Paul Murphy is CEO of Clarify, platform that makes media content extraction and search easy for developers to integrate into their applications. Paul's career in the software operations industry has spanned twenty years and three continents, in companies focused on voice recognition, text-to-speech and emerging voice processing techniques. In a previous blog 'Why Should Your Business Invest in Deep Learning?' I argued that, sooner or later, you were going to invest in deep learning. This week, we’re going to look at how to do that.  Once again, I want to go back to the early 80s. That’s when spreadsheets were introduced and quickly changed how businesses operated. Spreadsheets allowed business owners to write programs that helped them run far more efficiently than they could before. Spreadsheets allowed them to organize data and perform rudimentary computation. If you could do it with a pen and paper, you could do it with a spreadsheet.  While custom applications were much more powerful, it turns out that spreadsheets - with some basic algebra - solved the vast majority of the problem. Spreadsheets were used for the simple analysis jobs where it did not make sense to hire professionals. Even today they fill that gap.  Unfortunately, deep learning is more complicated. It depends on far more advanced maths, and requires more data than we’re used to handling. But if it’s valuable, we’re all going to want to use it. In fact, not using it will turn out to be an existential threat. The question then is: “How?”

Should I Build or Buy?

If you can afford it, you can hire scientists and engineers to build deep learning systems tailored to your business. They will use your existing data to train systems that will then allow you to predict the future or - more prosaically - find patterns to help you make smarter decisions. The more data you process, the smarter your systems will get. This will make it very difficult - if not impossible - for late or smaller competitors to catch up.  If you can’t build it, you’ll need to buy. This will be either generic deep learning capacity or specific tools and services.  Generic deep learning systems, like H2O are similar to spreadsheets. Like spreadsheets, they allow you to enter any data and perform any computation on that data. Unlike spreadsheets, they can’t be manipulated by ordinary business people. Only data scientists can develop the specialized “formulas” needed to train the system.So are only big companies are going to be able to leverage this technology? Not at all!  Luckily things have changed a great deal since the 80s with respect to software sales and – more importantly – distribution. Today, we can build very specialized software and sell it to anyone with an Internet connection. In fact, instead of having to install and manage your own systems, API-based machine learning systems can be had for a fraction of the cost and effort. This allows even small companies to leverage the expertise of expensive scientists and engineers without having to recruit them outright. Without this, we wouldn’t have our current Cambrian explosion of deep learning applications!  Today, most online deep learning applications are relatively generic. Siri is a deep learning system trained on massive amounts of speech data. My company, Clarify uses deep learning to extract data from video files so that the content of video libraries can be searched. And Microsoft’s How Old uses deep learning to guess the age of people in photographs.  Soon enough though, services specific to your industry and challenges will emerge. At that point, I hope you’re paying attention!  If not, your competitor may quickly be predicting a future in which your company no longer exists.Paul Murphy will be speaking at the next Deep Learning Summit taking place in San Francisco on 28-29 January 2016. Early Bird tickets are available until 4 December. All previous Deep Learning Summits have sold out - book here now to avoid disappointment!
Would you like to be our next guest author? Find out how here!

This is a guest blog and may not represent the views of RE.WORK. As a result some opinions may even go against the views of RE.WORK but are posted in order to encourage debate and well-rounded knowledge sharing, and to allow alternate views to be presented to the RE.WORK community.