First month as a PhD student
I wanted to write a post to organise my thoughts after my first full month as a PhD student. Additionally, I've been required to explain my research area to many people with varying degrees of expertise from zero to hero. So, I'm using this opportunity to kill two birds with one stone.
What is my research about? (for non-specialists)
My research involves applying computer algorithms from the field of artificial intelligence (AI) in the context of everyday items containing embedded computing devices collectively known as the internet of things (IoT). IoT devices such smart energy monitors or security cameras often require real time data analysis capabilities to respond to peaks in electrical load, or to track a moving object for example. It's infeasible to send this kind of data to servers in remote data centres due to constraints on bandwidth and the time it takes to send a request and receive a response. As IoT devices become more mainstream, the issue of processing data closer to source will be paramount to ensure that smart devices can perform their "smart" tasks. I'm searching to find a solution where these "smart" tasks can be performed in-situ rather than by the cloud. This will most likely be achieved by applying specific algorithms suited to running on small, resource-constrained IoT devices and by distributing the machine intelligence decision making tasks amongst many connected IoT devices.
What is my research about? (for academics in my field and other techies)
I'm researching how to apply deep learning (DL) techniques in internet of things applications aiding the provision of real-time analytics and decision making capabilities at the edge of the network. This will alleviate the strain on bandwidth and reduce latency by circumventing the necessity to send large amounts of data to the cloud for processing. With many billions of devices coming online in the internet of things in the next decade, this is of vital importance to facilitate the nature of smart devices. The main challenge around deploying deep learning in IoT is that the devices have limited storage, little compute capability and in some cases, are restricted by limited battery life. I'm therefore searching for solutions to select suitable DL architectures (and/or modifying them) for various IoT applications as well as looking at distributing training/inference among a network of devices (often termed "fog", which is analogous to the distributed nature of the cloud, but on a smaller scale at the edge of the network).
I'm still working on a way to word this for people with very little technical experience at all.
Transitioning from university staff to student has been a fairly straightforward, however the sheer volume of reading for the first few months is a challenge I'm trying to get to grips with. Although I developed experience with machine learning over the last year or so for the learning analytics project I was working on, I'm currently entrenched in learning about various deep learning techniques. These focus on artificial neural network (ANN) data structures that were initially introduced to mimic the brain's interconnected neurones. They are computationally expensive to train but don't require much pre-processing of the input data unlike other traditional machine learning techniques. There are many, many variations of these ANNs, each of which has it's own benefits and drawbacks. In the context of my research, I need to understand how best to employ these networks on small, low-powered devices.
Additionally, the mathematics involved in deeply understanding these algorithms has led me to review my linear algebra, multi-variate calculus and probability skills (which are lacking, but I'm getting there). I'm finding this a pleasurable challenge though. Academic reading at PhD level is proving to be the most challenging aspect, however. I think it will get easier as I become more familiar with the background knowledge to approach some of the journal papers in this area.
I'd like to think by mid-September, I will be in a good position to really start fleshing out the beginnings of a literature review but we'll just have to wait and see...