers from other factors, in order to enable reliable driver performance assessment. We compiled a data set for 331 drivers completing 7332 trips over 21 routes to obtain evidence of the impact of route, payload and driver behaviour on fuel economy. We then extracted various regression and neural models for fuel economy and used these models to remove the impact of route inclination and payload, allowing the impact of driver performance to be measured more accurately. All models demonstrated significant out-of-sample predictive ability. Neural models in general outperformed regression models, while amongst neural models radial basis models slightly outperformed multi-layer perceptron models. The significance of compensating for factors not controlled by the driver was verified by demonstrating large differences in driver performance ranking before and after compensating for route inclination and payload.

Paper Nr: 9
Title:

Stochastic Information Granules Extraction for Graph Embedding and Classification

Authors:

Luca Baldini, Alessio Martino and Antonello Rizzi

Abstract: Graphs are data structures able to efficiently describe real-world systems and, as such, have been extensively used in recent years by many branches of science, including machine learning engineering. However, the design of efficient graph-based pattern recognition systems is bottlenecked by the intrinsic problem of how to properly match two graphs. In this paper, we investigate a granular computing approach for the design of a general purpose graph-based classification system. The overall framework relies on the extraction of meaningful pivotal substructures on the top of which an embedding space can be build and in which the classification can be performed without limitations. Due to its importance, we address whether information can be preserved by performing stochastic extraction on the training data instead of performing an exhaustive extraction procedure which is likely to be unfeasible for large datasets. Tests on benchmark datasets show that stochastic extraction can lead to a meaningful set of pivotal substructures with a much lower memory footprint and overall computational burden, making the proposed strategies suitable also for dealing with big datasets.

Paper Nr: 12
Title:

Vision based Indoor Obstacle Avoidance using a Deep Convolutional Neural Network

Authors:

Mohammad O. Khan and Gary B. Parker

Abstract: A robust obstacle avoidance control program was developed for a mobile robot in the context of tight, dynamic indoor environments. Deep Learning was applied in order to produce a refined classifier for decision making. The network was trained on low quality raw RGB images. A fine-tuning approach was taken in order to leverage pre-learned parameters from another network and to speed up learning time. The robot successfully learned to avoid obstacles as it drove autonomously in a tight classroom/laboratory setting.

Paper Nr: 15
Title:

Risk-averse Distributional Reinforcement Learning: A CVaR Optimization Approach

Authors:

Silvestr Stanko and Karel Macek

Abstract: Conditional Value-at-Risk (CVaR) is a well-known measure of risk that has been directly equated to robustness, an important component of Artificial Intelligence (AI) safety. In this paper we focus on optimizing CVaR in the context of Reinforcement Learning (RL), as opposed to the usual risk-neutral expectation. As a first original contribution, we improve the CVaR Value Iteration algorithm (Chow et al., 2015) in a way that reduces computational complexity of the original algorithm from polynomial to linear time. Secondly, we propose a sampling version of CVaR Value Iteration we call CVaR Q-learning. We also derive a distributional policy improvement algorithm, and later use it as a heuristic for extracting the optimal policy from the converged CVaR Q-learning algorithm. Finally, to show the scalability of our method, we propose an approximate Q-learning algorithm by reformulating the CVaR Temporal Difference update rule as a loss function which we later use in a deep learning context. All proposed methods are experimentally analyzed, including the Deep CVaR Q-learning agent which learns how to avoid risk from raw pixels.

Paper Nr: 19
Title:

Exact and Approximate Rule Extraction from Neural Networks with Boolean Features

Authors:

Fawaz A. Mereani and Jacob M. Howe

Abstract: Rule extraction from classifiers treated as black boxes is an important topic in explainable artificial intelligence (XAI). It is concerned with finding rules that describe classifiers and that are understandable to humans, having the form of (I f...Then...Else). Neural network classifiers are one type of classifier where it is difficult to know how the inputs map to the decision. This paper presents a technique to extract rules from a neural network where the feature space is Boolean, without looking at the inner structure of the network. For such a network with a small feature space, a Boolean function describing it can be directly calculated, whilst for a network with a larger feature space, a sampling method is described to produce rule-based approximations to the behaviour of the network with varying granularity, leading to XAI. The technique is experimentally assessed on a dataset of cross-site scripting (XSS) attacks, and proves to give very high accuracy and precision, comparable to that given by the neural network being approximated.

