Artificial Intelligence

Dr. Lino Coria: Machine Learning in Real Life

Dr. Lino Coria shares his real-world machine learning experience with the Society for Canadian Women in Science & Technology (SCWIST), including insights into communicating results with key stakeholders, what happens when your solution isn't used as intended, and why, sometimes, deep learning is not the answer.


By Dr. Lino Coria, Computer Vision Engineer at Ecoation


My name is Lino Coria, and I am a Computer Vision Engineer at Ecoation. Alongside my colleagues, I develop computer vision algorithms to count and classify fruit (i.e. peppers, tomatoes, etc.). I also use this information to track anomalies and identify patterns of interest and concern in the greenhouses we collaborate with. My academic background includes a Master’s degree from McMaster University and a Ph.D. from the University of British Columbia. Throughout my career I’ve gained experience working in both industry and academia, allowing me to understand the challenges of machine learning in production.


Two months ago, the Society for Canadian Women in Science & Technology (SCWIST) invited me to speak at one of their virtual brown bag events. I chose “Machine Learning in Real Life” because I thought that sharing my experience could be useful to young engineers working in artificial intelligence (AI).

The last decade has seen extraordinary progress in the field of artificial intelligence. This is mainly due to the advancement in deep learning research and the availability of tools and platforms to train sophisticated neural networks. Those interested in machine learning no longer need to be experts in order to create useful applications related to speech recognition, natural language processing, and computer vision. Training a neural network, however, is just the first step when it comes to building a complete AI solution that will be used by thousands or even millions of people around the world. In this talk, I cover some aspects of machine learning that you usually don’t hear about in online tutorials.

How to successfully collaborate with everyone at your company?

One word: communication. When it comes to your peers, remember that you are all working towards the same goals even though you might have different strategies to tackle the problems. Make sure you identify how to effectively communicate with everyone you work with and acknowledge everyone’s contributions along the way. 

There will also be frequent occasions when you need to show your results to your manager and leadership. Even though you have run lots of experiments and produced dozens of plots and tables, you should not include everything in your presentation. Identify the key takeaway that you want people to get out of your work and pick one useful plot that drives the message home. For example, is the new method five times faster and equally as effective as the previous one? Build your presentation around that premise. This way everyone, including time-constrained leadership and less technically-focused team members, will appreciate the significance of your contribution.

What happens when people don’t use your product as expected?

In my experience, people will always use your product in ways you never imagined. You need to make sure you communicate with the end user to understand their needs and make it easy for them to use your product the right way. Ultimately, good, intuitive UX design is key to guide the user. Taking additional steps to develop standard operating procedure (SOP) guidelines will also be very useful in ensuring the product is used in the intended way.

Why, sometimes, deep learning is not the answer?

Deep learning is a very powerful tool but it is not the only one we have. Deep learning models require a lot of data for training and deep neural networks are computationally intensive. Before you choose to use deep learning, consider if there are other “old school” machine learning approaches that might be used to solve your problem. There is a chance that this is the kind of problem that does not need machine learning, and thinking ahead in this way has the potential to save you many hours of intense work. Explore all the options before you start.

Watch the full lecture here:

Dr. Lino Coria

Computer Vision Engineer

Ecoation Innovative Solutions Inc.

Similar posts