June 13, 2016
Take 3, Scene 8: The Present and Future of Machine Learning
On this international episode of Take 3, 3Pillar's Marius Banici and Sayantam Dey join us all the way from Romania and India, respectively, to discuss the present and future of machine learning.
Machine learning is a study of pattern recognition and computational learning in artificial intelligence. It has recently been used by major companies to help businesses analyze their data.
- Marius and Sayantam give us a brief overview of machine learning and discuss why it has been a topic in recent news
- We talk about the machine learning platforms offered by companies like Amazon, Google, and Microsoft, and how they are changing the way businesses use big data
- Marius and Sayantam bring more depth to the discussions they began with their earlier blog posts and dive into what's coming with the future of machine learning
About the Guests
Marius Banici is the Senior Director of 3Pillar Global's Advanced Technology Group. In this role, he is responsible for creating a culture of technical excellence and innovation throughout the company by leading 3Pillar's advanced technology teams in support of our Labs initiatives, engineering teams, and clients.
Sayantam Dey is the Director of 3Pillar's Advanced Technology Group. He has been with 3Pillar for a decade, delivering enterprise products and building frameworks for accelerated software development and testing in various technologies. His current areas of interest are data analytics, messaging systems, and cloud services.
Read the Transcription
Julia Slattery: So let's start with the basics, what is machine learning?
Marius Banici: Machine learning is a subfield of computer science that gives computers the ability to learn without being explicitly programmed. It uses algorithms that can learn from and make predictions on data instead of following strictly static instructions.
Sayantam Dey: Another way to think about it is we started off by telling the computer specifically what to do and how to process a set of instructions. Machine learning takes that one step further and says that okay this is the data that I have, deduce from it certain knowledge or a certain pattern and then if I give you a new input, predict what the output would be. So it goes from imperative style to a knowledge-based machine making style.
Julia Slattery: Why has it become such a hot topic of conversation recently?
Marius Banici: Well this is a Renaissance of the artificial intelligence. This is the result of the latest years’ massive advancements in some key areas. First, it's computing power, then cloud accessibility. Also, we see huge volumes of data generated by the penetration of Internet connectivity and powerful mobile devices. Lastly, algorithms and tools evolved and are available to the general public. We can say that it's the perfect storm for artificial intelligence, and machine learning is a key part of it.
Sayantam Dey: And the key aspect is the availability of the data that is there. So back when we didn't have almost all of this data and the computation part, the focus was on creating perfect modeling. So let's say your local weatherman, if they were trying to predict the weather, they will take the data and try to build the perfect mathematical model that would take your last week's weather patterns and predict the next week's weather patterns. Now, with the amount of data and with the amount of processing power that we have to deal with cloud and GP and other such infrastructure, the focus has gone on to simpler items that get better with a lot of data. So now your typical weatherman does not need to spend so much time on building the perfect model, but rather on making sure that model gets enough data to keep improving itself and its accuracy.
Julia Slattery: So companies like Amazon, Google, and Microsoft offer machine learning as a service for businesses to better understand their data. Can you describe the platforms that they offer and how this is changing the way businesses use big data?
Marius Banici: Yes, so these big players have seen the need for and potential in getting more meaning from data. And as part of their cloud offering, they created services – that can be easily incorporated by companies – for using data analytics and machine learning to improve their products. If a company is using one such major service provider, we will find a relatively painless way to incorporate it and then have a better understanding and better serve their clients with personalized interaction. It allows us to incorporate user context and aggregate many data sources. And it's accessible to incorporate and use those new technologies in their future services and products.
Sayantam Dey: Okay, let's roll back and figure out why these services are being offered in the first place. So like we saw in the last decade or so, there has been an explosion in the number of tools and solutions that are there for business intelligence. Business intelligence focuses on trying to make sense of the data that you or a company has collected over the course of a certain time period – two years, five years. It tries to answer the question “Where are we today?” in terms of the business. If we can define some key performance indicators – what are those key performance indicators, how do those key performance indicators work, are we above a certain threshold, are we below a certain threshold. There are lots of players, but there very much are solutions in this space. The same thing is now beginning to happen in the BI space with all these guys offering up their services and even you could add Tesla as a player in there.
