Daniel Kohlsdorf
Data Scientist, Machine Learning Enthusiast and Programmer


Currently my interest is in the implementation of intelligent systems and to apply machine learning to behavior modeling, recommender systems and content search. I am particular interested in collaborative retrieval methods and word embeddings as well as probabilistic methods for recommender systems and document modeling.

My research interest during my PhD was focused on analysing sequential data recorded from biological signals such as dolphins or human gestures. In particular I was working on data driven methods for indexing, annotation and discovery of high dimensional sensor streams. My interest was to find and model patterns in dolphin communication and human activity data. Often behavioral data such as animal communication or human activity in everyday life show language like structures. My algorithms are designed to find the atomic units of such languages automatically from data as well as the temporal structure of their occurrence. In the end I believe that the analysis of continuous sensor data in terms of interpretable, language like structures can help to explain unexplored domains such as dolphin communication as well as help to build recognizers for complex domains such as activity recognition in every day life situations or false positive free gesture recognition. The systems I implemented for these purposes used convolutional unsupervised feature learning, hidden Markov models as well as grammar induction using alignment based learning.

In general, I aim to create intelligent agents that discover how data is organized naturally and communicate their findings in human understandable form. Therefore, my research interest lie in artificial intelligence, data mining, pattern recognition and human computer interaction.