Interns: fresh approaches to fascinating challenges

Read time: 2 mins

There's always a buzz about the place when the interns visit.

Work-in-progress today, a thesis next year, then a future changing the world

Work-in-progress today, a thesis next year, then a future changing the world

While we're very much focused on our commercial proposition, we also have a strong research programme. This is because of our association with University College London: we're an AI spin-out with support from its Technology Fund, and one of our co-founders is Prof Jun Wang, UCL's  Chair Professor of Computer Science.

A big part of this commitment involves sponsoring interns during their studies - and a big part of their studies involves the thesis that they will eventually present.

So, they recently came to MediaGamma HQ in bright, bustling Clerkenwell to present their work-in-progress, under the watchful eye of Jun and Dr. Shuai Yuan, our VP of Data Science and himself a UCL alumnus.

The problems they're trying to crack are interesting because they're challenging. They also hold huge commercial promise.

Here's a summary:

  • Image factors for real-time bidding. One intern is looking at how image factors on a web page or mobile app might influence user behaviour. Every pixel on a screen has an RGB value - that is, how much red, green and blue go into making that colour. The research is finding out whether these values can have any effect on how users behave. Does what they see affect how they respond to the ad and if so, what exactly affects behaviour, and how? The bidding algorithm should take these factors into account when determining bid prices. 
  • Personalisation of e-commerce. On the home page of most e-commerce sites you'll see 'blocks' of content, promoting different products, holidays, offers and so on. If a site has, say, 50 blocks, then which combination of five blocks at the top of the page creates the highest likelihood that someone will click something? This is an example of the wonderfully-named 'multi armed bandit' problem.
  • User segmentation. If many thousands of people visit your site, and you want to promote the right products to them, then you need some way of segmenting them. But what's the best way to do this? You could start simple with just geography, or demographics, or go to the other extreme and create incredibly detailed, bespoke segments. But then, how can you be sure the segments you create today will still be relevant tomorrow? Next week? Next year? Or, given real-time bidding, how about in the new few minutes? Our interns are looking at ways in which algorithms not only identify audience segments completely automatically based on people's digital behaviour, they will even create names for those segments.
  • General architecture for reinforcement learning. How do you address the challenge of massive scale with personalisation? Moreover, how do you do this in an adaptive, flexible, intelligent way, in real-time? It's a huge engineering challenge, and one that our interns have taken up. This is exactly the kind of problem we relish: we like the hard problems because they're the problems worth solving.

It's so refreshing to see our innovative interns take on these really tough challenges. If you share our passion for world-class problem-solving, and you'd like to know more about working with us, then take a look at our Careers page or get in touch.