Monday, 18 June 2012

How to think outside the box - literally

This blog post is about innovation and creativity. More precisely it's about a stupid little technique, that once helped me be creative and achieve something unexpected: come up with a quantum physics technique, and publish a paper, with barely any training in physics.

Most big ideas and breakthroughs are unexpected. They are not part of your plan for next week, in case of a company, probably not on your roadmap for next year. It is important to have a roadmap, but it is just as crucial to realise that your potentially biggest achievements are not yet on it. Keeping that in mind, what can you do right now to trigger that next big idea coming? This may be your moment: follow these instructions. It may take some time, but bear with me. I mean, really, try this, if you can't commit yourself to try something creative right now, when you have nothing better to do than read this crappy blog post, then when will you?

My thinking box
Step 1: To be creative, you want to go beyond your confort zone: you want to think outside the box. What's the first thing you need for that? A box. That's right. If you want to think outside the box, you have to have the box first. So go and get one now. A cardboard box, large or small, will easily do. I used the one pictured on the right, which I found in the recycling bin in the office. Resume reading when you found your box.

Step 2: Now that you have the box, what's inside? Inside the box are things that you are confortable with thinking about. If you have a to do list, a project plan, a roadmap for next year(s), those all go inside the box. Everything that you anticipate to happen to you and your work, everything predictable, those all go inside the box.

Step 3+: When you're ready and have the box, you can start thinking outside of it. You will likely need a pencil, to take notes, I simply sketch stuff on the wall of the box. You may prefer a quiet place to think: this time, try a busy place: remember, you're outside your comfort zone now. The rest is up to you. Let me know if you how you saved the world using this technique.

Wednesday, 6 June 2012

My big fat data-driven wedding: how to decide when to get married

There's all this hype about the importance of using data to support critical business decisions. But few things can be more important in someone's life than choosing the partner they plan to spend the rest of their lives with. So look no further for a data-driven approach to support this critical decision: when should you stop looking and finally get married?The formal setup of the problem is as follows: there are \( N \) potential marriage partners out there, and I want to choose the one and only partner of my life. My goal of course is to settle with no-one but the very best partner to marry.I can 'date' each candidate sequentially in a random order. When I date someone, I can evaluate how she ranks in comparison to all previous partners I dated. Based on this, I can either go on and marry her in which case the process stops, or reject her and carry on dating more partners. Once a partner was rejected she cannot be called back again. The problem is hard, because it's assumed that I know nothing about a partner before I actually dated her. Some might immediately recognise the typical problem of exploration versus expoitation hidded in here. So when should I stop dating and marry someone?Let's say I want to maximise the probability of ending up with the best partner as wife. Firstly, there's no point in selecting someone who's not the best so far, because that also implies she cannot be the best overall. Also, the earlier you marry, the higher the chance of missing out on someone without even dating her. It turns out the optimal strategy is as follows: date the first \( k \) candidates and reject them irrespective of their 'performance'. Then, keep on dating partners, and choose the first one, who ranks above everyone I dated before. The threshold \( k \) depends on \( N \), assymptotically it's about a third of all candidates ( \( k \approx \frac{1}{e}N \) see slides for details).I gave a talk about this theory in our lab at Cambridge not long after getting married, and did a quick back-of-the-envelope calculation to see how close my marriage decision was to the optimal. The theory predicts based on my criteria I should've dated roughly 196,132 partners before getting married. Well, I leave it to you to guess whether I did that or not :) But at the end of the day I ended up making the best decision, so it doesn't matter.I got recently reminded of this problem, as I was thinking about making instant decisions in the PeerPerks platform, which as it turns out  can be modelled as a more complicated version of the optimal marriage problem, called probabilistic submodular multiple-choice secretary problem. In PeerPerks, we allow brands to offer freebies to influential users of social media. Our job is to make sure the promotion generates maximal success based on various indicators. In PeerPerks, every once in a while a user turns up on the website and authenticates using their twitter or facebook. At this point we have access to their social media profile, so it's like we've dated that person: we have a little insight into which topics she's interested in, what kind of people she's connected to, etc. Based on this instant evaluation our algorithm decides whether or not the person qualifies to receive a perk or not.