Keywords: Control Systems Design, Adaptive Control, Optimization
Abstract: This paper merges the advantages of data-driven control and sliding mode control in terms of applying and modifying two combinations of model-free adaptive control and sliding mode control suggested by Ebrahimi et al. in 2018 to the position control of tower crane systems. The modifications concern the classical definition of the control error or the tracking error and appropriate proofs are adapted and summarized. The two controllers are validated experimentally and compared in the control of the three positions specific to tower crane system laboratory equipment.

Keywords: Predictive Control, Optimization
Abstract: This paper focuses on the (re-)construction of the optimal solution for the multi-parameter quadratic programming (mpQP) problems. Optimization problems of this nature are widely employed in the formulation of modelbased predictive controllers (MPC) for discrete linear systems, wherein input and state constraints are imposed. This study examines the geometric characteristics of the explicit solution of an mpQP problem and introduces a novel convex-concave lifting technique to synthesize an equivalent mpQP problem. Whenever the solution corresponds to a PWA function, the present approach maintains the structure and control laws for the original systems. A new (less complex) cost function and a corresponding feasible domain are constructed through lifting for the equivalent optimization problem. The effectiveness of this strategy is demonstrated through an illustrative example.

Keywords: Nonlinear Systems, Control Systems Design, Robust Control
Abstract: In this paper, a new approach is introduced which combines Eigenvalue Assignment, State Dependent Riccati Equation (SDRE) and Sliding Mode Control (SMC) methods for nonlinear systems. In the classical SDRE based SMC (SDRE-SMC) approach, a nonlinear system is frozen at each time instant to obtain a linear-like structure model that is used to design a sliding surface (SS) at each time instant. This mechanism produces a state-dependent SS to hold the states on the SS. The approach proposed here is built on this mechanism and offers a new way to design a state-dependent SS for nonlinear systems so that the pointwise eigenvalues of the closed-loop system matrix of the control-free dynamics in the regular form can be kept in a specified disk. This gives a great advantage to shape the transient response characteristics. The performance of the nonlinear controller approach proposed here is investigated in simulations.

Keywords: Optimization, Linear Systems, Control Systems Design
Abstract: The paper deals with an optimal tracking problem in the presence of disturbances. A linear system in discrete time is considered. The paper presents a procedure with computational advantages. Step variant functions are used for approximation of the disturbance and reference variables. Simulation results are indicated.

Keywords: Control Systems Design, Linear Systems, Optimization
Abstract: This paper presents an enhanced approach for synthesizing a robust static output feedback H-infinity controller for linear discrete-time systems with polytopic uncertainties and external disturbances. While this problem has been extensively studied in the literature, the proposed method distinguishes itself through the utilization of parameter-dependent Lyapunov functions and novel bounding techniques for bilinear terms. By employing a more flexible and accurate characterization of system dynamics and uncertainties, our approach achieves improved controller performance with less conservatism compared to existing methods. The formulation of the controller design problem involves converting the nonconvex optimization into a convex minimization one using a congruent transformation and the cone complementarity technique. This leads to a set of linear matrix inequality conditions that guarantee the existence of an effective robust output feedback H-infinity controller capable of mitigating the effects of uncertainties and disturbances on the system. Numerical simulations show that our proposed method outperforms existing results in terms of disturbance attenuation rates.

Keywords: Robust Control, Optimization, System Identification and Modeling
Abstract: This paper proposes an alternative nonconvex approach to fit single-input and single-output transfer function models on magnitude frequency-based response measurements, with additional constraints such as upper boundness, stability, minimum phase and validity with respect to provided data. The proposed solution presents an improvement over the established Log-Chebyshev convex fit which is used as a well-placed starting point and refines the optimization to become less conservative and feasible in cases where (near) singularities are present and limit its sole application. As such, our contribution provides advantages in cases such as in mu-synthesis where the uncertainty model should have a low order to guarantee no conservativeness of the structured singular value approximation. Two case studies are described and discussed which show brought improvements.

