Aya Shabbar

This paper discusses the importance of value distribution in a multi-agent system in a distributional temporal-difference (TD) reinforcement learning network. Value distribution is the distribution of the random return received by a reinforcement learning agent. Temporal-difference (TD) learning, on the other hand, is a fundamental method for solving the reinforcement learning problem, and it can tackle the temporal credit assignment problem. Whereas standard TD learns a single prediction-the average expected rewarda distributional TD network learns a set of distinct predictions. Each of these is learned through the same method as standard TDby computing a reward prediction error that describes the difference between consecutive predictions. Consequently, this research is an in-progress study that aims to study a multi-agent system under a distributional TD network, with respect to the value distribution. The research, therefore, holds a theoretical and a mathematical form for a future development.

In this paper, Dynamic Programming Algorithms are used for evaluating and finding the optimal policies. More precisely, this research focuses on the use of some specific concepts such as MDPs and the Bellman Equations to determine how good a given policy is and how to find an optimal policy in a Markov Decision Process. DP is a general approach to solving problems by breaking them into sub-problems that can be solved separately, cached, then combined to solve the overall problem. The two required properties of Dynamic Programming that this research rely on are both optimal substructure and overlapping subproblems, which are satisfied both by Markov Decision Process.

Determining Causality across variables cam be challenging step but it is important for strategic actions. In other words, the problem of learning causal influences from data has attracted much attention. Standard statistical methods can have difficulty learning discrete causes, which interacting to affect a target, because the assumptions in these methods often do not model discrete causal relationships well. An important task then is to learn from data. In this paper, I will summarize the concepts of causal models in terms of Bayesian probabilistic, with a case study to detect causal relationships using Bayesian structure learning. This paper, additionally, relies on Sprinkler dataset to conceptually explain how structures are learned with the use of Python library bnlearn.

The use of object detection algorithms is becoming increasingly important in autonomous vehicles, and object detection at high accuracy and a fast inference speed is essential for safe autonomous driving. A false positive (FP) from a false localization during autonomous driving can lead to fatal accidents and hinder safe and efficient driving. Therefore, a detection algorithm that can cope with mislocalizations is required in autonomous driving applications. This paper proposes a method for improving detection accuracy while supporting a real-time operation by applying YOLOv3, which is the most representative of one-stage detectors, with redesigning the loss function. In addition, by using the predicted localization uncertainty during the detection process, the proposed schemes can significantly reduce the FP and increase the true positive (TP), thereby improving the accuracy. Compared to a conventional YOLOv3-tiny, the proposed algorithm, YOLOv3, improves the mean average precision (mAP). Nevertheless, the proposed algorithm is capable of real-time detection faster than 42 frames per second (fps) and shows a higher accuracy than previous approaches with similar fps. Therefore, the proposed algorithm is the most suitable for autonomous driving applications.
Keywords: object detection, YOLOv3, real-time systems, keras, deep learning

This paper presents an obstacle avoidance approach for e-puck module by using Fuzzy Logic controller.
The input from eight (8) IR sensors and the output of the motor speed will be used to construct the Fuzzy
Logic rules. Test environment, e-puck robot and Fuzzy algorithm was model and programmed with C#
programming language using Rider software. The Fuzzy system for e-puck robots was validated in a few
environments. The result shows the e-puck module can avoid those static obstacles successfully until it
reaches a goal point. The robot performance in terms of distance and time was recorded when the robot
works in simple, average and complex obstacle environments.

Deep Learning and Image Processing is a key concept in today's world of computational art, where artists employed AI algorithms to generate visuals. This paper explores AI-generated images, using Convolutional Neural Networks software as a paradigm of symbolic AI creative systems, and contextualizes the use of modern image processing technologies to create visual artworks. It discusses the methodologies and strategies used to make art using AI algorithms, manipulating them with Processing software tool. The discussion focuses on CNN (Convolutional Neural Network) and Processing software (Java) as the main technologies used in distinct fields to generate images. My conception of technical images provides a conceptual framework for examining the qualities and attributes of AI-generated images.