September 28, 2018
PhD in Computer Science
Chief Researcher, Systems and Machine Learning Group
NEC Laboratories Europe
Research for Statistical Relational and Neural Network Learning.
Our current research topic is on developing machine learning models for several complex systems, including digital health, retail, financial and also social systems that can be modeled as graph structured data.
My expertise is specifically in rational machine learning, and trying to combine relational learning with neural network based deep learning approaches that are very good at high dimensional and noisy data. The goal is to develop machine learning models with the data that is found in many application domains that can be modeled as a set of entities, and set of relationships between those entities, and creating new knowledge and new insights from the data. We think that the technology that we are developing will have a high impact on these systems, improving the lives of many, and making societies safer. For example, in many application domains it is important to discover latent connections between entities.
In the digital health domain, for instance, there might be latent connections between patients because of their genetic markup or the type of imaging data that's available for them. So our technology aims to discover these latent connections and improve the discovery of new drugs, diseases in patients, and aims to personalize medicine.
Challenge to Combine Different Types of Data.
One of the major challenges of our work is combining high dimensional and noisy data such as images, text and speech with the data that you typically find in company databases: relational data. For pattern matching, deep learning works really well and we have made very big strides in this direction. But then in situations where you don't have this type of data or where you also want to perform reasoning, incorporate human background knowledge, and understand the outcome of your algorithms, it is very crucial to combine existing pattern matching approaches based on deep learning with reasoning and relational learning approaches that are still being developed in the AI community.
Importance of Collaboration with Domain Experts
Machine learning researchers realize that in order to deliver very effective models and to deliver essentially a product that is meaning, it is necessary to collaborate with domain experts. For instance, in the digital health domain, it is extremely important to collaborate with the medical professionals to actually understand the data and, and to come up with new ideas. It is very important for machine learning researchers to be very good in a particular domain, and also understand the language of the people that are working in that particular domain.
It is very difficult to say what our research topic will be in 10 years from now for it is very rapidly evolving and changing field, but there are certain topics that will increase importance in the future, like explainable AI and explainable machine learning.
Explainable AI means that it's not only good at predicting something, but also be able to explain why the prediction was made. Since one of my main research goals is to combine relational AI and neural network based model free AI, I'm very excited to continue working on these topic in the future.
Another example is personalized medicine. Presently, medication is prescribed to all of the people in exactly the same way, regardless of their genetic markup, etc.
We want to help doctors to make better decisions with our technology in the future, that are more personalized and tailored to a particular individual, like looking at the genetic markup or medical imaging data of that particular individual.
We must have a lot of domain expert researcher who understand the particular domains, such as personalized medicine, and be able to interact with the people who are experts there, to have the breakthrough.
Dr. Mathias Niepert
Dr. Mathias Niepert received a PhD degree from Indiana University, USA.
He was a research associate at the University of Washington under the supervision of Pedro Domingos. Since 2015 he has been a senior research scientist at NEC Laboratories Europe. Mathias has published over 40 papers in leading conferences, journals, and workshops including ICML, NIPS, AAAI, IJCAI, and UAI. Mathias is an expert in machine learning for structured data and has won several best paper awards, a Google faculty research award, and several national research awards in the US and Germany.