Keywords: Robotics, Modeling, Simulation and CAD Tools, Other Topics
Abstract: Kinematics modelling of robotic systems is a fundamental requirement used by the underling control system. In this paper, we introduce a complete and analytic direct and inverse kinematic model for an 18 DoF quadruped robot. The proposed solution, based on simple trigonometric functions, is analytically determined, allowing the full control of the translation and orientation of the footholds and of the body. Considering the fact that the dynamic model of the robot requires position, velocity and angular acceleration profiles as references, this article also presents the applicability of a trajectory generator with zero initial and zero final conditions for the first and second derivatives. As performance evaluation, we have determined the accuracy of the model by implementing a static gait controller on a A1 Unitree quadrupped robot, where the trajectory of each foot is calculated based on Bsplines, given as reference to the inverse kinematics model.

Keywords: Robotics, Machine Learning, Intelligent Systems
Abstract: This paper presents a study on the mapless navigation of autonomous mobile robot using Deep Reinforcement Learning in an intralogistics setting. The task is to make an autonomous mobile robot learn to navigate to a goal without prior knowledge of the environment. In this paper, a controller using the Soft Actor-Critic algorithm is designed, trained, and applied for navigating the robot equipped with 360° LiDAR and front camera sensors. The controller is successfully validated in a fully observable environment under extensive simulations. Furthermore, we investigate the performance of the proposed controller in a partially observable environment and possible limitations. We use a 3D Temporal Convolution Network for processing the time series image data from visual observations. Besides Partial Observability, we also address the problem of sparse positive rewards in training the Deep Reinforcement Learning algorithm with a combined approach of Automatic Curriculum Learning and Dual Prioritized Experience Replay.

Keywords: Machine Learning, Robotics, Computer Vision
Abstract: We consider a problem in which the trajectory of a mobile 3D sensor must be optimized so that certain objects are both found in the overall scene and covered by the point cloud, as fast as possible. This problem is called target search and coverage, and the paper provides an end-to-end deep reinforcement learning (RL) solution to solve it. The deep neural network combines four components: deep hierarchical feature learning occurs in the first stage, followed by multi-head transformers in the second, max-pooling and merging with bypassed information to preserve spatial relationships in the third, and a distributional dueling network in the last stage. To evaluate the method, a simulator is developed where cylinders must be found by a Kinect sensor. A network architecture study shows that deep hierarchical feature learning works for RL and that by using farthest point sampling (FPS) we can reduce the amount of points and achieve not only a reduction of the network size but also better results. We also show that multi-head attention for point-clouds helps to learn the agent faster but converges to the same outcome. Finally, we compare RL using the best network with a greedy baseline that maximizes immediate rewards and requires for that purpose an oracle that predicts the next observation. RL achieves significantly better and more robust results than the greedy strategy.

Keywords: Manufacturing Systems, Robotics, Real Time Applications
Abstract: This paper deals with digital twin (DT) approach for a processing technology (PT) running on a mechatronic processing system (MPS) assisted by a mobile cyber-physical robotic system (MCPRS). The hardware architecture consists of the MPS, four workstation (WS), line-shaped, and MCPRS. MCPRS has in its structure a wheeled mobile robot (WMR) equipped with robotic manipulator (RM) having on the end effector a mobile visual servoing system MVSS). The workpiece (WP) is moved along the four stations for processing, and at the end, if it does not pass the primary quality test (PQT), it is picked up by the MCPRS, transported to the first station for reprocessing or scrapping. If the WP does not pass the second quality test (SQT), then it is stored as scrapped. WP that passes the SQT, through the same processing operations, will be brought to the quality standard. Thus, the WP will go through the MPS twice, for processing/reprocessing (P/R). The virtual world that serves as the subsystems of MCPRS. Additionally, the virtual world includes hybrid modeling with synchronized hybrid Petri nets (SHPN), simulation of the SHPN models, modeling of the MVSS, and simulation of the discrete-time trajectory-tracking sliding-mode control (DT-TTSMC) of MCPRS. The real world, corresponding to the virtual world, consists of communication, synchronization and control of the MPS and MCPRS’s subsystems (WMR, RM and MVSS), the graphical user interface (GUI) and a supervisory control and data acquisition (SCADA) system, implemented on a remote PC.

