Teaching Machines to Learn


Not only was she pushing herself outside of her comfort zone exploring this new field, Dotter was also getting to ask questions of purpose and meaning, questions at the intersection of both computer science and biology.

How can we teach a computer to learn the way we do? While many people have wondered that very question for years, not many people in the world today have approached the question of automated segmentation using machine learning — an area at the forefront of technological studies.

Marissa Dotter, a 2017 PLNU physics and engineering graduate, is one of the few exploring this area.

Dotter has interned with the Space and Naval Warfare Systems Command (SPAWAR) for three years. During that time, she has worked with machine learning, a field of computer science that teaches computers to learn without being explicitly programmed. At SPAWAR, a San Diego based organization within the U.S. Navy, Dotter first began working with classification, a sector of machine learning used to teach computers to identify a category set or class that a new observation or data point belongs to. For instance, a researcher could present a computer with an array of animal images (dogs, cats, horses, and turtles) and ask the computer to identify which category to place a new image of a cat. The goal is to have the computer identify the new image on its own with as much accuracy as possible.

While it seems pretty simple for a human to learn, since most of us come to understand simple classifications at a young age, it is very difficult for a machine to be able to understand these differences on its own.

“With machine learning, the whole concept is, at least with imagery, you can input an image or a bunch of images that are mainly what you want and the computer can learn characteristics and structures about those images,” Dotter explained. “Research has been going on for a long time on how to classify an image. But now, we’re looking at what we can do with it, if we can transfer knowledge to new data sets, if we can do character recognition, and if we are able to coherently read words in an image.”

Dotter’s current work involves segmentation, teaching computers to segment out images and identify only parts and pixels of images, but not the whole. This is to provide the computer with a more condensed and detailed understanding of what an image is. For Dotter, it involves writing algorithms and coding frameworks, which she admits was difficult at first.

“For quite a few weeks, I was just sitting at the computer reading,” she said. “That’s all I did — reading, reading, reading, and trying to build this on my own because that’s the best way to learn. My mentor told me to just try it, so I did.”

The challenges of this cutting-edge research and the opportunity to use her gifts and talents to problem solve are ultimately what captivated Dotter. A turning moment was when she presented a publication she wrote with the help of her supervisor, titled “Visualizations of High Dimensional Image Features for Classification,” at the 2016 Applied Imagery and Pattern Recognition (AIPR) conference in Washington, D.C. There, she contributed as an expert among a group of individuals with Ph.D.s in the field.

Related Article: A look at the gender gap in many science, technology, engineering, and math (STEM) fields.

“I think that was when it hit me,” she said. “I could see myself in this field. I could do this.”

Not only was she pushing herself outside of her comfort zone exploring this new field, Dotter was also getting to ask questions of purpose and meaning, questions at the intersection of both computer science and biology.

“These machines are supposed to model our biological neurons, so there are questions around cognitive neuroscience, such as, do we understand how we learn?” said Dotter. “Then there are questions of the actual algorithms that were written. Do we really understand what’s happening to these images as we push them through a network? Then, again, there’s the questions of the end. Are we figuring out what we’re actually supposed to be figuring out?”

The goal of machine learning is to make it possible for machines or computers to learn from their experiences, to be cognitive of their surroundings, and to ultimately be able to make decisions. This can be used well in robotics, the military, healthcare, and many other fields and areas. With the vast, unimaginable amounts of data being created every day (billions upon billions of gigabytes of imagery from satellites and more), tasks like categorizing or identifying images would be very time-consuming for human beings, to say the least. Through algorithms, computers can be taught to process the data for us.

“There’s just so much data out there and there has to be a way to use it to our advantage,” Dotter said. Of course, the ethical implications of machine learning are not lost on her.

“One of these years, machines are actually going to closely resemble people and the way we think, and that’s always weird to think about,” she said. “Not only what I have done and what’s been done, but where this is taking us. We’re basically trying to recreate in a machine the powers and abilities God gives humans, and that’s really crazy to think about.”

Related Article: Are We Becoming Obsolete? Robot Workers and the Future of Business

Though she does not believe machines will ever be capable of having emotions, feelings, or consciousness — a soul — as humans were created to have, there is a strong sense of responsibility she feels with her work and what is to come of it. Because of this, Dotter continuously seeks to rely on God and to have faith in Him and His will.

“I’ve always relied on knowing there’s a plan for my life,” she said. “God has a plan for me. He has a purpose and He instilled in me a calling. And He’s going to create opportunities for me to live out my calling as long as I’m willing and have the courage to go after them and try them.”

During her time at PLNU, the relationships Dotter formed provided her with a foundation and support system that allowed her to grow and practice asking deep and meaningful questions. And as she steps into this new chapter of her journey, she seeks to keep faith and her identity in Christ at the forefront.

“I’ve been given the ability to understand certain parts of education like math and engineering,” she said. “We’re all given different abilities from God. My talents and my intelligence are gifts, and I have to use them to continuously find His purpose.”

Related Article: PLNU alum blends art and science in entrepreneurial pursuit.