'I was, and still am, fascinated by all things related to perception and cognition. There is an enormous amount of data about the brain and the behaviour it produces. I considered going into computational cognitive neuroscience, which is concerned with building computational models that match this data in one way or another. While these models can provide insight about how the brain works at the detailed level, they typically don't produce truly intelligent behaviour. In AI research we take the opposite approach: we focus on getting the system to actually perform a difficult task – whether it is recognizing objects in images, understanding natural language, or any of the myriad manifestations of intelligence - while only occasionally glancing at biology. This is what appeals to me most. I chose for Amsterdam because it is the most technical programme I could find.'
'When you move your eyes even a tiny bit, the image projected onto your retina changes completely, in the sense that every photoreceptor measures a light intensity value that is different from what it measured before. Nevertheless, we can easily see that the scene has not changed. More generally, humans are able to perceive “sameness” at many levels of abstraction, despite large changes in superficial appearance.
In my research, I investigate this phenomenon from a mathematical perspective. Using the mathematics of symmetry, group theory, I came up with an algorithm that can learn to separate the invariant essence from superficial characteristics. The mathematics turns out to be quite similar to that seen in particle physics; the algorithm decomposes a signal into “elementary particles” each of which describes a single intrinsic property of the scene. Furthermore, the decomposition into invariant “what” and co-variant “where” information is somewhat reminiscent of the ventral / dorsal information streams in the primate cortex. We're now looking at applications of this theory to visual object recognition and video understanding.'
'I've made a few good friends during my studies (shared deadlines provide for a great bonding experience ;-). The department is very open; there is ample opportunity for doing research projects with faculty and phd students.'
'I recently finished an internship with Google DeepMind. Most time goes in being a Phd student at the UvA. in my spare time I work on the company I've co-founded, called Scyfer. My responsibilities include reading papers that I find interesting, thinking about ideas that fascinate me, and scribbling formulas on pieces of paper, before tossing them in the bin. Every once in a while I'll have a good idea, in which case I do some programming and write a paper about it. If the paper gets published, I get to travel the world to visit conferences and meet interesting people. I'm also looking forward for my opportunity to stay at OpenAI in San Fransisco where I'm going soon for a while'
'Learn mathematics, and learn to appreciate it. If you like math, it will be easy to learn. If you've learned some math, this master program will be a breeze, and you will be well prepared to do great work in AI. Q.E.D.'