How do machine learning models function with little data? | Humboldt Professor for AI Suvrit Sra

In maths lessons we learn to calculate the peaks and troughs of linear functions and, in doing so, are dealing with a very simple optimisation problem. It gets more difficult when several parameters have to be drawn up in order to find the best solutions to a specific problem, such as determining ideal prices for maximising profit in business or the best location for a logistics centre. In machine learning (ML), far more complex optimisation processes are key. To enable autonomous vehicles to differentiate between humans and road signs or computers to compose music that sounds like Chopin or Beethoven, an AI model has to be created and “trained” with data. By constantly optimising the target functions and algorithms, the model “learns” until it is finally able to process huge volumes of data, acquire patterns and laws and, ideally, make correct statements about unknown data. In order to adapt the model to new data, multiple optimisation problems have to be solved.
Based on his fundamental methodological work on different optimisation problems, Suvrit Sra has ushered in major progress in machine learning in the last few years. He is a driving force in collaboration between mathematicians and machine learning specialists, which benefits both fields.
As a professor for Resource Aware Machine Learning at the Technical University of Munich, Suvrit Sra’s methodological expertise is set to strengthen fundamental research into machine learning at the university, which already holds a vanguard position in artificial intelligence nationalwide. The research focus of Sra’s Humboldt Professorship for Artificial Intelligence will be placed on the robustness, reliability and resource efficiency of ML. Moreover, cooperation is planned between Suvrit Sra and the computer scientist, Stefanie Jegelka, who has also been selected to hold a 2022 Humboldt Professorship at the Technical University of Munich. In addition, the university hopes to sustainably shape ML research in Germany on the strength of this collaboration.
#HumboldtProfessor #ArtificialIntelligence #MachineLearning