Keywords: Robotics, Virtual and Augmented Reality, Automatic Control Education and Training
Abstract: In recent years, the manufacturing sector has undergone a significant transition from a paradigm of robotic autonomy to human-robot collaboration. This approach allows operators from different domains to rapidly adapt to a new way of working and reduces redundant learning costs, while also enabling intuitive robot control. To further enhance this paradigm, integrating Mixed Reality (MR) technology has emerged as a promising approach to creating and working in a virtual workspace while having a safe environment for training. This work develops a framework, based on digital twin representation, to integrate the MR with real-life equipment and humans. This allows for the control of a collaborative robot in an environment with objects or humans perceived by the eye-in-hand visual sensor. The efficiency of the proposed framework is emphasized by a comparison of Unity simulations versus the real-time behavior of the collaborative robot.

Keywords: Predictive Control, Robust Control, Robotics
Abstract: Quadruped robots and many other types of legged robots have been gaining popularity in recent years due to their advantages over wheeled robots. Firstly, quadruped robots have excellent mobility and stability. With four legs, the quadruped robots can move more efficiently over rough terrain and uneven surfaces, making them a prime candidate for use in disaster response situations, search and rescue operations and exploration missions. Furthermore, quadruped robots don't have holonomic constraints, thus making them more maneuverable than any other type of robots. This makes them ideal for use in confined spaces. The problem of quadruped control can be divided in low level control and high level control, the former concerning each joint angle position, velocity, acceleration and torque to achieve a desired motion for each leg, while the latter is concerned with the required whole-body linear and angular velocities to follow a specified trajectory. In this paper we present a high level control strategy using Dynamic Window Approach (DWA) and a modified dynamic model for the whole-body dynamics of the quadruped. Furthermore, we study the effects of modified mass-inertia matrix on the overall performance of the algorithm. Our tests are conducted in a simulated environment and in real world.

Keywords: Cloud Computing, Distributed Systems, Computer Science Education and Training
Abstract: Plagiarism is a well-known problem that impacts the academic community. The escalation of source code plagiarism affects the quality of computer science and technology education. This article approaches a novel architecture for an academic plagiarism detection system focused on projects and source code. A part of the detection process is based on a few already existing open-source tools. All the system components use the OpenStack Cloud platform services for scalability, parallel processing of files, and load balancing. The solution also integrates with an identity and access management system, which contains users, roles, and cohorts. The proposed solution is an efficient approach for both organizing assignments at which students upload projects and for detecting plagiarism in all the submitted assignments.

Keywords: Distributed Systems, Software Engineering, Cloud Computing
Abstract: Peer review represents the status-quo when it comes to evaluating research articles that are submitted to conferences and journals. The significance of a computer science article is given by the prestige of the publication and is correlated with the inclusion in the ISI Web of Science index. This paper discusses the issues of the current paper publication paradigm and proposes a decentralized approach to the paper dissemination and the peer review processes. On the one hand, decentralization and transparency are obtained by employing smart contracts, through blockchain technology. On the other hand, an optimization of the paper rating system is obtained by employing a system of expert badges, based on NFTs, which ensure that the peer review process is just and that only specialists in the fields associated to the contributed paper offer proficient feedback. Other proposed facets include the remuneration of reviewers, a method of allowing the proposed system to expand based on the community's input, and a solution for allowing the organization of conferences.

Keywords: Cloud Computing, Databases and Data Mining, Software Engineering
Abstract: Most research addressing OCR through machine learning techniques is focused on the actual algorithms and on using the MNIST data set as the de facto benchmark. Little effort was made to extend the data set or to build an entirely new one. Furthermore, support for characters other than English ones is mostly limited. This paper presents an OpenStack based approach that aims to overcome this last limitation by providing a community-oriented solution for developing and maintaining richer, language agnostic, community-shared data sets for OCR based applications. The proposed architecture is integrated with OpenStack services and relies on new Cloud perspectives, such as Function-as-a-Service (FaaS), to achieve a greater degree of flexibility. The included modules allow users to upload their own data sets, select or fine-tune their desired pre-processing methods, and derive the required features for their target character set. Both the input and the output data are stored using OpenStack specific data services and are shared for all the users of the Cloud deployment. An interesting feature is that the underlying FaaS functionality would also allow interested parties to upload their own pre-processing and feature extraction stages.

