Jennifer Marsman is excited about machine learning. As principal software development engineer and evangelist at Microsoft, she has been involved in everything from language processing to AI - and with her team, she's even experimenting with her own lie detector.
Coming from a background in natural language processing and the Natural User Interface platform, she has been on the cutting edge of machine learning for the past 14 years. And now machine learning is bigger than ever.
"There's been a huge resurgence in machine learning, with the cost of data storage going down and the computational power has gone up."
She believes that those two factors contribute to the fact that machine learning is now able to make sense of all of the data - parsing, finding correlations and patterns to be able to make predictions with data.
With machine learning comes many possibilities, and Marsman says Microsoft has focused on democratising AI.
"We want to make it accessible to everyone. So even if you're smart developer and you're good at what you do and you don't have a background in data science or machine learning, you can still use machine learning models."
Microsoft's three types of machine learning
Marsman says Microsoft takes three approaches to machine learning: cognitive services, Azure machine learning and the CNTK, or the deep learning toolkit.
She says cognitive services, that use algorithms developed in Microsoft research, are pre-built in the cloud and can solve common problems like facial detection, emotion detection are freely available to everyone.
Those facial detection capabilities can map and match faces, giving information such as gender and age, right down to information such as whether a person is wearing glasses or not.
"We even had swim goggles in our dataset so if you need to know swim goggles, we have an API for that!"
She says that similar technology can be applied to text analytics; parsing logs such as enterprise support tickets and customer calls. Mining and extracting that data through sentiment analysis can reveal key topics can unveil key problems that organisations need to solve.
Azure machine learning provides more insight by allowing developers to build and train their own machine learning models.
"It may be a problem that's specific to your own business. For example, I have all this transaction data and I want to figure out four logs on this piece of equipment. Here's all the information from IoT sensors about my machine to be able to predict when we're going to have a machine or equipment failure."
She says that predictive maintenance can be useful before companies face failure and start bleeding money from unexpected breakdowns.
The final offering is the CNTK, or deep learning toolkit. Marsman says that it's open source, and can be used for a variety of problems.
"The kind of problems deep learning is really good for is computer vision, machine translation, speech recognition - a lot of these things were done using deep learning specifically."
"Computer vision is a really exciting problem to think about in terms of autonomous driving - Ford has made an announcement that they're going to support level 4 fully autonomous vehicles by 2021. Those cars have no steering wheels, no brakes - you get in and you just go," she says.
"When you think about it, you need computer vision to do that. Unless you want to do a massive infrastructure overhaul where the car needs to be able to talk to the traffic light. I don't think that's feasible because that's a huge infrastructure cost to enable that to work. But computer vision is something where the car can see and react like a human, things like that can be possible," she continues.
How one question spurred a lie detector experiment
Marsman says that the importance of data privacy should also be talked about much more, and in this case participants were made fully aware of what data would be collected and how it would be used.
Marsman is also developing a lie detector, which reads from 14 different spots on the scalp.
"When you tell the truth, that activates the recall centres in the brain. When you lie, that activates the creative centres in your brain. I started thinking, if I had a thing that reads from 14 different points, could I maybe distinguish between the two of those?"
This question led to experimentation trying to solve a fun problem, one that attracted the attention of the Azure Machine Learning team. The project is still in development, but is an example of trying new and radical things that push the boundaries.
She says that Microsoft's democratisation of AI is also really exciting, particularly it will encourage people to start using it more and more.
"I encourage everyone to think about 'how can machine learning help your business or your life?' Even in the consumer space, people are using bots to improve their lives. I love the idea of having a personal assistant that can help. We're poised at the edge of possibility right now and I'm really excited to see what the next 10 years will bring."
Women in tech
As a woman in technology, Marsman says Microsoft has always been supportive. When she left college, she said she visited Microsoft and found it inspiring to see the female presence.
"There were so many female faces there when I walked around, whereas some of the other places I interviewed where it was all older white men."
Marsman says technology as a career is full of onramps and offramps as things change and evolve, meaning that everyone is learning everything at the same time - a benefit for those starting and stopping their work in various areas.
And advice for women in tech? Marsman says that it's all about learning.
"If you want to get started, I think the best thing to do is just build something. Just try. Use search engines to see if someone is doing something similar, look for existing code. You don't have to reinvent the wheel. Just build and try and you learn from your mistakes."
"Taking a risk - saying yes to a challenge even if you don't know how to solve it right away, just try. You learn from your failures. I don't think we celebrate failure enough in the industry. Sometimes it's a success, sometimes it's not. You learn so much in the process that it's worth it. "