Md Abdullah Al Hafiz Khan

Graduate Research Projects

• Scalable Human Activity Recognition

  Proposed Transfer learning enabled Convolutional Neural Network (CNN) to recognize Human Activity. Accelerometer sensors signals are fed in the CNN as input. Eight (8) activities are cognized without or a limited amount of labeled activity data in the target domain.

  Tools & resources: Python, Pandas, Numpy, Scipy.

• Radar Based Activity Recognitino (RAM)

  In this project, we investigated radar sensing technology to detect high-level activity like walking, cooking, brushing etc. I developed our recognition algorithm. Processed the received array of radar sensor signals. Extracted meaningful features from the pre-processed signal and applied machine learning algorithm to recognize human activities.

  Tools & resources: Python, Numpy, Scipy, Bamble Bee Radar Sensors.

• Collaborative Opportunistic Human Activity Recognition (COAR)

  In this project, I proposed a finite state machine based novel model for opportunistically selecting different sensor data according to the activity context when the user is equipped with multiple smart devices. The finite state machine controls the data source selection which in our case is either the smartphone or the smartwatch. For inferring the activity context of the user, I used maximum entropy based classifier as the conditional independence of the features are not known.

  Tools & resources: Android, Python

• Sense Presence

  In this project, I developed a novel speaker estimation algorithm based on unsupervised clustering of overlapped and non-overlapped conversational data to determine the number of occupants in a crowded environment. I also designed a hybrid approach combining acoustic sensing opportunistically with locomotive model to further improve the occupancy detection accuracy.

  Tools & resources: Python, Numpy, SciPy.

• Appliance State Identifications

  In this work, I proposed novel techniques for detecting appliance and its state using acoustic signature of the appliance. We correlate an appliances inherent acoustic noise with its energy consumption pattern individually and in presence of multiple appliances. We initially investigated classification techniques to establish the relationship between appliance power and acoustic states for efficient energy disaggregation and abnormal events detection. we propose a probabilistic graphical model, based on a variation of Factorial Hidden Markov Model (FHMM), for multiple appliances energy disaggregation to improve the accuracy of our disaggregation algorithm.

  Tools & resources: Python, Numpy, Scipy.

Undergraduate Research Project

• Admission Control Algorithm for reliable web service

  Server admits a request when it can reserve a sufficient amount of resource to achieve the desired quality. If not, broker [Broker takes the decision of admission or rejection of a client into the system. The decision process is based on the maximization of total satisfaction of all clients.] Should contact another server to make the request. There are thousands of services, are available for selection. Searching and finding the most suitable service to match the clients functional and QoS requirements may be better performed by an automated system module such as QoS broker, which contacts the selected servers to get QoS information to make the final decision according to QoS information. QoS broker is a mediator between a server and client during the service establishment phase, performing service admission and resource assignment based on both client QoS requirement and several loads. Our topic mainly concerns on the device an algorithm that select reliable server for services.

  Tools and Resources: Java