Healthcare is a fundamental part of every individual's life. The healthcare industry is developin... more Healthcare is a fundamental part of every individual's life. The healthcare industry is developing very rapidly with the help of advanced technologies. Many researchers are trying to build cloud-based healthcare applications that can be accessed by healthcare professionals from their premises, as well as by patients from their mobile devices through communication interfaces. These systems promote reliable and remote interactions between patients and healthcare professionals. However, there are several limitations to these innovative cloud computing-based systems, namely network availability, latency, battery life and resource availability. We propose a hybrid mobile cloud computing (HMCC) architecture to address these challenges. Furthermore, we also evaluate the performance of heuristic and dynamic machine learning based task scheduling and load balancing algorithms on our proposed architecture. We compare them, to identify the strengths and weaknesses of each algorithm; and provide their comparative results, to show latency and energy consumption performance. Challenging issues for cloudbased healthcare systems are discussed in detail.
Increasing availability of location-based applications and sensor devices have necessitated quick... more Increasing availability of location-based applications and sensor devices have necessitated quicker analysis of moving object data streams in order to identify patterns. The efficiency of currently available algorithms used in pattern detection is not adequate to handle large scale data streams that are increasingly available. We focus on the particular problem of flock detection in moving object data and our goal is to detect flocks quickly and using fast algorithms. Firstly, we employ a triangular grid to reduce the search space of clustering algorithms which has a significant effect in case of dense objects. As a second step, we implement a modified flock membership function and pipeline creation that ensures better memory and time performance during cluster detection. We show that this refinement also improves the rate of flock detection. Finally, we parallelize our algorithm to further enhance the handling of massive data streams. Based on an extensive empirical evaluation of these algorithms across a variety of moving object data sets, we show that our method is significantly faster than the existing comparable methods over sliding windows. In particular, it requires lesser time to identify flocks and is 2-4 times faster thus confirming the efficiency and effectiveness of our approach.
Growing technologies like virtualization and artificial intelligence have become more popular now... more Growing technologies like virtualization and artificial intelligence have become more popular nowadays because they are more handy and accessible on mobile devices. But lack of resources for processing these applications at the user end and the limited energy of mobile devices are still significant hurdles. Collaborative edge and cloud computing are one of the solutions to this problem. An optimal offloading strategy is required to balance transmission latency for the cloud and limited resources at edge servers. We have proposed a multi-period deep deterministic policy gradient (MP-DDPG) algorithm to find an optimal offloading policy to the collaborative cloud network including the central cloud server, edge cloud servers, and mobile devices constrained by minimization of computation, transmission delay, and energy consumption. The novelty of this algorithm lies in partitioning the task to offload in multiple time slots and reusing cloud and edge resources in every slot, rather than taking a single offloading decision and running out of remote resources by offloading a single large task. Our results show that MP-DDPG achieves the minimum latency and energy consumption in the collaborative cloud network. INDEX TERMS Collaborative cloud computing, computation offloading, latency, energy efficiency, deep reinforcement learning, multi-period deep deterministic policy gradient
Emotion Recognition plays an important role in understanding human behavior. It finds its utility... more Emotion Recognition plays an important role in understanding human behavior. It finds its utility in various domains such as healthcare, automobile industries, understanding social interactions, fraud detection, and many more. Analyzing a person's emotions in a controlled environment with various devices has been challenging since it adds to human anxiety, which manipulates the readings. This presents a need to devise ways to recognize and study emotions in a wireless manner. We devised a system that recognizes the emotions using Heart Rate Variability (HRV) of the subjects which is estimated from their videos using Remotephotoplethysmography(rPPG). Our emotion recognizer has 93.27% accuracy.
Increasing availability of location-based applications and sensor devices have necessitated quick... more Increasing availability of location-based applications and sensor devices have necessitated quicker analysis of moving object data streams in order to identify patterns. The efficiency of currently available algorithms used in pattern detection is not adequate to handle large scale data streams that are increasingly available. We focus on the particular problem of flock detection in moving object data and our goal is to detect flocks quickly and using fast algorithms. Firstly, we employ a triangular grid to reduce the search space of clustering algorithms which has a significant effect in case of dense objects. As a second step, we implement a modified flock membership function and pipeline creation that ensures better memory and time performance during cluster detection. We show that this refinement also improves the rate of flock detection. Finally, we parallelize our algorithm to further enhance the handling of massive data streams. Based on an extensive empirical evaluation of these algorithms across a variety of moving object data sets, we show that our method is significantly faster than the existing comparable methods over sliding windows. In particular, it requires lesser time to identify flocks and is 2-4 times faster thus confirming the efficiency and effectiveness of our approach.
Uploads
Papers by jui mhatre