Keywords: Cloud Computing, Distributed Systems, Software Engineering
Abstract: There is an increasing need for computational and storage capabilities for complex distributed applications. Existing solutions need to be deployed in an environment that allows for an increase in performance, scalability, and availability. This paper takes looks at the state-of-the-art regarding methods that take existing applications and make them more efficient by using Cloud services. The novelty of the paper consists of a proposed framework for deploying applications on three major Cloud providers (i.e., Amazon's AWS, Google Cloud and Microsoft Azure) and on the OpenStack open-source Cloud. After the main services from the four Cloud providers are identified, different deployment methods are described depending on the Cloud services and on the requirements of the application. Also, some examples of migrations are discussed with reference to specific Cloud provider services. The proposed solution for Anything-as-a-Service (YaaS) is a straightforward framework for taking different types of applications and migrating them to the Cloud. Therefore, the deployed applications benefit from Cloud features such as resource pooling, availability or scalability, while also being wary of the incurring costs.

Keywords: Cloud Computing, Distributed Systems, Internet of Things
Abstract: Ambient Assisted Living (AAL) is starting to become the norm as more and more smart devices and sensors are installed in people's homes. This is an important aspect in improving the quality of life, especially for the elderly and for people with disabilities.

The solution presented in this paper approaches this paradigm in a large scale deployment context, as it proposes an AAL architecture on three layers (Edge, Dew, and Cloud), which is capable of handling high amount of sensor data in near real-time. Using a distributed system of rules engines, the system is able to take actions based on the acquired data. An important novel concept is the "new home deployment", which facilitates the integration of the devices from a new installation site - the software setup of the new edge infrastructure is done by using a dedicated Cloud service. In order to showcase the practicality and efficiency of the architecture, this paper presents the intricacies of the employed services and service-interaction when deploying the solution on the OpenStack Cloud platform.

Keywords: Distributed Systems, Cloud Computing
Abstract: In this paper, we present a novel architecture for an auto-scaling MPI cluster that leverages the capabilities of OpenMPI and OpenStack frameworks. Unlike traditional MPI clusters, which are typically static and expand to the maximum available resources, our proposed architecture aims to dynamically allocate resources in a cloud environment. This strategy is essential to mitigate high costs associated with static provisioning and ensure that the parallel tasks running on the cluster can fully benefit from the computational power of the cloud when needed.

By integrating OpenMPI, a high-performance MPI implementation, with the flexible resource management capabilities of OpenStack, we enable the dynamic allocation and de-allocation of compute resources based on the cluster's workload demands. This auto-scaling mechanism allows the cluster to scale up or down in response to changing computational requirements, optimizing resource utilization and reducing costs.

Keywords: Predictive Control, Machine Learning, Adaptive Control
Abstract: In this work, we present a new method to tune a Model Predictive Controller (MPC) with the help of a Reinforcement Learning (RL) algorithm called Double Q-Learning. In this algorithm, two function approximators with different sets of parameters are trained simultaneously. First, the nonlinear MPC is parametrized in the weights of its cost function and unknown parameters of its equality and inequality constraints. Then, it is defined as the action-value function of the Double Q-Learning algorithm. By randomly switching between two sets of parameters in the MPC, we show that the exploration of the proposed algorithm increases. Since model error terms are added to the baseline stage cost, thanks to more exploration, less model mismatch is obtained. With this, less bias in the MPC controller is achieved compared to an MPC-based Q-Learning algorithm. Simulation results on a coupled tanks system show that not only the training process resulted to be faster than observed for the MPC-based Q-Learning method, but also the final control performance was found to be more desirable.

Keywords: Predictive Control, Robust Control, Linear Systems
Abstract: This work introduces a stochastic model predictive control scheme for dynamic chance constraints. We consider linear discrete-time systems affected by unbounded additive stochastic disturbance. To synthesize an optimal controller, we solve two subsequent stochastic optimization problems. The first problem concerns finding the maximal feasible probabilities of the dynamic chance constraints. After obtaining the probabilities, the second problem concerns finding an optimal controller using stochastic model predictive control. We solve both stochastic optimization problems by reformulating them into deterministic ones using probabilistic reachable tubes and constraint tightening. We prove that the developed algorithm is recursively feasible and yields closed-loop satisfaction of the dynamic chance constraints. In addition, we will introduce a novel implementation using zonotopes to describe the tightening analytically. Finally, we will end with an example illustrating the method's benefits.

