I am part of the Advanced Processor Technology (APT) group here at The University of Manchester and I am supervised by Mr. Graham Riley and Dr. Mikel Lujan . My research is part funded by ARM and IS-ENES2 Project.
My research topic is "Efficient execution of Convolutional Neural Networks on low power heterogenous systems". If I had to sum up my research in three words, it would be Convolutional Neural Networks ( a.k.a "ConvNets"), Heterogenous Computing and Energy-Efficiency.
- ConvNets I became interested in Neural Networks during my undergraduate degree. Back then, I designed a Neural Net to act as a adaptive filter to cancel echoes from a voice signal with trailing echoes. Later, I joined The University of Manchester to do my Masters and my disseration revolved around "Transfer Learning" and "ConvNets" in the context of facial information processing. This project focussed on understanding whether transferring features in deep ConvNets would result in better feature extraction due to disentangling of informtation components (such as a person's identity, person's expression) found in the raw pixel representation of an image. Now, I am interested to get down to the nitty-gritty of the implementation of ConvNets and its execution behaviour on resource constrained devices.
- Heterogenous Computing We are now in the Many-core era with CPUs, GPUs, FPGAs all working together. I am interested in efficiently making use of these resources in the deploying algorithms such as ConvNets. One of the development boards I will get to play around with, is the Jetson TX1 that contains NVIDIA Maxwell GPU with 256 NVIDIA CUDA Cores and a Quad-core ARM Cortex-A57 MPCore Processor.
- Energy Efficiency I am interested in techniques to measure energy consumption during the inference of a ConvNet and mapping code phases to its energy profile and developing methods to exploit this information to enable better energy efficiency.
Teaching AssistantApart from my research, I work part-time as a Teaching Assistant for the following modules:
- COMP61011 Foundations of Machine Learning (PG)
- COMP61021 Modelling and Visualization of High Dimensional Data (PG)
- COMP60611 Parallel Programs and their Performance (PG)
- COMP60621 Designing for Parallelism and Future Multi-core Computing (PG)
- COMP14112 Fundamentals of Artificial Intelligence (UG)
- COMP61342 Computer Vision (PG)
- COMP37212 Computer Vision (UG)
- COMP-PGT-INTRO COMP-PGT-INTRO (PG)
Conferences and Summer Schools
- GPU Technology conference (GTC) - April 2016
- Advanced Computer Architecture and Compilation for High-Performance and Embedded Systems (ACACES) - July 2016
- Poster presenter & 2 min oral talk - ARM Research Summit- September 2016
- Lisbon Machine Learning Summer School (LxMLS) - July 2017
- Poster presenter - ARM Research Summit - September 2017
- Extended abstract publication & Poster presenter- IEEE International Symposium on Workload Characterization (IISWC) - October 2017
- Poster presenter & 2 min oral talk - Neural Information Processing Systems Conference (NIPS) - December 2017
- FastPath Workshop Speaker International Workshop on Performance Analysis of Machine Learning Systems, ISPASS - April 2018
Student BloggerI also actively post monthly blogs about my experiences as a postgraduate research student.
Student VolunteerI engage in fund-raiser activities with Manchester RAG and open day activities as CS student volunteer. I was also a Business Ambassador for the TeenTech event held in Manchester, April 2016.
- Student school and faculty representative
IEEE Xplore (2017)
Fine-grained energy profiling for deep convolutional neural networks on the Jetson TX1
IEEE Xplore (2014)
Echo Canceller Using Error Back Propagation Algorithm
International Journal of Computer Applications (2014)
Performance Evaluation of Error Back Propagation Algorithm for Non-Linear Classification and Function Approximation in VHDL Platform