You may be aware that machine learning is being used to recommend new music for you to listen to or to train automatic cars to recognize people and objects on the street. Other applications are less well known, though they will arguably make a bigger impact on society as machine learning paves the way for advances in science, medicine and healthcare.
A few weeks ago, Elon Musk and Mark Zuckerberg had a widely discussed Twitter feud about the potential dangers of a world where our everyday technologies are largely powered by artificial intelligence (AI). However, many in the AI community argued that such doomsday-thinking overlooks the benefits – and pitfalls – of AI in its current state.
At Elsevier, we’re increasingly using machine learning to help researchers, engineers and clinicians do their valuable work. It can help them find funding, bring them the right information when they need it and support the treatments they give patients. Below, five of our machine learning experts discuss the work they’re doing and how it can change the way science and medicine are done.
1. It can reveal trends that are important to your work
Deep Kayal is a Machine Learning Engineer with Elsevier’s Content and Innovation Team, which works on the automatic extraction of data from research papers for various products at Elsevier.
Natural language (processing) to me is the frontier of machine learning these days. Text is difficult to process, because it is not continuous and it is subjective: words mean different things to different people. But this is exactly what makes it fun. For example, we’ve built a functionality that distills the topic of a paper based on its contents so that we can now do large-scale analyses of which topics are cited most – and can therefore be considered to be trending. This helps researchers decide where they should focus their research on, it helps funding bodies decide where to invest their money in, and it helps publishers decide which journals should cover what next. The data that we have at Elsevier is a goldmine of information for a lot of interesting and intelligent things. The fact that we own this data and can therefore guarantee its quality really sets us apart.
2. It can improve the quality of science
I’m a biologist by training and was brought into Elsevier to bridge the work we do in life sciences and healthcare with machine learning. One of the models I’m working on will be able to identify whether a sentence in an abstract is a result, method, hypothesis or a goal. This is important to compare papers with each other and understand the validity of the science. Non-reproducibility is a big problem, which has enormous implications for drug companies, who can’t always trust new research coming out of academia and often have to do studies all over again. If machine learning could improve the quality of science, this would have huge implications on the price and speed of drug manufacturing.
3. It can tell you what research you should be reading
Finne Boonen is a Data Scientist in Elsevier’s Big Data group, where she currently works on Newsflo, a service that measures the impact of researchers and their institutions by tracking media coverage of research.
There is a lot of potential in what we can do at Elsevier with data and metadata of scientific literature. One example is to combine search with recommendations, so that researchers can get exactly the right paper they need to read next based on their research interests. The challenge there is to make sure that what we recommend contains enough new information to avoid that researchers are trapped in a filter bubble. This is a problem for everyone – just think about all the discussions about fake news during and after the American elections – but especially for scientists, the filter bubble is very insidious. You have to avoid that the papers by well-known researchers get an exponential amount of citations at the cost of lesser-known academics’ work. At the same time, there are certain seminal articles that scientists in certain fields need to be aware of. Such questions are really interesting to solve.
4. It helps you follow the money
Dr. Georgios Tsatsaronis is the Principal Natural Language Processing (NLP) Scientist with the Content and Innovation Team.
One of the most exciting projects that we’re working on right now will allow us to easily extract information on who funded a given study. This will make life a lot easier for funding bodies, who can then easily find the output of the research they’ve sponsored, but also for researchers, who often have to report back to funding agencies, costing them a lot of time. What I like about working on machine learning here at Elsevier is that we can combine theory with practice. We do research, have patents and even publish papers, and we then get to use this new knowledge to develop new functionalities for products that can actually help improve research.
5. It lets you search for images – and use pictures to diagnose medical conditions
Sujit Pal is Technology Research Director in the Labs group, where he focuses on search, natural language processing, machine learning and distributed processing.
One of my projects was doing image classification for ClinicalKey, allowing users to search for both images and text – a game changer for medical professionals. The challenge with automating healthcare technology is that the stakes are very high, since lives are on the line. Therefore, we have to reach accuracy levels of 90 percent and higher. At the same time, in healthcare you also get the biggest bang for your buck with machine learning. For example, computers are now as good or even slightly better as humans in predicting diabetic retinopathy (diabetes induced blindness) by looking at pictures of the retina. More and more, machines will fulfill this kind of service role. This does not mean that computers will replace doctors, though. It simply helps them save time on routine jobs, which they can then spend on the things they are best at, be it a niche specialization, diagnosing a rare illness, or simply engaging with their patients.
“Don’t treat it like magic – it’s just math”
Still not convinced about the potential of machine learning? “Don’t take our word for it, go try it yourself,” Helena advised. Coursera and fast.ai provide excellent MOOCS (massive open online courses) to help you get started. “The learning curve is significantly lower nowadays than it was even a year ago,” she said. “Don’t treat it like magic – it isn’t. It’s just math.”
comments powered by Disqus