So at this point, I think, as an analyst, what is most useful to me is if I need to run experiments on certain sets of data that might be really big – like even gigabytes or terabytes, in the case of Amazon – I can take the data that is already stored on their stream, which Amazon makes available to me as long as I'm using Amazon infrastructure. And then I can run multiple experiments on it to see which is the best of those that performs with a given set of data. So as an analyst, that speeds up my work considerably because I don't have to invest in getting the infrastructure and setting up and maintaining it. So that's the major advantage.
Julia Slattery: You mentioned Tesla there, could you expand on that?
Sayantam Dey: Yeah, so they are trying to open up a platform called Mobile Eye. It's not out there in the public domain yet, but they are also trying to provide services like Amazon and Microsoft.
Marius Banici: Yes, they are democratizing using artificial intelligence by making it available to everyone and open sourcing it as much as possible.
Julia Slattery: So how do you see machine learning impacting not only the way businesses perform, but also the way data is used and understood?
Marius Banici: I think machine learning is the key for mastering the volume and complexity of the data that is produced. It will push forward pervasive computing and will bring us new ways to understand our world and make discoveries, and also automate many things in our lives.
Sayantam Dey: Yeah, like I was mentioning in the previous question, we saw the rise of the BIE big companies trying to use BI. So companies that have data and understand the performance metrics would now like to make some bets on what future strategies they might take in the sales, or the marketing, or even in the everyday business operations. These guys are poised to sort of take on the machine learning aspect of it and say okay, fine, we know these are our KPIs, we know that this is what we do well, this is what we don't do well, based on this can we demonstrate and offer our operating parameters in the future. So for example, the classic case of the customer churn, as in “How long am I able to keep a customer on my commerce site or on my portal that I have?” These are questions that are very important because they define overall the sales strategy and market strategy – I mean how you contact your customers, which customers you contact. So like Marius said, it's going to touch every aspect of the business sooner than later.
Julia Slattery: What does the future of machine learning look like? You kind of touched on this, but could you expand on it a bit?
Marius Banici: The same way we see today that most companies are shifting to cloud computing and it became ubiquitous, I think in a few years, machine learning will be operated in most of the software that we build.
Sayantam Dey: I agree with that. I mean, it's going to take the shape of how BI has become ubiquitous in business. People will figure out where and how to use machine learning in their business. There are certain aspects of it that are very glamorous right now. For example, chatbots. There’s a lot of hype around chatbots and it's glamorous, Facebook is doing it, Microsoft is doing it, Google is doing it, but eventually people will figure out what is the best chatbot to use and where to use them. So yeah, it's going to get into that slope of enlightenment pretty soon.
Julia Slattery: You both have written about machine learning related topics for the 3Pillar website in the past. Can you touch on what those blog posts are about, and some of the tools or uses cases that you wrote about?
Sayantam Dey: My blog posts center around statistical analysis, which is a precursor to machine learning. We can say the machine learning is actually an extension of statistics. Statistics deal with mathematical models, which we call parametric models, and machine learning takes it to the next level or takes it to another branch of it and looks at non-parametric models where you don't make any assumptions about the data. So that's what I have been writing about.
I think one thing that we probably should touch upon is that none of this is based in math and in statistics. I don't know if our listeners are looking to partner with people to work on machine learning problems. I think it's important for them to, at this point, not to get swayed by tools; instead, they should be talking to people who have a background in statistics and who have a background with solving mathematical and related problems.
Marius Banici: Yeah, I think around the solutions that we build, we can point to typical use cases; one is understanding natural language. We can have very good results with the available technologies and libraries and everything to get plain English text and be able to understand it and come up with smart answers. The second thing is aggregation of different sources of data and then creating services that incorporate social media, web, IoT, and everything around us for a better experience. And that is again about machine learning. When we speak about the tools used, for example, by Amazon, a major advantage is how easy it is to incorporate it. You will be able, with a set of data, to use machine learning to learn from your data and be able to drive predictions based on the past experience. And that can happen quite easily in a matter of weeks, but as Sayantam said, with the right knowledge on theoretical models behind it, statistics, probability, mathematics.