Paper Nr: 21
Title:

Active Recall Networks for Multiperspectivity Learning through Shared Latent Space Optimization

Authors:

Theus H. Aspiras, Ruixu Liu and Vijayan K. Asari

Abstract: Given that there are numerous amounts of unlabeled data available for usage in training neural networks, it is desirable to implement a neural network architecture and training paradigm to maximize the ability of the latent space representation. Through multiple perspectives of the latent space using adversarial learning and autoencoding, data requirements can be reduced, which improves learning ability across domains. The entire goal of the proposed work is not to train exhaustively, but to train with multiperspectivity. We propose a new neural network architecture called Active Recall Network (ARN) for learning with less labels by optimizing the latent space. This neural network architecture learns latent space features of unlabeled data by using a fusion framework of an autoencoder and a generative adversarial network. Variations in the latent space representations will be captured and modeled by generation, discrimination, and reconstruction strategies in the network using both unlabeled and labeled data. Performance evaluations conducted on the proposed ARN architectures with two popular datasets demonstrated promising results in terms of generative capabilities and latent space effectiveness. Through the multiple perspectives that are embedded in ARN, we envision that this architecture will be incredibly versatile in every application that requires learning with less labels.

Paper Nr: 26
Title:

Breast Cancer Automatic Diagnosis System using Faster Regional Convolutional Neural Networks

Authors:

Lourdes Duran-Lopez, Juan P. Dominguez-Morales, Isabel Amaya-Rodriguez, Francisco Luna-Perejon, Javier Civit-Masot, Saturnino Vicente-Diaz and Alejandro Linares-Barranco

Abstract: Breast cancer is one of the most frequent causes of mortality in women. For the early detection of breast cancer, the mammography is used as the most efficient technique to identify abnormalities such as tumors. Automatic detection of tumors in mammograms has become a big challenge and can play a crucial role to assist doctors in order to achieve an accurate diagnosis. State-of-the-art Deep Learning algorithms such as Faster Regional Convolutional Neural Networks are able to determine the presence of an object and also its position inside the image in a reduced computation time. In this work, we evaluate these algorithms to detect tumors in mammogram images and propose a detection system that contains: (1) a preprocessing step performed on mammograms taken from the Digital Database for Screening Mammography (DDSM) and (2) the Neural Network model, which performs feature extraction over the mammograms in order to locate tumors within each image and classify them as malignant or benign. The results obtained show that the proposed algorithm has an accuracy of 97.375%. These results show that the system could be very useful for aiding physicians when detecting tumors from mammogram images.

Paper Nr: 28
Title:

A Sequential Heteroscedastic Probabilistic Neural Network for Online Classification

Authors:

Ali Mahmoudi, Reza A. Moghadam and Kurosh Madani

Abstract: In this paper, a novel online classification algorithm called sequential heteroscedastic probabilistic neural network (SHPNN) is proposed. This algorithm is based on Probabilistic Neural Networks (PNNs). One of the advantages of the proposed algorithm is that it can increase the number of its hidden node kernels adaptively to match the complexity of the data. The performance of this network is analyzed for a number of standard datasets. The results suggest that the accuracy of this algorithm is on par with other state of the art online classification algorithms while being significantly faster in majority of cases.

