Abstract

Autonomous systems have seen tremendous progress since the last decade and have the potential to become transformative military technologies in the coming few years. Future intelligence, surveillance and target acquisition systems would generate much larger volumes of real-time data which would be nearly impossible to process with human mental capacity. Machines would be challenged by the uncertainty and ambiguity in such unstructured data leading to difficulty in decision making. It must be understood that human operators would eventually be task saturated handling the unmanned systems and thus, there is a pressing need to utilize AI to free up human mental capacity in a flexible and adaptable way. This paper articulates how effective integration of soldiers, robots and artificial intelligence may contribute to future military advantage and warfighting systems. A cognitive analysis framework is presented to analyse the need for explicit allocation of cognitive responsibilities between the team members to achieve specific objectives while understanding the trade-off between performance and cognitive overload. A case study is presented to highlight the importance of flexible human robot interaction and demonstrate the impact of human robot collaboration in defence scenario. The paper enunciates key challenges for future ground forces for developing competent human-machine teams.
Full Paper

Abstract

As organizations keep investing in cyber security technologies generously due to growing demands for that technologies, human weaknesses remain a basis of data breaches across organizations network. Human weaknesses are still a hard challenge to overcome if organizations want a robust cyber security shield. In this research we tried to address human limitations that lead to data beaches or cyber-attacks against workplace network depending on literature reviews in the same field of the research. First, we write about cyber security as a technical matter where organizations invest huge budget to implement needed software and hardware to protect their information assets. Then we talk about cyber security as a human matter. After that we address human behavior related to security starting with improper practices related to password. The next topic is leaking information by employees or insiders intentionally or unintentionally. Next, we focus on the risk of accessing organization network using personal WIFI-enabled devices which suffers from lack of security software and updates and how they may be exploited penetrate the network. Then we talk about social engineering as a threat. Next topics are reaction to suspicious activity, cyber Awareness for employees as a shield, and cyber Security policy.
Full Paper

Abstract

Computer-aided breast cancer diagnostic systems’ credibility relies profoundly upon the accuracy of the models correlating the various enigmatic descriptors to the correct class labels. Consequently, breast H&E samples’ classification is one of the most crucial problems of computer vision in the medical field. As an outcome of the accelerated evolution in computational resources, Convolutional neural networks emerged to elicit more intricate features. Nonetheless, CNNs are data greedy and inclined to overfit in the medical field due to the deficient supply of labeled patches. Although Transfer Learning assists in reducing the massive training-data requirement of CNNs, it fails to separate domain-specific layers from the general ones, which often leads to the worst accuracy on unseen samples. The shortcoming mentioned earlier is often attributable to the lack of interpretability of the features extracted at each layer. Moreover, the high number of hyper-parameters associated with ensembles of pre-trained models greatly magnifies the search space, which substantially increases the training time. To overcome these problems, we propose an adaptive particle swarm optimization hyper-parameter selection method that focuses on deriving an ensemble rule for a fixed length of the best-trained models. We first fine-tune a set of pre-trained models on low-resolution images and determine the best combination of them based on the variance of classification errors amongst them rather than their validation loss. Then, we use a PSO whose learning rate is adjusted with a Mamdani fuzzy inference system to infer the ensemble policy. Finally, we train the ensemble thoroughly as a noisy-student model using hyper-parameters obtained from the two search sessions with high-resolution and weakly classified images. The outcomes of our method were compared to different state-of-the-art techniques implemented on the ICIAR 2018 dataset. The proposed approach substantially outperformed previously published procedures.
Full Paper

Abstract

This paper discusses key artifacts generated by Microsoft Windows which may be examined during a forensic investigation. A brief history of Microsoft Windows operating system is presented, followed by a description of the internal structure of hard disks. An analysis of artifacts created due to user’s interaction with applications and the subsequent interfacing of these applications with the operating system and the underlying hardware (primarily the hard disk and volatile memory) are detailed out from an investigator’s perspective.
Full Paper