Keywords: Predictive Control, Control Systems Design, Biomedical Engineering
Abstract: In the operating rooms and the intensive care unit, it is crucial to manage the patient's hemodynamic status, which includes factors like cardiac output and mean arterial pressure. Anesthesiologists confront a difficult task while monitoring high-risk patients. Cardiac output optimization has been found to enhance the result of high-risk patients in terms of hospital stay, mortality rate, post-operative problems, etc. The application of standard control approaches is restricted because the mean arterial pressure response of a patient using vasoactive medicines is modeled by a first-order dynamical system with time-varying parameters and a time-varying delay in the control input. In order to circumvent implementation challenges, this work develops an approximation technique that describes the system using a higher-order model. Predictive control is therefore used to comprehend the practical application of higher-order hemodynamic systems. The effectiveness of this strategy is demonstrated by the simulations and outcomes that are given.

Keywords: Predictive Control, Automotive Control Systems, Control Systems Design
Abstract: Predictive control based on an informative system trajectory, instead of a physics-based model, has received significant attention in recent years. This paper investigates the potential of using such data-driven control for vehicle dynamics control and autonomous path following. By considering the path following problem in the error space, the underlying system is approximately linear and existing results for data-driven predictive control can be applied. Also, scheduling based on longitudinal speed can be readily included. The proposed control algorithm was tested on two different lane change maneuvers in a high-fidelity simulation environment.

Keywords: Predictive Control, Automotive Control Systems, Control Systems Design
Abstract: The cooperative adaptive cruise control (CACC) functionality received significant interest in the state-of-the-art due to its advantages in optimizing traffic flow. The model-based predictive control (MPC) strategy was used in various studies due to its advantages in improving the performance of the vehicles (reducing the travel costs, improving the quality of the travel by reducing sudden accelerations, and ensuring the stability of the platoons). Moreover, MPC solutions are built to maximize the advantages of vehicular communication by sharing predictions on states of vehicles (e.g., velocities, accelerations, trajectories). In addition, MPC is also used to compensate for the disturbances added by communications. Thus, this paper proposes a CACC strategy for a vehicle platoon. The solution is based on the distributed MPC (DMPC) strategy, and the controller is proposed in discrete time, ensuring predecessor-follower string stability for the whole platoon.

Keywords: Machine Learning, Optimization, Industrial Applications
Abstract: In the current paper, a strategy for handling a stochastic optimization problem based on metaheuristic techniques is presented. This optimization problem is defined based on the impact of environmental factors such as temperature and irradiance on photovoltaic panels. The objective aims to solve the defined problem by finding an SVM model of a photovoltaic panel. To enhance the efficiency of power generation, this model will be utilized for the purpose of predicting the voltage linked to the Maximum Power Point of the panel.

Keywords: Automotive Control Systems, Predictive Control, Intelligent Systems
Abstract: The development, enhancement, and integration of intelligent functionalities aimed at automating vehicles are increasingly becoming a fundamental aspect of modern life. There is a growing emphasis on improving the everyday driving experience and reaping the benefits that these advanced cars offer. In light of these advancements, this study proposes a novel method for extracting trajectories, which serves as a valuable tool in implementing and evaluating complex algorithms and functionalities. Obtaining a real trajectory enables the observation of how a vehicle moves along a given road. Moreover, the extracted trajectory can also facilitate the development of additional functionalities such as lane keeping, velocity profile, and other advanced techniques. This multifaceted approach allows for the exploration and refinement of various intelligent systems, ultimately contributing to the overall improvement of autonomous driving capabilities. The proposed method of trajectory extraction opens up new possibilities for advancing the field, providing a solid foundation for implementing and evaluating intricate algorithms. Through the utilization of real trajectories, valuable insights into vehicle behavior can be gained, which can further enhance the performance of autonomous vehicles.