Short Papers
Paper Nr: 2
Title:

Semi-automatic Segmentation of MRI Brain Metastases Combining Support Vector Machine and Morphological Operators

Authors:

Gloria Gonella, Elisabetta Binaghi, Paola Nocera and Cinzia Mordacchini

Abstract: The objective of this study is to develop a semi-automatic, interactive segmentation strategy for efficient and accurate brain metastases delineation on Post Gadolinium T1-weighted brain MRI images. Salient aspects of the proposed solutions are the combined use of machine learning and image processing techniques, based on Support Vector Machine and Morphological Operators respectively, to delineate pathological and healthy tissues. The overall segmentation procedure is designed to operate on a clinical setting to reduce the workload of health-care professionals but leaving to them full control of the process. The segmentation process was validated for in-house collected image data obtained from radiation therapy studies. The results prove that the allied use of SVM and Morphological Operators produces accurate segmentations, useful for their insertion in clinical practice.

Paper Nr: 6
Title:

Neural Sequence Modeling in Physical Language Understanding

Authors:

Avi Bleiweiss

Abstract: Automating the tasks of generating test questions and analyzing content for assessment of written student responses has been one of the more sought-after applications to support classroom educators. However, a major impediment to algorithm advances in developing such tools is the lack of large and publicly available domain corpora. In this paper, we explore deep learning of physics word problems performed at scale using the transformer, a state-of-the-art self-attention neural architecture. Our study proposes an intuitive novel approach to a tree-based data generation that relies mainly on physical knowledge structure and defers compositionality of natural language clauses to the terminal nodes. Applying our method to the simpler kinematics domain that describes motion properties of an object at a uniform acceleration rate and using our neural sequence model pretrained on a dataset of ten thousand machine-produced problems, we achieved BLEU scores of 0.54 and 0.81 for predicting derivation expressions on real-world and synthetic test sets, respectively. Notably increasing the number of trained problems resulted in a diminishing return on performance.

Paper Nr: 10
Title:

On the Design of a Heuristic based on Artificial Neural Networks for the Near Optimal Solving of the (N2–1)-puzzle

Authors:

Vojtěch Cahlík and Pavel Surynek

Abstract: This paper addresses optimal and near-optimal solving of the (N2–1)-puzzle using the A* search algorithm. We develop a novel heuristic based on artificial neural networks (ANNs) called ANN-distance that attempts to estimate the minimum number of moves necessary to reach the goal configuration of the puzzle. With a well trained ANN-distance heuristic, whose inputs are just the positions of the pebbles, we are able to achieve better accuracy of predictions than with conventional heuristics such as those derived from the Manhattan distance or pattern database heuristics. Though we cannot guarantee admissibility of ANN-distance, an experimental evaluation on random 15-puzzles shows that in most cases ANN-distance calculates the true minimum distance from the goal, and furthermore, A* search with the ANN-distance heuristic usually finds an optimal solution or a solution that is very close to the optimum. Moreover, the underlying neural network in ANN-distance consumes much less memory than a comparable pattern database.

Paper Nr: 13
Title:

Calibration Techniques for Binary Classification Problems: A Comparative Analysis

Authors:

Alessio Martino, Enrico De Santis, Luca Baldini and Antonello Rizzi

Abstract: Calibrating a classification system consists in transforming the output scores, which somehow state the confidence of the classifier regarding the predicted output, into proper probability estimates. Having a well-calibrated classifier has a non-negligible impact on many real-world applications, for example decision making systems synthesis for anomaly detection/fault prediction. In such industrial scenarios, risk assessment is certainly related to costs which must be covered. In this paper we review three state-of-the-art calibration techniques (Platt’s Scaling, Isotonic Regression and SplineCalib) and we propose three lightweight procedures based on a plain fitting of the reliability diagram. Computational results show that the three proposed techniques have comparable performances with respect to the three state-of-the-art approaches.

