Past performance is a very reliable guide to your career success

There are some things it's definitely better not to know. Like a prediction of one's h-index.

 We fed a neural network with data about the publication activity of physicists and tried to make a “fake” prediction, for which we used data from the years 1996 up to 2008 to predict the next ten years. To train the network, we took a random sample of authors and asked the network to predict these authors’ publication data. In each cycle the network learned how good or bad its prediction was and then tried to further improve it. 

Concretely, we trained the network to predict the h-index, a measure for the number of citations a researcher has accumulated. We didn’t use this number because we think it’s particularly important, but simply because other groups have previously studied it with neural networks in disciplines other than physics. Looking at the h-index, therefore, allowed us to compare our results with those of the other groups. 

However, that our prediction is better than the earlier ones is only partly due to our network’s performance. Turns out, our data are also intrinsically easier to predict, even with simple measures. You can for example just try to linearly extrapolate the h-index, and while that prediction isn’t as good as that of the network, it is still better than the prediction from the other disciplines.

Physicists are very predictable

I have a new paper on the arXiv, which came out of a collaboration with Tobias Mistele and Tom Price. We fed a neural network with data about the publication activity of physicists and tried to make a "fake" prediction, for which we used data from the years 1996 up to 2008 to predict the next ten years.