In the resource constrained delay sensitive collaborative computing paradigms of Industrial Internet of Things (IIoT), Deep Neural Network (DNN) splitting schemes are promising. Multi-application multi-task (MAMT) scenarios in AI enabled industries bring novel challenges in deciding the split criterion for DNN tasks as opposed to Single -application multi-task (SAMT). This work proposes a novel algorithm that takes into account the varying architectures of DNN tasks and model them as Multi-choice Multi-dimensional Knapsack (MMKP) problem. Layer profiling is used to create an initial greedy solution and to optimize it this work adapts epidemic theory on how a pathogen infection follows an optimal path to spread itself in determining the optimal split solution for MAMTs. The complexity of O(mn) signifies that our approach for MAMTs is efficient compared to that of SAMT with similar complexity. Results show that the proposed approach performs optimally and at par with the prevalent approaches for both SAMT and MAMT.

Industries increasingly depend on deep learning-based predictive quality systems to anticipate product quality using process data. However, the dynamic nature of production processes poses a significant challenge. These changes can render previously trained models ineffective, and thus, Memory Aware Synapses (MAS)-based continual learning approach for edge-enabled industries is adopted. This work extends MAS with knowledge transfer and a time-efficient MAS hyperparameter selection mechanism using a hybrid grid search and Particle Swarm Optimization mechanism. The experimental analysis demonstrates that the proposed mechanism reduces the overall model loss by 39.03% and 27.18% compared to MAS mechanisms, respectively, for two real-world use cases. An Android-based mobile application is designed as a ready-to-use solution for the mobile edge device.

Edge-enabled Deep Learning (DL) solutions for Predictive Quality Inspection (PQI) of products in Industry 4.0 are mostly designed for static manufacturing environments. In general, modern manufacturing processes are dynamic in nature. In this context, continual learning-based model retraining accommodates the dynamism for PQI of multiple processes (tasks) using a single DL model. However, the impact of the task ordering in sequentially arriving tasks and solution to reduce this impact on the overall PQI is yet to be solved. To this end, a novel mechanism using a light-weight similarity analysis module is introduced in the quality prediction system at the resource-limited edge. Sequential training of tasks above a similarity threshold (γ) is preferred, and dissimilar tasks are overlooked to train a separate model. This enables a PQI system to hover over training efficiency and model sustainability. The experimental results validate the impact of task order and the effectiveness of the proposed similarity-based analysis to reduce this impact by 70% on the model's overall performance in the real-world use case of plastic bricks.

Mobile sinks were introduced in wireless sensor networks (WSNs) to mitigate the infamous hotspot problem. However, routing in mobile-sink-based WSNs requires frequent updation of sink location information to all the sensor nodes; which is an energy-expensive process for resource-constrained WSNs. Therefore, it is required to develop a green routing protocol that can minimize the energy overhead in sink location updation as well as reduce the data delivery delay. This article proposes a virtual-infrastructure-based delay-aware green routing protocol (DGRP) that creates multiple rings in the sensor field and limits the updation of mobile sink location information to the nodes belonging to the rings only. Simulation results show that DGRP outperforms existing routing protocols in terms of energy consumption and throughput. In addition to this, DGRP results in $\approx 26$ %, $\approx 39$ %, and $\approx 35$ % improvement in data delivery delay for a varying number of sensor nodes, sink speeds, and network sizes, respectively, when compared with the state of the art.

In today's world, Network and System Security are of paramount importance in the digital communication environment. To avoid breaches, it is badly needed for a security administrator to detect the intruder and prevent him from entering into the network. Machine Learning techniques are used to solve these types of problems, but they are not highly able to generalize as they fail to obtain relation among the features. Several works have also been done in Deep Learning using Artificial Neural Networks, Deep Neural Networks, RNN, etc. are not computationally efficient. This paper suggests a new machine learning model for intrusion detection that uses LSTMs and CNNs. This work uses CNN to choose feature characteristics from the input data, and send these features to LSTM for sequence analysis, and to address the imbalanced data set problem, Based on the total number of training examples in each class, each example will have its weight calculated based on cost function method. The raw input data format is transformed into a matrix format(image) to further decrease the computation cost. To test the efficiency of the CNN-LSTM model this work uses a conventional NSL-KDD dataset. The computation time has been reduced to 1/10ththe time that a fully connected layer took to train. The experimental results show that the model achieves an accuracy of 99.6% and a Detection rate of 96.75% while training.

From many e-commerce websites such as Amazon, Flipkart and Netflix to online advertisements, recommender systems are used to recommend certain products to their users depending on their preferences and interactions with the products. Sequential recommender system is useful to model the short term behavior of the users depending on their latest interactions. Many techniques ranging from Markov chain models to convolutional neural networks (CNNs) have been used to solve the sequential recommendation problem. The most efficient state-of-the-art-model that uses CNNs is the top-N sequential recommendation system [1]. This model fails to capture the skip behavior in the union-level patterns as it doesn't take the interaction between distant items in a sequence into consideration. This paper adopts the basic architecture proposed in Caser [1] and introduction of the joint embedding of two items, formation of a 3D tensor with all the embeddings and a introduction of a new convolutional block have been made. This novel approach improves the mean average precision (MAP) from 0.1507 to 0.1884 which is significant and gives rise to the most efficient sequential recommendation system that uses CNN.

Shared Sensor Network (SSN) refers to a scenario where the same sensing and communication resources are shared and queried by multiple Internet applications. Due to the burgeoning growth in Internet applications, multiple application queries can exhibit overlapping in their functional requirements, such as the region of interest, sensing attributes, and sensing time duration. This overlapping results in redundant sensing tasks generation leading to the increased overall network traffic and energy consumption. Existing approaches operate on data sharing among various tasks to minimize the upstream traffic. However, no existing work attempts to prevent the redundant task generation to reduce the downstream traffic. Moreover, the allocation of suitable sensor nodes to meet the Quality of Service (QoS) requirements of the queries is still an open issue. This article proposes an end-to-end query processing framework (named, QueryPM) that first, calculates the functional requirements similarity among queries to prevent the redundant task generation. Then, it takes the QoS and functional requirements into account while allocating the tasks on the sensor nodes. Extensive simulations on the proposed approach show that downstream traffic, upstream traffic, and energy consumption reduced to 60%, 20--40%, and 40%, respectively, as compared to state-of-the-art mechanisms.

There are two major challenges in mobile sink based wireless sensor networks (WSNs) concerning to the query-driven scenarios, first, dissemination of queries to their respective region of interests (RoIs), and second, routing the data towards mobile sink. Due to sink mobility, routing of data packets to the sink becomes difficult because sink's query injection location and data collection location (current location of sink) may not be the same. Moreover, advertising mobile sink's location by flooding, introduces extensive burden on sensor nodes. In this paper, we propose a virtual ring infrastructure based query-driven ring routing protocol (QRRP) to reduce the overhead of updating mobile sink location information as well as routing the data towards current location of the sink. QRRP takes the advantage of proposed angle based routing to route the queries from mobile sink to their respective RoIs, and data from sensor nodes to the sink. Simulation results on the proposed approach show that energy consumption and data delivery delay are significantly reduced as compared to state-of-the-art mechanisms.