Paper Nr: 16
Title:

Unsupervised Detection of Sub-pixel Objects in Hyper-spectral Images via Diffusion Bases

Authors:

Alon Schclar and Amir Averbuch

Abstract: Sub-pixel objects are defined as objects which due to their size and due to the resolution of the camera occupy a fraction of a pixel or partially span adjacent pixels. Unsupervised detection of sub-pixel objects can be highly useful in areas such as medical imaging, and surveillance, to name a few. Hyper-spectral images offer extensive intensity information by describing a scene at hundreds and even thousands of wavelengths. This information can be utilized to obtain better sub-pixel detection results compared to those that are obtained using RGB images. Usually, only a small number of wavelengths contain the information that is required for the detection. Furthermore, the intensity images of many wavelengths are noisy and contain very little information. Accordingly, hyper-spectral images must be pre-processed first in order to extract the information that is needed for the sub-pixel detection. This extraction process produces an image where each pixel is represented by a small number of features which allows the application of fast and efficient detection algorithms. In this paper we propose the Diffusion Bases (DB) dimensionality reduction algorithm in order to derive the essential features for the sub-pixel detection. The effectiveness of the DB algorithm facilitates the application of a very simple algorithm for the detection of sub-pixel objects in the feature space. The proposed approach does not assume any distribution of the background pixels. We demonstrate the proposed framework for the detection of cardboard objects in airborne hyper-spectral images of a desert terrain.

Paper Nr: 23
Title:

A Low-power, Reachable, Wearable and Intelligent IoT Device for Animal Activity Monitoring

Authors:

L. Duran-Lopez, D. Gutierrez-Galan, J. P. Dominguez-Morales, A. Rios-Navarro, R. Tapiador-Morales, A. Jimenez-Fernandez, D. Cascado-Caballero and A. Linares-Barranco

Abstract: Along with the proliferation of mobile devices and wireless signal coverage, IoT devices, such as smart wristbands for monitoring its owner’s activity or sleep patterns, get great popularity. Wearable technology in human life has become quite useful due to the information given (sleep hours, heart rate, etc). However, wearables for animals does not give information about behaviour directly: they collect raw data that is sent to a server where, after a post-processing step, the behaviour is known. In this work, we present a smart IoT device that classifies different animal behaviours from the information obtained from on-board sensors using an embedded neural network running in the device. This information is uploaded to a server through a wireless sensor network based on Zigbee communication. The architecture of the device allows an easy assembly in a reduced dimension wearable case. The firmware allows a modular functionality by activating or deactivating modules independently, which improve the power efficiency of the device. The power consumption has been analyzed, allowing the 1Ah battery to work the system during several days. A novel localization and distance estimation technique (for 802.15.4 networks) is presented to recover a lost device in Doñana National Park with unidirectional antennas and log-normalization distance estimation over RSSI.

Paper Nr: 24
Title:

Challenging the Intuition about Memory and Computation in Theories of Cognition

Authors:

Jochen Kerdels and Gabriele Peters

Abstract: In this position paper we argue that the concepts of memory and computation as they are commonly used in theories of cognition are strongly influenced by our intuitive understanding of the corresponding concepts in contemporary computer systems, leading to an implicit loss of biological plausibility. To support our argument we provide an alternative perspective on memory and computation that allows a closer comparison of the capabilities of computer programs running on computer systems and neurobiological systems showing that computer programs exhibit a computational flexibility that is difficult to reconcile with neurobiological constraints. We end this paper by offering a neurobiologically plausible perspective on memory that views memory as a dynamic, distributed process that is an intrinsic part of a neurobiological network that integrates information, e.g., sensory information.

Paper Nr: 25
Title:

Glioma Diagnosis Aid through CNNs and Fuzzy-C Means for MRI

Authors:

I. Amaya-Rodriguez, L. Duran-Lopez, F. Luna-Perejon, J. Civit-Masot, J. P. Dominguez-Morales, S. Vicente, A. Civit, D. Cascado and A. Linares-Barranco

Abstract: Glioma is a type of brain tumor that causes mortality in many cases. Early diagnosis is an important factor. Typically, it is detected through MRI and then either a treatment is applied, or it is removed through surgery. Deep-learning techniques are becoming popular in medical applications and image-based diagnosis. Convolutional Neural Networks are the preferred architecture for object detection and classification in images. In this paper, we present a study to evaluate the efficiency of using CNNs for diagnosis aids in glioma detection and the improvement of the method when using a clustering method (Fuzzy C-means) for pre-processing the input MRI dataset. Results offered an accuracy improvement from 0.77 to 0.81 when using Fuzzy C-Means.

