Every couple years a form of technology becomes the Next Big Thing. We’ve been through the 3D printing, AR, and VR phases and in 2016, it seems like it’s machine learning’s turn.
However, unlike individual concepts, machine learning is an infrastructure that could last a generation. Major corporations are banking on machine learning for their futures. Here in Dublin, startups are already exploring new frontiers in the technology. For them, this could be a potential gold mine.
“[We are] trying to replicate the way we learn,” says Arturo Calvo of the ADAPT Research Centre, a collaboration of Trinity College Dublin, University College Dublin, Dublin Institute of Technology, and Dublin City University. “Learn how we make decisions.”
Machine learning essentially allows computers to make a decision, even if designers did not specifically program it for that choice. If you ever used an old customer service chatbot, you probably noticed it could only answer certain questions with certain answers. Machine learning will allow the bots to react and adapt. Artificial Intelligence becomes a whole lot more possible as a result.
Parsa Ghaffari, CEO and founder of Dublin-based AYLIEN, says three factors led to machine learning’s growth. Each factor plays to a strength in Dublin’s tech community: interested startups looking for ways to use data, established research institutions, and a strong foreign direct investment culture searching for great talent.
- Big Data
Think of big data like books. After the invention of the printing press, the proliferation of books helped raise literacy rates across Europe and the world. Big data from social media, smart devices, and the internet of things makes it easier for computers needing to learn.
With all that data there is simply too much for humans to sift through, creating demand for machine learning. Demand creates opportunity. That is where several Irish startups have stepped in.
— AYLIEN (@_aylien) November 9, 2016
Ghaffari’s AYLIEN uses machine learning and artificial intelligence to scan text and photos to seek context. Unlike a computer, a human could read an article or document and understand its meaning. However, with the right programming a computer could scan hundreds of thousands of documents.
Ghaffari believes Ireland could develop a machine learning industry faster than other European countries because of its small size and well-trained workforce. With data becoming a major national resource, the question becomes how to exploit it.
Luckily, Dublin has a significant number of startups already looking for ways to harness that data. Logograb developed technology that can identify a brand just based on a picture of the logo. Iconic Translation Machines, which spun out of Dublin City University (DCU), uses automation to lower the costs of language translation. Other companies in the field include KatanMT, a SaaS-based machine translation company; Wattics, which developed self-learning software for energy management; and Belfast-based Amplihae, which uses machine learning to automate wide-area networks.
Opening.io, the so-called “Robot Recruiting Firm” helps companies narrow down candidates during a hiring process. Opening.io Co-founder and CTO Adrian Mihai says machine learning algorithms read resumes and match a candidate to a job. They can also use real-time data to determine the proper salary for the candidate.
“The field is raising a lot of interest,” says Mihai of Dublin’s potential in machine learning.
- Better Algorithms
Machine learning is as much about understanding the human mind as it is about understanding computers. Our brains process information using past lessons and experience to make decisions. Ghaffari points to recent breakthroughs in algorithms that have made it possible to really understand languages, images, and sounds.
Doctors Rozenn Dahyot and Naomi Harte of Trinity College Dublin hope to unlock other senses. Harte’s research focuses on understanding speech, while Dahyot focuses on visual signs. Both fields require multiple senses to give the context needed for decisions. For example, understanding speech requires visual cues as well as hearing.
The ADAPT Centre provides workspace for PhD and postdoctoral students to develop and research machine learning technologies. Declan McKibben leads ADAPT’s Design and Innovation Lab, which helps connect academic work with private industry. He acknowledges that commercialising products means working on a faster timeline than academia typical does.
— The ADAPT Centre (@AdaptCentre) November 9, 2016
“The simple question no one asks is: ‘is it any good?’” says McKibben. “Can we use this? Can it make us money? Can it squeeze a cost out of an existing process?”
Back in October, ADAPT held a showcase highlighting more than two dozen students’ research projects into machine learning. Ideas included PEEP (Personalized Event Engagement Portal), which lets people attending conferences take part in multiple sessions at the same time. Another student partnered with RTÉ to conduct sentiment analysis during the Irish elections.
Aniruddha Ghosh, a post-grad student at DCU developed the Sarcasm Magnet, which looks for sarcasm on Twitter, a task that is not easy for computers. He sees a potential business in his work. The neuronetwork system operates without a lot of hardware and it can do high-level analysis without much oversight. In fact, by adapting his technology Ghosh believes he could help analyze the mental health of a person based on their tweets.
— SarcasmMagnet (@onlinesarcasm) November 11, 2016
For machine learning to reach its full commercial potential more blue skies research is needed. Harte and Dahyot point out that behind all of the machine learning technology is mathematics developed hundreds of years before anyone knew an application for it.
“There has to be room for knowledge for knowledge’s sake,” says Harte.
- Better Infrastructure
You might not expect it, but the same equipment that makes World of Warcraft and other computer games look so cool is also fueling the machine learning boom. Rather than surfing the web using a CPU (Central Processing Units) like most of us do, machine learning entrepreneurs and researchers use GPUs (Graphics Processing Units). The advanced graphics cards can bring in audio and video, analyze it, and provide a response on the same chip. That speeds up the process.
Dublin has a strong pool of data scientists, according to Arturo Calvo. That encouraged a number of major corporations to set up their data teams here. For example, Linkedin is growing its natural language processing team in Dublin. Microsoft, IBM, and Accenture are all expanding their data science teams in Dublin.
Last year, the online fashion store Zalando launched its Fashion Insights Centre in Dublin’s Grand Canal Quay, to tackle ambitious data projects. “Dublin was the hot spot for data scientists and data engineers” says Simon O’Regan, Product Owner of Fashion Data at Zalando, lured more by the projects rather than the fashion.
According to him, investment from multinationals like AmDocs, where he used to work, and government policies encouraged people with advanced degrees to move here. Zalando promised to hire 200 people within three years. So far, according to O’Regan, they’ve hired close to 80 employees in Dublin.
Calvo and his colleague, Dr. Killian Levacher, launched the Dublin machine learning meetup group, which shows how big that talent pool is. More than 1,300 people joined the group, half from industry and half from the universities.
“This is the place to be at this moment,” says Levacher.
Calvo and Levacher made a specific effort to focus on the ethics of machine learning. They acknowledge automation will lead to job losses and other challenges. That may be the best reason why Dublin can become the machine learning hub of Europe. This technology takes more than just cutting edge research and strong programming skills. It takes the humanity of a country that’s seen turbulent times to know how to plot the best path forward.