Research Spotlight: Dr. Muhammad Abdul-Mageed
University of British Columbia
Dr. Abdul-Mageed, a Computational Modelling theme Co-Lead for the EFL SSHRC PG, answered some questions for us regarding his work with natural language processing (NLP) and deep learning, his recent appointment as a Canada Research Chair (CRC) in Natural Language Processing and Machine Learning, and how his projects address some of the pressing issues at the forefront of society today.
Can you tell us about what first led you to be interested in natural language processing and machine learning?
For me, it is a story of language and computers. I realized early on how much I am fascinated with language, as for example in learning foreign languages, their structures, comparing between them, and growing my vocabulary. One manifestation of this is my interest in poetry, reading it and even experimenting with writing `poetry’ at times. One thing about poetry, especially of the type that employs meter and rhyme, etc. is that it follows rules. For me, this is intimately related to engineering. In other words, reading and writing poetry is a process that is similar say to dealing with objects in real world, understating and even simulating it. This brings up the question of computers. Like many others, I would lose myself sitting in front of a computer exploring almost everything my hands can get to, whether it is learning new software or even tinkering with computer hardware.
And then I realized how much I can do to actually use computers first to analyze language, then to model it and that is basically what NLP is about. It is about teaching machines to understand and generate human language. So, you teach the computer to read a piece of text and you want to identify what type of emotion that text is expressing (so that is understanding), or you want to translate it into a foreign language (which involves generation). Now in order to build really good models of language (or speech, image, video, etc.), you need good machine learning. In other words, you need to develop and apply methods that works well under different types of conditions: with big data, small data, noisy data, etc. Machine learning is the field to do that. And I fell in love with especially deep learning methods just because they are powerful and have all this human inspiration. It is about mimicking information processing in the brain. If we understand how the human brain works, perhaps we can instil some of that knowledge in machines. Deep learning does that.
Can you share a bit about Learner.ai, the new educational platform you recently launched?
So we are undergoing a real revolution where advances in AI, machine learning, and deep learning are becoming very pervasive. They have applications in almost all science and technology as well as the social sciences and humanities. A lot of people find these fields and methods useful and interesting and wish they knew more about them. The problem is that these fields can look quite scary to an outsider. After all, there is a lot of math involved, then you need engineering skills. So, as an educator, I love empowering my students with technical knowledge that is not quite accessible without much facilitation. I first took on making a lot of material more intuitive in courses I teach, such as my deep learning for NLP or machine translation courses, just by trying to boil these concepts down to the the first principles as well as by visualizing them and making them relevant to what students already know from before. Then I saw this huge demand outside the classroom, meaning people from different institutions or even professionals who have interest in machine learning. So, I figured one way to serve this audience would be through an online platform where we provide hands on content and offer live workshops. That materialized in Learnera.ai. We have already offered our first workshop based off Learnera and we were excited to see the massive interest, with more than 250 people attending both in-person and online. We plan to iteratively provide novel content and training workshop through Learnera.
How do you feel being named a Canada Research Chair (CRC) will impact your future research?
I currently have a very wide network of national and international collaborations, and I am very grateful to have a great team of students and post-docs in my own group here at UBC. I hope the CRC will allow me the time and capacity to strengthen these collaborations, scale up the training I offer, and perhaps start new collaborations. I am also hoping the CRC will allow me the bandwidth to collaborate with my colleagues in building and strengthening structures within UBC, such as the Language Sciences Global Research Institute and our AI Center, CAIDA.
What are you currently working on within the EFL team?
My group is involved in several projects. We are developing methods to enable development of better speech and language processing models that can be used in many educational applications. For example, we are developing large language models that improve Arabic, African, and Indigenous language processing. We are also developing machine translation models that allow best use of content across many of these languages and afford more engaging communication between people from different backgrounds. We are also developing more equitable speech processing models that can improve experiences of second language speakers of English, for example, to interact with machines and other humans.
In an ideal future, what kind of projects/work would you like to be doing?
My goal is to carry out meaningful research that can have positive impact on the lives of large populations. In particular, I really want to articulate projects that serve diverse populations, including those historically marginalized. We live through a great time of social awareness and we need to keep the momentum as to focus on social justice. UBC currently has important initiatives that directly target issues of equity, diversity, and inclusion and I do think that no matter what your field is, you can contribute. So, we are developing models that can hopefully be used by several groups and that can help make discoveries to ease human suffering and improve well-being.
I also work on developing methods that I hope will enable more efficient use of resources. Some of the most impactful methods developed within deep learning right now are still energy-hungry. With all our concerns about climate change, it is not sustainable to continue working with such models. So, one research thread in my group is about developing environment-friendly models that consume less energy.
Dr. Muhammad Abdul-Mageed is an Assistant Professor in the School of Information and Department of Linguistics and an Associate Member in Computer Science at the University of British Columbia. He is also a member of the Center for Artificial Intelligence Decision making and Action (CAIDA) and the Institute for Computing, Information, and Cognitive Systems and Director of the Deep Learning & Natural Language Processing (DLNLP) Group at UBC.
In June 2022, Dr. Abdul-Mageed was named a Canada Research Chair in Natural Language Processing and Machine Learning. He also delivered a talk on Deep Learning and Turjuman, a neural machine translation toolkit that he and his team developed, at the 2022 EFL Annual Meeting in Toronto.
Read more about Dr. Abdul-Mageed’s CRC appointment and his research below:
Turjuman
Preprint: https://arxiv.org/pdf/2206.03933.pdf
Demo: https://demos.dlnlp.ai/turjuman/
GitHub: https://github.com/UBC-NLP/turjuman
Documentation: https://turjuman.readthedocs.io/en/latest/
Twitter thread: https://twitter.com/mageed/status/1535306948370845696?s=20&t=PcXaO7YDpuEndFLLyDuQjg
Deep Learning
https://bit.ly/3m6Op2C
Presentation at 2022 EFL Annual Meeting: https://ensuringliteracy.ca/muhammad-abdul-mageed-presentation/