Paper Nr: 27
Title:

Sampling Frequency Evaluation on Recurrent Neural Networks Architectures for IoT Real-time Fall Detection Devices

Authors:

F. Luna-Perejon, J. Civit-Masot, L. Muñoz-Saavedra, L. Duran-Lopez, I. Amaya-Rodriguez, J. P. Dominguez-Morales, S. Vicente-Diaz, A. Linares-Barranco, A. Civit-Balcells and M. J. Dominguez-Morales

Abstract: Falls are one of the most frequent causes of injuries in elderly people. Wearable Fall Detection Systems provided a ubiquitous tool for monitoring and alert when these events happen. Recurrent Neural Networks (RNN) are algorithms that demonstrates a great accuracy in some problems analyzing sequential inputs, such as temporal signal values. However, their computational complexity are an obstacle for the implementation in IoT devices. This work shows a performance analysis of a set of RNN architectures when trained with data obtained using different sampling frequencies. These architectures were trained to detect both fall and fall hazards by using accelerometers and were tested with 10-fold cross validation, using the F1-score metric. The results obtained show that sampling with a frequency of 25Hz does not affect the effectiveness, based on the F1-score, which implies a substantial increase in the performance in terms of computational cost. The architectures with two RNN layers and without a first dense layer had slightly better results than the smallest architectures. In future works, the best architectures obtained will be integrated in an IoT solution to determine the effectiveness empirically.

Paper Nr: 29
Title:

Multi-dataset Training for Medical Image Segmentation as a Service

Authors:

Javier Civit-Masot, Francisco Luna-Perejón, Lourdes Duran-Lopez, J. P. Domínguez-Morales, Saturnino Vicente-Díaz, Alejandro Linares-Barranco and Anton Civit

Abstract: Deep Learning tools are widely used for medical image segmentation. The results produced by these techniques depend to a great extent on the data sets used to train the used network. Nowadays many cloud service providers offer the required resources to train networks and deploy deep learning networks. This makes the idea of segmentation as a cloud-based service attractive. In this paper we study the possibility of training, a generalized configurable, Keras U-Net to test the feasibility of training with images acquired, with specific instruments, to perform predictions on data from other instruments. We use, as our application example, the segmentation of Optic Disc and Cup which can be applied to glaucoma detection. We use two publicly available data sets (RIM-One V3 and DRISHTI) to train either independently or combining their data.

Paper Nr: 30
Title:

Multitask Learning or Transfer Learning? Application to Cancer Detection

Authors:

Stephen Obonyo and Daniel Ruiru

Abstract: Multitask Learning (MTL) and Transfer Learning (TL) are two key Machine Learning (ML) approaches which have been widely adopted to improve model’s performance. In Deep Learning (DL) context, these two learning methods have contributed to competitive results in various areas of application even if the size of dataset is relatively small. While MTL involves learning from a key task and other auxiliary tasks simultaneously and sharing signals among them, TL focuses on the transfer of knowledge from already existing solution within the same domain. In this paper, we present MTL and TL based models and their application to Invasive Ductal Carcinoma (IDC) detection. During training, the key learning task in MTL was detection of IDC whereas skin and brain tumor were auxiliary tasks. On the other hand, the TL-based model was trained on skin cancer dataset and the learned representations transferred in order to detect IDC. The accuracy difference between MTL-based model and TL-based model on IDC detection was 8.6% on validation set and 9.37% on training set. On comparing the loss metric of the same models, a cross entropy of 0.18 and 0.08 was recorded on validation set and training set respectively.

