Mahishadal Raj College, established in 1946, is the second oldest college in Purba Medinipur district. It offers undergraduate courses in arts, commerce and sciences. It is affiliated to Vidyasagar University.
An inventory model of a decaying item is proposed considering dynamic credit period to the consumers, fuzzy life time of the units, time-dependent fuzzy decaying rate, and freshness dependent dynamic imprecise demand. Decaying rate of units depends on its lifetime and is considered as a dynamic monotonic increasing function of the same. To encourage the consumers, the retailer provides a credit on a part of the purchase and due to the dynamic decreasing nature of the freshness of units, the credit interval is considered as dynamic increasing. The unit cost, consumer credit period, and credit amount have impact on demand. As demand is fuzzy, some units may left at any cycle ending and are put up for sell in lot in a lump sum price. The model is formulated using fuzzy mathematical tools to decide the optimal cycle length, unit selling price, and order quantity to optimize the annual fuzzy profit from a cycle. A modified form of particle swarm optimization is applied to get the best decision, in which, using fuzzy credibility measure the objectives corresponding to two feasible solutions are compared to select the better one. Proper fuzzy simulation approaches are followed to determine the required credibility measures.
Human activity recognition with data privacy protection and low response time is a significant research domain. A human activity recognition model based on Gaussian differential privacy (GDP) and federated learning (FL) is proposed in an edge-cloud computing environment. We consider two different cases of GDP-based FL method. In case 1, Gaussian noise is clipped with the local datasets of the edge devices participating in the FL process, and in case 2, Gaussian noise is clipped with the local model updates during the FL process. In both cases, leakage of original information is prevented by adding noise to the local datasets or local model updates. We perform activity data collection and analyze the collected data for performance evaluation. The results illustrate that for our dataset, the proposed activity recognition model achieves a prediction accuracy of 0.82-0.84 and 0.83-0.84 for case 1 and case 2, respectively. The results also illustrate that the proposed GDP-based FL approaches achieve above 80% prediction accuracy for well-known activity recognition datasets. The results also show that the proposed approaches have 13-17% and 54-57% lower response time than the edge-cloud and cloud-only frameworks, respectively.
A novel hyperheuristic is designed for solving Generalized Traveling Salesman Problems (GTSP) comprising real and imprecise cost matrices. A novel procedure named K-node rearrangement (KNR) is designed for group sequencing and several strategies are designed for the optimal selection of destinations from distinct groups of the same. The operations are iteratively applied in a nested way for an optimal schedule. 3-opt is applied periodically to enhance the quality of the output. Its efficiency is examined using instances from GTSPLIB having a maximum of 89 groups and 442 destinations, achieving 100
The Stackelberg game strategies are outlined for a wholesaler-retailer-consumer supply chain of an item in a finite planning schedule, where consumers' demand is influenced by price, credit period, and inflation. As the profit of each party depends on the consumers' demand, the retailer offers some credit period to its consumers and the wholesaler shares this burden by offering some credit period to the retailer. The retail price depends on the selling price of the wholesaler and both price are decision variables. It is also assumed that a portion of consumers are defaulters, and is increased with the length of the credit period. The model is formulated to optimise individual profits from the planning horizon following the Stackelberg game approach, where the wholesaler is the leader and the retailer is the follower. The performance of Artificial Bee Colony algorithm is enhanced by the introduction of four additional perturbation rules and their collaborative use following a Q-learning strategy. The algorithm is implemented, tested, compared with state-of-the-art algorithms, and applied in a nested way to determine the optimal policies of both players simultaneously. The model is illustrated using different numerical examples, and some managerial insights are outlined.
