Incremental Learning from Data Streams

Since the data streams are characterized as open-ended and often infinite sequences of data examples, of which data distribution is likely to change through time, this indicates the requirement for alternative learning algorithms to typical (batch) learning algorithms that learn from stationary datasets. The goal of the proposed course is to teach the students about the state-of-the-art algorithms that are used to perform learning from data streams. The course will guide the students through the major open challenges in the field (supervised learning, data compression, concept drift detection, clustering from streams, specialized evaluation statistics). With this knowledge, the students will be able to apply their machine learning skills to a specialized and useful area that is connected to the abundance of data in our everyday lives (bank/weather/financial transactions, sensor readings etc.).
The course will be organized by mixing lectures with hands-on lab exercises that the students will do in the Statistical package R. The lab exercises will include applying the acquired knowledge on their own problem and stimulating a competition between students to achieve the best possible learning results.