Paper Nr: 31
Title:

Identify Theft Detection on e-Banking Account Opening

Authors:

Roxane Desrousseaux, Gilles Bernard and Jean-Jacques Mariage

Abstract: Banks are compelled by financial regulatory authorities to demonstrate whole-hearted commitment to finding ways of preventing suspicious activities. Can AI help monitor user behavior in order to detect fraudulent activity such as identity theft? In this paper, we propose a Machine Learning (ML) based fraud detection framework to capture fraudulent behavior patterns and we experiment on a real-world dataset of a major European bank. We gathered recent state-of-the-art techniques for identifying banking fraud using ML algorithms and tested them on an abnormal behavior detection use case.

Paper Nr: 11
Title:

Learning Method of Recurrent Spiking Neural Networks to Realize Various Firing Patterns using Particle Swarm Optimization

Authors:

Yasuaki Kuroe, Hitoshi Iima and Yutaka Maeda

Abstract: Recently it has been reported that artificial spiking neural networks (SNNs) are computationally more powerful than the conventional neural networks. In biological neural networks of living organisms, various firing patterns of nerve cells have been observed, typical examples of which are burst firings and periodic firings. In this paper we propose a learning method which can realize various firing patterns for recurrent SNNs (RSSNs). We have already proposed learning methods of RSNNs in which the learning problem is formulated such that the number of spikes emitted by a neuron and their firing instants coincide with given desired ones. In this paper, in addition to that, we consider several desired properties of a target RSNN and proposes cost functions for realizing them. Since the proposed cost functions are not differentiable with respect to the learning parameters, we propose a learning method based on the particle swarm optimization.

Paper Nr: 18
Title:

Demand Forecasting using Artificial Neuronal Networks and Time Series: Application to a French Furniture Manufacturer Case Study

Authors:

Julie Bibaud-Alves, Philippe Thomas and Hind B. El Haouzi

Abstract: The demand forecasting remains a big issue for the supply chain management. At the dawn of Industry 4.0, and with the first encouraging results concerning the application of deep learning methods in the management of the supply chain, we have chosen to study the use of neural networks in the elaboration of sales forecasts for a French furniture manufacturing company. Two main problems have been studied for this article: the seasonality of the data and the small amount of valuable data. After the determination of the best structure for the neuronal network, we compare our results with the results form AZAP, the forecasting software using in the company. Using cross-validation, early stopping, robust learning algorithm, optimal structure determination and taking the mean of the month turns out to be in this case study a good way to get enough close to the current forecasting system.

Paper Nr: 22
Title:

Prediction and Classification of Heart Disease using AML and Power BI

Authors:

Debmalya Chatterjee and Saravanan Chandran

Abstract: Machine Learning (ML) is transforming the industries from delivering normal products to deliver intellect products. Large sets of data points are analysed by the computers and the relationship modelling is applied in a predictive way in real time to obtain accurate results. Machine Learning is adopted in healthcare problems for increasing efficiencies, saving money, and saving lives. The cost of medical treatment is reduced and the healthcare processes are optimized throughout the organization with the support of ML. ML improves healthcare delivery and patient health. Machine learning improves diagnosis and treatment options, also empowers individuals to take control of their health. Diagnosis advancements, predictive healthcare, medicines, and helping patients through ML interface produces better results. Heart Disease relates to many numbers of medical complications related to the heart. In recent years, ML has spread its knowledge in every field. In healthcare, the usage of ML has been significantly increased. This research work aims at the prediction of heart disease and classification of heart disease using Machine Learning algorithms. The experimental results are classified into five heart disease stages using values 0, 1, 2, 3, and 4, value 0 for no heart disease and 4 for severe heart disease. The Area Under the Curve (AUC) values depict the accuracy level of the prediction using this proposed model. The results are displayed using the data set in the form of charts that is easy to analyse the number of people having chest pains. The ML analytical report added up in the form of charts or other visuals, the results are reported informatively. This analysis is helpful for doctors and the medical industry for several case studies.