${vImageAlt}

Profile: Jason Bernard

Learning About Plant Growth Using Artificial Intelligence

Jason Bernard, PhD Student in Computer Science (Project 3.5)

P2IRC Research: Inferring Models of Plant Growth Using Artificial Intelligence

Can you tell me a bit about yourself?
I’m a PhD student here at the University of Saskatchewan. I received my Masters in Information Systems at Athabasca University. I’m an older student, so I’m experienced from working in industry and can write some very complex computer programs because I learned a lot while working for different tech companies. 

Can you provide a brief summary of your research?
Behind every natural process, there is an algorithm that approximates how it works. For GIFS and P2IRC we focus on plant growth, but this can be applied to other areas and extends beyond plant growth to biological processes within the human body, geological processes, and more. One way to model these algorithms is with L-systems – these are basically a way of converting a series of symbols into a model or simulation. This canola plant video is an L-system.  However, how do you find the L-system that models the growth of a specific type of plant? Currently this is done by human experts and is complicated and time consuming. So, we are using artificial intelligence to find the algorithms that produce these models. 

How did you become involved in the P2IRC project?
My masters research focused on technologically-enhanced learning. While I think that is valuable and I’m proud of it, I wanted to work on research that has more of a global impact. When I was looking to do my PhD, I noticed that this offer was associated with combating world hunger which is meaningful to me, especially considering the growing issues with food security. Furthermore, I agree with the GIFS mission that food security is a precursor to political security.

What do you enjoy most about your research?
In August 2016 I attended the first P2IRC Symposium and began thinking about finding L-systems using Artificial Intelligence. I was told this is a really hard problem – maybe impossible. I like a good challenge, and this work is not easy. I also like the idea that my research could help feed somebody someday.

Why is your project important?
This project reduces the cost and effort required to grow better crops for P2IRC. Our piece of the project empowers other experts to further their work more easily. It allows plant physiologists to build simulations of plant growth based on various factors in advance. In the case where a prediction says it will be successful, then a trial is still necessary, as no simulation will ever replace the need to actually grow the crop. However, the major gain is in reducing the number of trials that growers and plant physiologists need to do by predicting when a trial will fail. This comes with huge savings in terms of money and effort. 

From the universal perspective, the project reveals scientific knowledge algorithmically and helps find models that aren’t currently mechanically understood.

What was one of the biggest challenges you’ve faced as a researcher and how did you overcome it?
I’ve studied this paper where the authors discuss that inferring L-systems as immensely complicated. At the beginning of this project, I would have agreed. The basic idea of my work is that if there is a model to describe something, it must exist somewhere in the space of all the models. So, we search that space, but it is almost infinite. Technology has lagged here because this research is very challenging – we must find a way to shrink that space so that it is searchable.

What is one of the most interesting findings that you’ve had to date?
When I first started, the best inference tool we had were simple L-systems with 2 symbols. Now, we’re up to 72 symbols and have improved the state of the art by hundreds of factors. We had two scientific discoveries or ‘firsts’ that I’m really proud of. This doesn’t just apply to plant scientists; our work makes other computer scientists’ work easier as well.

The first scientific discovery we’ve made are algorithms for inferring stochastic L-systems and homomorphic L-systems. The stochastic L-system essentially increases the number of ‘strings’ that could replace a symbol. The homomorphic L-system is when you can’t even see the symbols because they are masked. This second discovery is especially important for making the research practical. In reality, when a natural process is observed, the mechanisms (which are the symbols) cannot usually be directly seen.

How have you grown or developed as a professional because of your research?
I’ve improved on some of my skills, such as writing. I also have gotten better at stopping myself from chasing down rabbit holes. 

Research Gate: Jason Bernard
Google Scholar: Jason Bernard

Share this story