Transfer Learning has become one of the standard methods to solve problems to overcome the isolated learning paradigm by utilizing knowledge acquired for one task to solve another related one. However, research needs to be done, to identify the initial steps before inducing transfer learning to applications for further verification and explainablity. In this research, we have performed cross dataset analysis and network architecture repair for the lane detection application in autonomous vehicles. Lane detection is an important aspect of autonomous vehicles’ driving assistance system. In most circumstances, modern deep-learning-based lane recognition systems are successful, but they struggle with lanes with complex topologies. The proposed architecture, ERF-CondLaneNet is an enhancement to the CondlaneNet used for lane identification framework to solve the difficulty of detecting lane lines with complex topologies like dense, curved and fork lines. The newly proposed technique was tested on two common lane detecting benchmarks, CULane and CurveLanes respectively, and two different backbones, ResNet and ERFNet. The researched technique with ERF-CondLaneNet, exhibited similar performance in comparison to Resnet-CondLaneNet, while using 33% less features, resulting in a reduction of model size by 46%.

The heterogeneous nature of the internet-of-thing (IoT) is gaining popularity and, simultaneously, faces rising security issues. The distributed denial of service (DDoS) attack is the most significant security threat addressed in the research. The research proposes edge-heterogeneous IoT (HetIoT) centric defense IDS that aids the HetIoT infrastructure in detecting and blocking victim traffic near the network edge. The Edge-HetIoT defense IDS helps to address significant issues such as performance and security due to proximity to the local network . The research focuses on six learning techniques, including five machine learning (ML) classifiers, namely, ID3, NB , RF , LR , and AdaBoost , and the proposed deep learning (DL)-based hybrid model (i.e., CNN+LSTM). These learning techniques are trained and tested using the real-time benchmark-dataset CICDDoS2019 and consider binary and multiclass (14 classes) classification. The performance is analyzed and evaluated against six classifiers to determine which classification model performs best in detecting and classifying various DDoS attacks. The proposed DL-based hybrid model outperforms when compared against ID3, NB, RF, LR, and AdaBoost. The proposed DL-based hybrid model successfully detects and classifies MSSQL , NetBIOS, TFTP, NTP, Syn, and Portmap attacks with 100% precision, recall, and f1-score. The overall weighted average precision, recall, and f1-score for the proposed DL-based hybrid model are 92%, 89%, and 90%, respectively.

In this paper, we have developed minimum Total mean squared error (MSE) joint linear transceiver design for Single-user multiple-input, multiple-output (SU-MIMO) systems employing an improper constellation schemes. It assumes Imperfect channel state information (CSI) is available at both ends of transmitter and receiver and design transceiver is subject to a total power constraint. Channel mean and transmit correlation information are consider as channel state information. Joint design of transceiver is consider as optimization problem, the optimum closed-form precoder and decoder are derived by minimizing a total MSE based on modified cost function. The performances degradations due to imperfect channel estimation and/or transmit correlation are demonstrated and superiority of proposed method is proved by simulation results.