Tackling Hard Problems Head-On: 5-Minute Interview with Hilary Mason, General Manager for Machine Learning at Cloudera

“Bad ideas are the gateway to the best ideas,” said Hilary Mason, General Manager for Machine Learning at Cloudera.

Problems that look impossible intrigue Mason, the founder of Fast Forward Labs (now Cloudera Fast Forward Labs). If clients only bring up obvious, good ideas, she worries that they might miss out on high-impact ideas that may be hugely valuable.

In this week’s five-minute interview (conducted at GraphConnect 2018 in NYC) we discuss how Mason uses applied machine learning, graph technology and specifically Neo4j to help customers accelerate and embrace machine learning and AI opportunities.

How do graphs fit into your problem-solving strategy?

Hilary Mason: In my keynote at GraphConnect, I spoke about the power of metaphor in thinking about the problems we need to solve.

That metaphor can then drive the architectural decisions we make. We work pretty broadly in machine learning across a variety of industries, use cases and techniques. And within those, we find that there are a few really powerful metaphors that tend to resonate, and graphs are one of the dominant metaphors. It just makes a lot of intuitive sense that the world is represented in nodes and edges and the characteristics of the relationships between those nodes.

In my talk, I tried to draw out that interplay between the way we think about building machine learning systems and the metaphors we choose, particularly around graphs to connect them all together.

Can you tell us about a recent project where Neo4j has been particularly effective?

Mason: We do a wide variety of work, and we really like hard problems, the ones where at the beginning of the project, you’re not sure if they’re possible or not. I always say that we try to take it from being a science problem to being an engineering problem.

There a couple of areas where we’ve found that the graph metaphor, and specifically Neo4j, has been useful in turning problems into an engineering problem. One of them was building a tool for a bank to get the right information to the right traders at exactly the right moment to support better trading decisions, particularly with respect to commodities but broadening out to other kinds of securities.

There’s a really interesting graph metaphor here: the relationship between a commodity or a particular company, the relationship between a company and a person and the relationship between any of these things and what’s actually in the news. There are also really interesting pieces of the machine learning analysis that sit on the news feed to understand what entities are represented and the relationship between those entities and how that can eventually impact what they want to trade.

That’s one project where Neo4j was part of the infrastructure that allowed us to go from an idea, where we didn’t know if the idea would actually be useful, to the point that we could execute at a high enough level of quality to support people who are extremely professional in their domain, all the way to an app that those people use in their day to day work.

How do you feel about exploring less obvious uses for graph technology?

Mason: I’m a huge fan of having bad ideas and I’ll tell you why. If you feel comfortable enough to have bad ideas as well as good ideas – as well as ideas that may be a little out there, that’s a good sign. If you are able to get some things that might be a little too risky or a little too dangerous to otherwise share and consider them.

Whenever I walk into a company and I ask them what data science and machine learning use cases are they looking at, if I see a list of obvious, good ideas, things we know we can solve, things that we can calculate exact ROI for, I get worried because I think they are definitely missing out on some of the more high-impact, potentially hugely valuable ideas. Bad ideas are the gateway to the best ideas.

What do you think the future of machine learning will look like?

Mason: People always ask me where the future of AI and machine learning are going. I have a couple of answers that might seem to be in conflict but are actually the same thing.

One is that the technology will continue to evolve and the capabilities – what is actually possible, what is actually cheap and accessible – will continue to change.

We are not at the end or even at a plateau of innovation here. New stuff is becoming possible all the time and that is going to continue. On the other hand, I think a lot of the hype, and the excitement that was generated, perhaps out of proportion to the technical capabilities, will go away.

I’m saying both that we will see huge technical innovation and we’ll see a lot less excitement about it. We’ll know we have success and maturity for machine learning and AI when nobody cares about machine learning and AI anymore, when the technology has faded into the background and we’re excited about what we can actually do with it.

Is there anything on the machine learning horizon that you’re looking forward to?

Mason: I’m particularly excited about where we’re going with the developer experience for data science machine learning applications.

Today, we have a very disjointed process of doing exploration, developing models, deploying them, monitoring them to make sure they haven’t drifted too far and retraining them. This is an area where, I think if you look a year or two ahead, we will have tooling to support this across a variety of different metaphors so that data scientists are able to, in one toolset, go from a bad idea to a good idea to a model and to a production system without relying on other people to support them in that work. That’s something I’m personally very excited about.


Want to share about your Neo4j project in a future 5-Minute Interview? Drop us a line at content@neo4j.com