I investigate properties of HMAX, a computational model of hierarchical processing in the primate visual cortex. High-level cortical neurons have been shown to respond highly to particular natural shapes, such as faces. HMAX models this property with a dictionary of natural shapes, called prototypes, that respond to the presence of those shapes. The resulting set of similarity measurements is an effective descriptor for classifying images. Curiously, prior work has shown that replacing the dictionary of natural shapes with entirely random prototypes has little impact on classification performance. This work explores that phenomenon by studying the performance of random prototypes on natural scenes, and by comparing their performance to that of sparse random projections of low-level image features.

No description available.

No description available.

No description available.

No description available.

A dissertation submitted to the Faculty of Engineering and the Built Environment, University of the Witwatersrand, Johannesburg, in fulfilment of the requirements for the degree of Master of Science in Engineering, August 1,2019 / Nowadays, most organizations and platforms employ an intrusion detection system (IDS) to enhance their network security and protocol systems. The IDS has therefore become an essential component of any network system; it is a tool with several applications that can be tuned to specific content in a network by identifying various accesses (normal or attack). However, network intrusion detection system (NIDS) that focuses on revealing suspicious activities, is not effective in solving various problems such as identifying false IP packets and encrypted traffic. Hence, this work investigates the use of ensemble learning to solve these types of network intrusion detection problems (NIDPs). Random forest (RF), Decision Tree (DT) and Support Vector Machine (SVM) are introduced as classifiers based on Boruta and Principal Component Analysis (PCA) algorithms. In general, the main difficulties in using ensemble for the intrusion problem are to minimize false alarms and to maximize detection accuracy (Anuar et al., 2008). Additionally, the NIDP is divided into five categories, namely the detection of probe attacks, denial of service, remote to local, user to root and normal instances. Each problem is examined by one of the three aforementioned classifiers. In tackling these problems, the three classifiers achieved competitive results comparing to the works conducted by Balon-Perin (2012), Zainal et al. (2009) and Kevric et al. (2017). The results revealed that ensemble learning achieved more than 99% accuracy in demarcating attacks from normal connections. Particularly, RF, DT and SVM allowed to safeguard the NIDS from known and unknown attacks by developing reliable techniques. The KDD99 and NSL KDD datasets have been used to implement and measure the system performance (Fan et al., 2000; Dhanabal and Shantharajah, 2015). / PH2020

Image categorization is often performed manually, which can be a time consuming and a very difficult process, especially for human iris images. Previous researchers have been working on predicting ethnicity from texture features of iris images using other methods. This thesis is one of the the first to present a solution of iris image categorization using artificial neural networks, specifically for human iris images with discernible and complicated textures. The work will allow users to quickly and automatically categorize human iris images by using supervised and unsupervised learning algorithms. Contributions of this solution include a fast and accurate way to apply iris matching and solve the time consuming problems. The solution aims to find efficient and appropriate artificial neural network algorithms that can categorize iris images based on texture features. Detailed algorithms, specific techniques, performance analysis, limitations and future work will be also provided in this thesis.

Due to some technical and environmental constraints, expanding the current electric power generation and transmission system is being challenged by even increasing the deployment of distributed renewable generation and storage systems. Energy storage can be used to store energy from utility during low-demand (off-peak) hours and deliver this energy back to the utility during high-demand (on-peak) hours. Furthermore, energy storage can be used with renewable sources to overcome some of their limitations such as their strong dependence on the weather conditions, which cannot be perfectly predicted, and their unmatched or out-of-synchronization generation peaks with the demand peaks. Generally, energy storage enhances the performance of distributed renewable sources and increases the efficiency of the entire power system. Moreover, energy storage allows for leveling the load, shaving peak demands, and furthermore, transacting power with the utility grid. This research proposes an energy management system (EMS) to manage the operation of distributed grid-tied battery micro-storage systems for stationary applications when operated with and without renewable sources. The term "micro" refers to the capacity of the energy storage compared to the grid capacity. The proposed management system employs four dynamic models; economic model, battery model, and load and weather forecasting models. These models, which are the main contribution of this research, are used in order to optimally control the operation of the micro-storage system (MSS) to maximize the economic return for the end-user when operated in an electricity spot market system. Chapter 1 presents an introduction to the drawbacks of the current power system, the role of energy storage in deregulated electricity markets, limitations of renewable sources, ways for participating in spot electricity markets, and an outline of the main contributions in this dissertation. In Chapter 2, some hardware design considerations for distributed micro-storage systems as well as some economic analyses are presented. Chapters 3 and 4 propose a battery management system (BMS) that handles three main functions: battery charging, state-of-charge (SOC) estimation and state-of-health (SOH) estimation. Chapter 5 proposes load and weather forecasting models using artificial neural networks (ANNs) to develop an energy management strategy to control the operation of the MSS in a spot market system when incorporated with other renewable energy sources. Finally, conclusions and future work are presented in Chapter 6.

No description available.

The production control system of a shop consists of three stages: due-date prediction, order release, and job dispatching. The literature has dealt thoroughly with the third stage, but there is a paucity of study on either of the first two stages or on interaction between the stages. This dissertation focuses on the first stage of production control, due-date prediction, by examining methodologies for improved prediction that go beyond either practitioner or published approaches. In particular, artificial neural networks and regression nonlinear in its variables are considered. In addition, interactive effects with the third stage, shop-floor dispatching, are taken into consideration. The dissertation conducts three basic studies. The first examines neural networks and regression nonlinear in its variables as alternatives to conventional due-date prediction. The second proposes a new cost-based criterion and prediction methodology that explicitly includes costs of earliness and tardiness directly in the forecast; these costs may differ in form and/or degree from each other. And third, the benefit of tying together the first and third stages of production control is explored. The studies are conducted by statistically analyzing data generated from simulated shops. Results of the first study conclude that both neural networks and regression nonlinear in its variables are preferred significantly to approaches advanced to date in the literature and in practice. Moreover, in the second study, it is found that the consequences of not using the cost-based criterion can be profound, particularly if a firm's cost function is asymmetric about the due date. Finally, it is discovered that the integrative, interactive methodology developed in the third study is significantly superior to the current non-integrative and non-interactive approaches. In particular, interactive neural network prediction is found to excel in the presence of asymmetric cost functions, whereas regression nonlinear in its variables is preferable under symmetric costs. / Ph. D.