Please tell us a little bit about your educational background.
I completed my undergraduate studies in mechanical engineering in 2015. From 2016 to 2019, I worked as a Software Engineer developing chatbots, websites, analytic dashboards, and a privacy-preserving conferencing app. In 2018 I co-founded AISaturdays Lagos, a community of enthusiasts who meet every Saturday for 16 weeks to learn about Artificial Intelligence (AI). As we dived deep into the concepts and mathematics of machine learning, I realized how little I knew. To expand my knowledge of AI, I decided to pursue a master’s degree in computer science. So, in 2019, I moved to Germany, and I have just recently completed my Master’s at Saarland University.
How did you come to CISPA?
During my first semester at the University of Saarland, I took the machine learning and cybersecurity course offered by Prof. Dr. Mario Fritz from CISPA. I remember walking into the CISPA building for the first time and telling myself that I would work here one day. At the end of the first lecture, I approached Mario to ask him if he would be interested in supervising my master’s thesis project. During my second semester, I did a remote summer internship at the University of Toronto and Vector Institute under the supervision of Prof. Dr. Nicolas Papernot, where I worked on a generative model extraction approach for speaker identification models. So I studied the vulnerability of speaker identification models to model extraction attacks, where an attacker with access to a large surrogate dataset can query the target model (victim) in the form of an API endpoint. The consequences of such an attack are devastating, because the attacker can successfully replicate the functionality of the victim model and can use this trained model to create adversarial examples for malicious purposes. You can imagine crafting an adversarial example that classifies one speaker as another speaker. The impact of such an attack would be insurmountable in real life, if it were successful. After my internship with Prof. Dr. Papernot, I joined Mario’s research group as a research assistant, where I started working on generating private synthetic cohorts of genetic data. Later I worked on my master’s thesis project, which focused on learning generative models for tabular data based on small samples.
You are a group leader of the “SautiProject”, a community of researchers. Can you tell us what the “SautiProject” is about and what you are working on?
SautiProject is an accented speech research group at TRI-AI, a non-profit organization I co-founded with my colleagues at AISaturdaysLagos. Our goal is to improve accented speech research in Africa. There are about 250 ethnic groups, and each group has its own language. Basically, there are three major languages (from the tribes Igbo, Yoruba, and Hausa), but among them, there are many other languages and accents.
The SautiProject is currently focusing on four projects: SautiDB, which is a crowdsourcing dataset collection platform. Anyone can go there, read some texts, and we capture their way of speaking. SautiClean is a post-processing tool for speech datasets. So, after we receive the audio, my team and I listen to hints of specifics in accents. The machine learning tool “SautiClassify” helps us with an accent classification. For example, I’m from the Yoruba tribe and anyone who is Nigerian probably recognizes that I’m Yoruba from my accent. By giving a speech example of my voice the tool should be able to classify that this is a Yoruba speech or not. The last part of the project is an accent translation tool to improve the online learning experience; it’s called SautiTranslate. It should be able to translate from one accent to another. We envisioned this as an educational tool.
In the beginning, you mentioned the project “AI Saturdays Lagos”, which you also co-founded. What is the project about, and what would a “typical” Saturday look like?
AI Saturday Lagos is a community of learners and researchers. We teach machine learning and AI-related subjects through Saturday Classes. Currently, I teach remotely. Typically, the classes take place over a 16-week period. So essentially, the goal is to democratize the knowledge of AI in Africa, for our students to distribute and develop tools that we can use and tools for all the people across the world to use.
A “typical” Saturday looks like this: Basically, before the pandemic, we would have something that looks like a room of 40 to 100 students with various knowledge backgrounds in Lagos. Since the pandemic, the classes have moved online. Many of our students are finishing their undergraduate studies, but we also have students in graduate programs, working students, or students in their transitioning periods, for example transitioning from one field to Machine Learning and Data Science. Our cohorts are usually 16 weeks long, and we go through open-source courses for example from Stanford University, Carnegie Mellon University and the University of Amsterdam. The classes are instructor-led, and it is a mix of practical and theoretical classes. For example, we would have a theoretical class in the morning from 10 – 12 pm and from 1 – 3 pm we have a hands-on class where students can practicalize the concepts learnt in the theoretical classes. Students are also given assignments and are required to work on a personal or group project. At the end of the cohort, we offer students who have met the requirements a certificate of completion.
