One of the main fuels for the computing technology revolution is the ability of engineers to create more advanced algorithms for pattern analysis and real world understanding. They want to give you the best search results, to translate text for you, to give you the best movie or book recommendations, etc. For decades, pattern analysis has been done by engineering algorithms specific for a certain use case; and even with increasing use of machine learning, engineers still design custom features and predictors that are used by machine learning algorithms to make their task easier. In order to apply machine learning, engineer spends a lot of time finding out what are those important features from which an algorithm will learn. To recommend you a movie, engineer might conclude that the movie genre you watch the most might be an important feature. The “real” learning revolution hasn’t started when we first applied machine learning with features designed by humans, but when we started to be able to apply it to solve problems end-to-end. Algorithms learn both what are the features that are good to solve the problem and how to actually solve the problem. To recommend you a movie, give the algorithm all raw data I have on you, and let it find what’s important.
Miloš is mostly involved in applying state of the art research in practical engineering problems. He has the most of the working experience in the areas of self-driving cars and ADAS, while recently he is mostly working on analyzing vehicle mobility and trajectory patterns. He is an engineer who is mostly interested and skillful in general pattern analysis, statistical estimation and machine learning - with highlights on deep learning, representation learning, sequence analysis and computer vision.