Over the course of March of the Machines, there has been a lot of talk about machine learning and deep learning, and the jobs arising from them, but what is it like to work in that field?
When we talk about emerging technologies and the future of tech, deep learning is an area that crops up again and again. It will be the driving force behind the development of AI and robotics, and already plays an essential part in the creation of tech we use on a daily basis. But what is it like to work in this evolving sector?
We asked Kevin McGuinness, research fellow at the Insight Centre for Data Analytics, Dublin City University (DCU), about what he’s doing with deep learning and how the area is changing.
What is your role within the Insight Centre?
I am a research fellow working on topics in computer vision and deep learning. I also teach graduate-level data analysis and machine learning for the School of Electronic Engineering. My research is funded by Science Foundation Ireland under the Starting Investigator Research Grant (SIRG).
My team is interested in developing novel methods for applying complex models in situations where there is limited training data available. We are looking at a variety of approaches, including unsupervised and semi-supervised learning, multi-objective deep learning, transfer learning and domain adaptation. Our goal is to apply these techniques to solve real-world problems in areas such as information retrieval, image understanding, segmentation and medical image analysis.
If there is such a thing, can you describe a typical day in the job?
I work on many different projects, so a day at the office for me could involve a variety of activities, including discussing research experiments and theory with my PhD students, creating slides and lecture notes for teaching, presenting papers on our own research internally and externally, working on project and grant proposals, attending project meetings, collaborating with Irish and international researchers, and reviewing papers for academic journals and conferences.
I also ensure to include as much time as I can to work on my own research ideas, which typically involves writing scientific software, training machine-learning models or evaluating performance on benchmark data sets.
What types of project do you work on?
Some examples of projects that I am currently involved in include:
- multimedia information retrieval and image similarity search
- modelling human visual attention (saliency)
- deep computer vision models for computer-aided diagnosis of diseases, using radiograph and X-ray images
- automatic segmentation of neonatal brain images
- large-scale video annotation tools for autonomous driving
- automatic image analysis and tagging
- deep models for interactive image segmentation
- automatic brand and logo detection in images
- hybrid recommender systems based on content and collaborative filtering
What skills do you use on a daily basis?
Strong mathematical and software development skills are essential for my role. I use Python and C++ for development and training models. For deep learning, I use Theano and Keras, although I have also worked with several other packages, such as Caffe and TensorFlow. I use tools from linear algebra, calculus and statistics for understanding and implementing algorithms.
Good communication and presentation skills are also critical for communicating research and teaching.
What is the hardest part of your working day?
The diversity of activities each day can make some days a little frantic. There is a lot of mental context-switching necessary to jump between writing code, composing lectures, reviewing papers and trying to stay on top of the latest research. Many of these tasks require a very different mental frame and it can take time to shift gears.
Do you have any productivity tips that help you through the working day?
It is important to decide which tasks are the most important for you (and not just the most urgent), and set aside time for these in your calendar early. If you do not schedule time for these activities, meetings will consume your time and you will end up trying to fit in these activities between meetings, which results in a lot of context-switching overhead.
Also, keep a to-do list. There are usually far too many tasks to remember, and trying to keep them all in your head is stressful.
When you first started this job, what were you most surprised to learn was important in the role?
The importance of communicating and publicising your research was something I had not really considered much before starting my academic career. It turns out this stuff is important.
Time management is also more important than I would have thought.
How has this role changed as this sector has grown and evolved?
The recent advances in machine learning and deep learning have been widely covered by the media. The public now seems to have a better appreciation of what makes some problems in AI challenging.
Deep learning has also had a strong impact and unifying impact on many research communities. There is now a lot of cross-pollination of ideas from computer vision, natural language understanding and speech recognition.
What do you enjoy most about the job?
I very much enjoy working on challenging problems and coming up with novel solutions. The process of doing this can be frustrating at times, but that makes it all the more gratifying when you finally develop a solution that improves the state-of-the-art.