With free-access communities like AISaturdays Lagos, low-threshold access to complex research areas is possible. After a few years of project work, do you see more women participating in the courses? What could be a reason why there are still fewer women than men in AI research?
I think, in our community, we definitely have a lot of women, but it’s not enough. I told my teammate that there were too many men. For example, in my SautiProject-Team, there is just me and Lebogang as female, which is 20 percent of our team. This is not great, and it’s not because there aren’t a lot of competent women, it’s just that sometimes it takes time and intention. There is still a lot to do. In our classes we have a fair number of women, from our last classes maybe 30%. It’s never going to be enough for me until there is gender balance. There is also a community in Africa called “She Code Africa”, founded by Ada Nduka Oyom. It's a non-profit organization focused on celebrating and empowering young girls and women in technology across Africa. They have thousands of women that are studying not only machine learning or AI, but also software development and cloud computing. It’s a huge community of women trying to bring women into STEM (Science, Technology, Engineering, Mathematics). Community driven research is one way to help us push towards the goal. I think we are recognizing that as a society, there should be more women in STEM, and there should be more Nobel laureates women, there should be more women making decisions. That’s my position. Balance in society is what we want, not exclude one gender or the other.
Where did your interest in AI research come from? Why did you choose AI as your research field?
I became curious about the question: “Do machines really think?”, after I inadvertently found myself in a class full of students talking about NASA curiosity robot in space during my second year in mechanical engineering. That was about 11 years ago. But this question stuck with me throughout my studies and thus began my growing interest in Artificial Intelligence. But the more I looked into it, the more specific my questions became and the more I tried to weigh the pros and cons. So now I am both excited and cautious about this. I think it's important not only to understand the benefits, but also to know the limitations and what to do about them. For example, one such benefit is the use of AI to determine if a baby is suffering from asphyxia, one of the most common causes of neonatal death. This AI is being used to improve the lives of newborns in Africa, and Ubenwa, an African company, is championing this cause. On the other hand, there are several language models that perform poorly on African languages because these low-resource languages are not present in the training corpus. Africans should be at the forefront of developing tools that benefit their society. To do that, we need to build our skills and knowledge about them. If we do not do that, we will be at a disadvantage. I am someone who cares deeply about education, technological advancement, and the development of my ecosystem. Nothing else strikes me as urgent as preparing Africans for the future of machine learning and artificial intelligence and using these tools to solve many of our local problems.
These limitations of current AI systems are also the reason for my frustration in the field and the fuel for my passion to work at the intersection of machine learning, privacy, and security. As you probably know, AI not only reflects the biases in our society very well, but exacerbates them. We need to develop tools that protect everyone from the harm spread by these tools, understand them better, develop formal proofs and robustness guarantees, and have better expectation settings for what AI can and cannot do. My work here at CISPA is central to addressing these problems.
Do you already have any plans for the time after your PhD?
I hope to work as a research scientist after my PhD, continuing to work on research projects that intersect privacy, security, and machine learning, and building socially aware AI tools that takes the context in which it is deployed into account. I want to better society, and I’m very passionate about data science for social goods. Last summer, for example, I was at Carnegie Mellon University for the Data Science for Social Good fellowship program where I worked with Vibrant Emotional Health to improve the call-routing system so that people who are in mental health crisis can get help in a timely manner. I hope to continue working on social projects and developing tools that benefit society.
Is there any advice you would give to young girls or anyone who is interested in AI but may be hesitant to take the next step?
I’m still learning myself, but a general piece of advice I have for anyone interested in AI is to spend time testing their own hypothesis. Don’t echo other people’s opinions just develop your own opinion about some existing issues. Challenges are everywhere, in every field, but as you grow, you learn. Be patient with yourself and trust your brilliance. A lot of things take time, and to quote my colleague Sahar: Progress is not linear. So, it’s okay if you don’t understand now; keep at it. It will all make sense with time. Be patient with yourself. You will fall, you will stumble; but try to pick yourself up and continue to move. Keep trying — I’m rooting for you.
For more details on the projects, check out these websites:
AISaturdays Lagos: https://github.com/AISaturdaysLagos/cohort7_structure