I am currently working on an unhealthy amount of different things in parallel. These are the three most important ones:
I do data mining for a friend’s company called Prezi. prezi.com is an online presentation tool based on an infinite, zoomable 2D canvas. The coolest part of the work is mining the spatial structure of presentations, but the regular stuff (spotting trends, regularities and outliers, user clustering, etc.) is interesting, too.
I help a sociologist colleague with his mathematical model of aiding the poor. Actually, it is just one specific situation that we model: Let’s say you can observe a poor person’s current wealth. You model his wealth as determined by two hidden variables: his effort and his luck. You want to compensate him for his bad luck, but not for his lack of effort. The paradoxical situation we are interested in is when decrease in luck implies a decrease in compensation. (Because a decrease of the observed ‘wealth’ variable causes you to downgrade your estimation on the hidden ‘effort’ variable.)
The open source translation memory service I mentioned in my last post like this is already fully functional, but the user interface really needs more polishing. Unfortunately, we lost much of our enthusiasm when we recently realized that a German startup called Linguee does something very similar, very professionally. But we will finish our system anyway. I don’t like unfinished projects.
Our system was originally not intended to be a for-profit service. It was an open source university project, a work of love, and a chapter in my PhD. Then we realized that it has profit potential, and increased our efforts. Seeing Linguee, we had to seriously downgrade our expectations of startup fame. But we are still interested in the project for the sake of it. The previous version has several thousand users who would love to see an upgrade, and Linguee currently only works for a few language pairs.
I am currently working on an unhealthy amount of different things in parallel. These are the three most important ones:
I do data mining for a friend’s company called Prezi. prezi.com is an online presentation tool based on an infinite, zoomable 2D canvas. The coolest part of the work is mining the spatial structure of presentations, but the regular stuff (spotting trends, regularities and outliers, user clustering, etc.) is interesting, too.
I help a sociologist colleague with his mathematical model of aiding the poor. Actually, it is just one specific situation that we model: Let’s say you can observe a poor person’s current wealth. You model his wealth as determined by two hidden variables: his effort and his luck. You want to compensate him for his bad luck, but not for his lack of effort. The paradoxical situation we are interested in is when decrease in luck implies a decrease in compensation. (Because a decrease of the observed ‘wealth’ variable causes you to downgrade your estimation on the hidden ‘effort’ variable.)
The open source translation memory service I mentioned in my last post like this is already fully functional, but the user interface really needs more polishing. Unfortunately, we lost much of our enthusiasm when we recently realized that a German startup called Linguee does something very similar, very professionally. But we will finish our system anyway. I don’t like unfinished projects.
Be careful not to fall prey to Sunk Cost fallacy.
Our system was originally not intended to be a for-profit service. It was an open source university project, a work of love, and a chapter in my PhD. Then we realized that it has profit potential, and increased our efforts. Seeing Linguee, we had to seriously downgrade our expectations of startup fame. But we are still interested in the project for the sake of it. The previous version has several thousand users who would love to see an upgrade, and Linguee currently only works for a few language pairs.