Learning about AI in the workplace: AI4ALL alum at ReWork Deep Learning Toronto

Guest post by Maxwell Jones, Carnegie Mellon AI4ALL ‘18

AI4ALL Editor’s note: Meet Maxwell Jones, a 2018 CMU AI4ALL alum. Maxwell recently won an AI4ALL Alumni Grant to support his travel to the ReWork Deep Learning Summit in Toronto. Read on to hear about what Maxwell learned at the conference and what kind of AI applications he’s most excited about.

Thanks to an AI4ALL grant, I was able to go to RE WORK’s Deep Learning Summit in Toronto in October. Deep learning simply refers to large neural networks that use a training data set to learn and predict outcomes. During the conference, I watched engaging presentations, interacted with a variety of companies, and attended workshops where company representatives led discussions on deep learning and neural nets. The presentations were evenly split between companies with large AI capabilities or sectors and researchers explaining their findings. The discussions were held by companies who already had a working and profitable neural network implemented.

Most of the attendees were employees of companies who wanted to learn more about deep learning to implement or improve it in some aspect of their business. The diversity in types of companies was greater than I had expected.

My favorite companies using AI at the conference included a bank that was implementing a neural network to better predict when users were going to need a loan, a storage company expanding its social media platform by using AI to increase its appeal, and a computer vision company that could recognize human movements from drinking to talking to drawing letters in the air with a finger.

The conference gave me a better idea of how AI could be used in the workplace. While I thought that researchers and companies would be at relatively the same place in terms of implementing deep learning, I was very wrong. Leading researchers are working on the optimization of deep neural networks, but most companies are still getting their first or second version of deep neural networks to work at all.

One of the most stimulating parts of the conference was a presentation by Dr. Geoffrey Hinton, widely regarded as the godfather of neural networks. His goal was to create a small neural network that performed as well as a large network with the same data. To do this, Dr. Hinton needed to find a way to transfer all the information that the large network had learned to the smaller one. He trained the small network with the training data twice: once with the correct answers and once with buffered answers from the large network. Buffered answers allow the small network to not only find the correct answer, but also rank how likely other options were. Telling the small neural network that a picture is a cat is helpful, but telling the network that the larger network classifies the picture as 10 times more likely to be a dog than a car and 1000 times more likely to be a car than a house gives the small network much more information from a single data point.

During the conference, I learned more about cutting-edge research in deep learning and saw how companies of all kinds implement it. I saw how the research aspects and business application of deep learning work together, and I plan to continue with deep learning as I pursue my education.

About Maxwell

Maxwell Jones is a senior at Thomas Jefferson High School in Alexandria, Virginia. His academic interests include math and computer science, and he plans to double major in those two subjects in college. He is also a writer and editor for The Paper — a national origami magazine, a three-year varsity basketball athlete, and a tutor for kids in his high school and a local elementary school.