Keywords: Fault Diagnosis and Fault Tolerant Control, Machine Learning, Industrial Applications
Abstract: This article proposes the use of Kernel Ridge Regression (KRR) for modeling and anomaly detection in the temperature control of textile dyeing processes. The anomaly being considered is the inability to heat or cool the dyeing machine temperature. It is concluded that anomaly detection results of the KRR model are satisfactory. The ultimate goal is to find problematic temperature controls rapidly and solve equipment failures of dyeing machines to increase right-first-time dyeing.

Keywords: Communication Systems, Real Time Applications, Internet of Things
Abstract: Time-Division Multiple Access (TDMA) medium access policy is widely used in wireless sensor networks where predictability is mandatory in communication. Such a policy is ideal for real-time applications but has an inefficient use of the communication channel. A significant improvement is given by the Greedy TDMA (GTDMA) access policy. In this paper, we present a complex simulation platform built on a popular sensor network simulator. We present an accurate analysis over the benefits of the GTDMA policy over the classic TDMA after intense simulations in many scenarios.

Keywords: Computer Architectures, Embedded Systems, Intelligent Systems
Abstract: This paper proposes a RISC-V extension, named SigWavy, meant to optimize the PWM control for general purpose or application specific designs. The RISC-V extension named above is a PWM Control Unit with a dedicated ISA extension set designed for configuring and driving up to 32 PWM signals. The extension is integrated into RiscPwm, an updated version of our previous work, the RisCanFd SoC, for taking advantage of CAN-FD, a massively used protocol in the areas of automation and mobility. Being configured with the dedicated ISA extension or with parameters directly extracted from CAN-FD commands, the proposed solution manages to configure/reconfigure PWM channels between 4.79x and 9.18x times faster than an ARM Cortex-M7 processor, although our SoC operates with a 6x lower frequency.

Keywords: Computer Architectures, Other Topics
Abstract: In this work, we present an original contribution which augments the Intel Nehalem multicore architecture with a selective high-latency arithmetic set-associative reuse buffer. The architecture is simulated using Sniper, which we adapted to estimate the power consumption, area of integration and chip temperature, including latency modifications for the newly added unit. The implementation of a set-associative reuse buffer is a new approach, along with the applicability in a multicore microprocessor, applied to long-latency arithmetical instructions targeting dataflow bottleneck and increasing CPU performance. Additionally, we have performed a manual design space exploration for the enhanced microarchitecture, by varying the associativity and the size of the proposed reuse buffer unit and evaluating the impact on the interested metrics. Our simulations on the Splash 2 benchmarks, revealed an average reuse rate up to 33.27% allowing a maximum speedup of 6.56%. While the energy consumption remains stable, we see an average chip temperature reduction of 2.8°C along with the increase in associativity.

Keywords: Internet of Things, Distributed Systems, Communication Systems
Abstract: The purpose of the Internet of Things is to gain momentum in all fields and to link devices equipped with sensors, processing power, software, and other technologies that can communicate and share data with other hardware through the Internet or other communications networks. This paper presents some different types of topologies used for CoAP protocol and analyzes their performance. The topologies we used for this paper are the following: box topology, linear topology, elliptical topology, and random topology. Our study shows differences of about 17% in power consumption concerning the topology chosen, which is not neglectable for the Internet of Things domain, where power consumption is a sensitive issue.

Keywords: Machine Learning, Agent - Based Systems
Abstract: This paper compares the performance of two reinforcement learning algorithms, Q-Learning and MAXQ-0, in learning to play an original game. An extension of MAXQ-0 algorithm, MAXQ-P is introduced, which enhances the variety of the tree nodes with simple, ordered and repetitive nodes. The hierarchical approach provided by MAXQ-P finds the optimal solution faster than the flat Q-Learning approach but converges more slowly. Furthermore, the performance of the MAXQ-P algorithm decreases after a certain number of episodes due to representation error in the weights of the model. To address this issue, the model is periodically tested with an exploration value of 0, and if the model successfully finds the solution, it is stored for future use. This study provides insights into the benefits and drawbacks of using hierarchical reinforcement learning algorithms for complex tasks and highlights the importance of carefully designing and training such algorithms for optimal performance.

