Over the weekend, I had the opportunity to visit the strange land of NIPS (Neural Information Processing Systems), a kingdom in the far corners of the West, where the Machine Learners dwell. Some of the noblemen there had invited me so that the locals could get a close-up look at one of us Visualization People at one of their workshops.
The Machine Learners are a friendly and peaceful people, if a little peculiar. They do not burden themselves with purposes or tasks. They are content with their work if it functions in itself, and only seem to worry about comparing to their own. It is a serene and seemingly simple world, though one that might feel a little constrained and self-centered to a well-traveled outsider.
The people of NIPS have strange and interesting practices. Among them is the Ski Break, which they observe from 10:30am to 3:30pm for the purpose of skiing on the NIPS Mountain (which is always covered in snow). While the Machine Learners gathered and dutifully observed the break, there was a remarkable lack of interest in the actual practice of skiing. On the slope, I saw mostly visitors from foreign lands, who easily outnumbered the locals. The Machine Learners, instead, gathered around their machines, of which they are evidently very fond, to discuss their intricate inner workings.
Foreign, to them, are the worries of the visualization world. When they need a metric, they can easily find one. Everything in their world can be measured, observed, and expressed entirely in their language. They do not have to contend with the complexities and incompleteness of information that plagues most of the rest of the world.
A strange language is spoken in NIPS, with many expressions whose meanings I was unable to discern. The Machine Learners often exchange glances and nod while they speak, as if to confirm that they do, in fact, understand each other’s language. They also fill their notebooks with strange symbols and equations that I have not seen used anywhere else, except for the alchemists.
NIPS is a rich and beautiful place, their king must be a proud man. When I arrived with the first rays of daylight, they were feasting on an extensive breakfast, brought to them by busy servants who were ready to fulfill their every wish. I immediately received the customary welcome gift of a champagne glass (though, alas, sans champagne), and their king had also generously arranged (through his vassals) to cover the expenses of my travels.
While he does not seem to encourage his own subjects to travel, their king does welcome outsiders into his empire. The Machine Learners I met showed a general curiosity towards visitors, if not a lot of knowledge of the happenings beyond their borders. To thank them for their hospitality, I shall speak to our Council of Elders to invite a delegation from NIPS to one of our next gatherings.
As I was bidding my farewells after night had already fallen, the king’s servants were preparing a lavish feast. Riding back to my humble inn, I could hear the Machine Learners’ songs for many miles.
Great post — dialogue with the Machine Learning (and statistics) community is vital if visualisation is going to realise its long term potential for exploratory data analysis, as well as communication. The problem is, how do you convince a member of the machine learning community that human visualisation can uncover information that algorithmic methods can’t? The best answer I know is to show them Anscombe’s Quartets (http://en.wikipedia.org/wiki/Anscombe's_quartet).
Although these examples were invented to show the limits of linear regression, the lesson is bigger than that: whatever statistical test (or machine learner) someone can come up with, we can come up with examples like these — ie that have the same statistical properties but are clearly different when visualised. (I’m pretty sure that this is a corollary of the No-Free Lunch thereom http://en.wikipedia.org/wiki/No_free_lunch_theorem, but haven’t managed to prove it yet)
joe