Specializing in engineering and architecture of large knowledge-intensive AI systems including a question answering ensemble for e-learning and research at RWTH Aachen that was able to defeat PhDs on factoid open domain question answering tasks as well as the IBM Sapphire dialog system for academic advising at the University of Michigan. Now working on IBM's Fabric for Deep Learning (FfDL) which is the open source heart of Watson Machine Learning (WML) for framework-agnostic distributed deep learning on top of Kubernetes that integrates with a wide ecosystem of related open source projects like Uber Horovod, the Adversarial Robustness Toolkit (ART), AI Fairness 360 (AIF360), H20.ai, Seldon and kube-batch.
My current expertise and research focus forms three pillars:
- Exploring more complex real-time AI workloads beyond pure deep learning training like deep reinforcement learning
- Integration of various AI solutions (FfDL, ART, AIF360, compression, XAI, visualization) and approaches (RL, DL, causal inference, prob. programming, etc.) in coherent AI fabrics
- Architecture of complex AI applications (i.e. orchestration & choreography of AI applications)
I currently work at the MIT-IBM Watson AI Lab in Cambridge, MA, USA.