Keywords: Machine Learning, Intelligent Systems, Agent - Based Systems
Abstract: The latest work in the field of deep reinforcement learning speaks highly about the advanced exploration techniques which avoid the greedy decisions of agents. Usually, reinforcement learning works by finding the optimal policy for a Markov Decision Process. In off-policy algorithms the agent learns a value function for this optimal policy, separate of the action choice, an example being the deep Q-learning algorithm. Algorithms based on a maximum entropy framework, like soft Q-learning, overcome the greedy behavior of the agent, effectively combining exploration and exploitation by adding an entropy term to the Bellman equation. This method, applied to the Lunar Lander environment, was compared to the classic deep Q-learning, using the same set of different random seeds and averaging multiple runs. An implicit exploration strategy proves to compensate for disturbances caused by intrinsic sources of non-determinism, such as random seeds. This paper highlights the sensitivity to intrinsic and extrinsic influences for deep reinforcement learning, with respect to exploration and repeatability.

Keywords: Biomedical Engineering, Machine Learning, Computer Vision
Abstract: Ultrasound imaging is an often employed technique in diagnosing of breast cancer, although the prediction reliability depends on the specialist’s experience. Computer Aided Diagnosis (CAD) systems have been introduced for the enhancement the quality and time invested in classifying breast ultrasound(BUS) images. Deep Convolutional Neural Networks based algorithms is considered one of the most successful strategy in breast ultrasound image analysis. Data limitation is one of the prioritizing issues at the current moment. This problem is referred by introducing transfer-learning-based models and stratification as a data augmentation technique for achieving a better accuracy of the classification. The paper has demonstrated that the deep feature extraction and feature selection can properly categorized the breast ultrasound images using the pre-training methods. A dataset containing 1578 breast ultrasound images was used for model training and testing, and the optimal level of achievement was reached by InceptionResNetV2 and DenseNet121 with an accuracy of 83% and an ”one over the rest” AUC score of 0.933 for DenseNet121, respectively 0.923 for InceptionResNetV2.

Keywords: Software Engineering, Machine Learning, Other Topics
Abstract: Successful software projects are now an important challenge, the main focus of the engineering community being to predict software failures based on the history of buggy classes. However, software defect prediction techniques are effective as long as there is enough data to train the prediction model. To mitigate this problem, cross-project defect prediction is used.

The purpose of this investigation is two-fold. First, replicate the experiments in the original articles and second, investigate other settings regarding defect prediction with the aim of bringing new insights and results on the best approach.

In this study, three supervised (Random Forest - RF, Logistic Model Tree - LMT, Naive Bayes - NB) and three unsupervised (Expectation Maximization - EM, DBSCAN, KMeans) approaches are investigated. The experiments used use preprocessed methods (normalization and feature selection). Two sets of experiments are performed considering all available features and using preselected features by Principal Component Analysis, each set of experiments being employed on both supervised and unsupervised methods.

The results of the replicated experiments confirm the original findings: when using supervised methods considering all features the best method is NB, followed by RF and LMT and similar to better results when considering fewer features (with PCA); when using unsupervised methods the results are not better than the original; however, when considering fewer features the results obtained with the newly considered methods (EM, DBSCAN, KMeans) are better than in the original paper.

Keywords: Machine Learning
Abstract: The paper presents a first series of results regarding the comparison of two deep neural networks designed, trained, and validated with the help of two deep learning frameworks, PyTorch and TensorFlow, in the classification of a large set of breast histopathology images. The original dataset consisted of 162 whole mount slide images of breast cancer specimens scanned at 40x. From that, 277,524 patches of size 50 x 50 were extracted (198,738 invasive ductal carcinoma negative and 78,786 positive). Invasive Ductal Carcinoma is the most common subtype of all breast cancers. The metrics compared in the training and validation phases were loss, accuracy, precision, recall and specificity. The architecture used for convolutional neural network has 8 layers: 3 convolution layers, 3 max-pooling layers, 1 linear layer and a sigmoid function. The number of epochs is considered a hyperparameter. It defines the number of times the entire data set must be worked through the learning algorithm. In our study we used 5 epochs. For the 8-layer network, PyTorch performed better in accuracy, precision, recall, and loss in both training and validation. In contrast, specificity is lower. Recall or sensitivity is around 73%, and specificity around 93%. Our estimates can be considered and continued since a diagnostic study must have both sensitivity and specificity of at least 70%. The results so far are promising, and we propose to continue the experiments in several directions.