CURRENT RESEARCH
My research has always been revolving around the intersection between philosophy and mathematics albeit in different ways across the years. Initially it was through set theory, which I viewed as metaphysics, more precisely a metaphysical playground which could develop strong mathematical techniques which could be used afterwards on more complex metaphysical problems. Later I turned to proof theory, which I found quite boring, but it was a stepping stone to AI. Of course, GOFAI was the first stop. When I learned about neural networks in 2013, I was captured. The idea of doing everything (Church-Turing!) on a single little neuron, duplicated and connected to build an enormous network was inspiring. To this day I am far less inspired by ideas such as transformers, which use specialized and hand crafted layers than by ideas such as dropout, embeddings and autoencoders which stay true to the original spirit of connectionism.
This is how we arrive at my main research direction today. For me, everything relevant is encodings, for the aim should be not just to build large systems but "smart", in lack of a better word. As all connectionists, I also believe that the main thing should be learning optimal weights for a simple neuron. As weights are multiplied by inputs, any change in weight can be offset by a change in input. This in turn means that weight can be left unchanged if the input is processed well enough, which in turn connects well with ideas of embodied cognition in philosophy, i.e. that the eye is not a simple "camera" but that it processes a certain amount of data by itself before sending it to the brain. Experiments on encoding transformations (which are de facto just better encodings, since an encoding of an encoding is just a new encoding), are interesting both philosophically and technically.
A second research direction I pursue is the history of cybernetics and artificial intelligence in Croatia, the USSR and the Eastern block, which my mind finds almost aesthetically pleasing, by recovering and reconstructing the lost ideas of Eastern block scientists. This, for me, always sparks up my imagination which conjures images of scientists of old working in impossible conditions under the looming threat of an apocalypse.
To this day, I am perhaps best known for my textbooks, most notably Introduction to Deep Learning from Springer and Logic and Philosophy high school textbooks from Element which I wrote together with some amazing coathors. The publishers were also great and I have nothing but words of praise for their immense help.