Advanced driver-assistant systems and automated driving functions have to deal with increasing complexity as their application evolves into several areas such as multi-lane highway junctions or urban areas. One of the main challenges is the assessment of interactions between traffic participants. In order to master such complex environments, the usage of deep learning algorithms is promising. In this session I will present how we at ZF use such algorithms in order to anticipate the future behavior of traffic participants and thus, improve comfort and road safety and finally enable advanced automated driving functions.

We've all heard about AI, ML, Data Classification, etc. All cool!But what about building decision-making assistants AKA Recommendation Systems? Ever heard of them? No? Great! We'll cover the basics of Math, then proceed with a brief overview of Classification, and finally, discuss how we can use Matrix Factorization techniques to build state-of-the-art Rec. Systems! Prerequisite to this presentation is a familiarity with Data Classification Systems. Anything else is not really required as we will introduce it through a series of examples.

Recent advances in natural language processing demonstrate the capability of large-scale language models (such as GPT-3) to solve a variety of NLP problems with zero shots shifting from supervised fine-tuning to prompt engineering/tuning. However, building large language models raises challenges on data preparation, training, and deployment. In addition, while the process is well-established for a few dominant languages such as English, its execution on localized languages remains limited. We'll give an overview of the end-to-end process for building large-scale language models, discuss the challenges of scaling, and describe some existing solutions for efficient data preparation, distributed training, model optimization, and distributed deployment. We'll use examples on localized languages such as French or Spanish using NVIDIA Nemo Megatron, a framework for training large NLP models optimized for SuperPOD hardware infrastructure.

Machine learning models are becoming obsolete and must be retrained - this is the current widespread tenor. After we have taken a closer look at this thesis, we will show how this can be done. Which components does a CI/CD pipeline for Machine Learning really need - and which are optional. How can the whole thing be implemented without building an entire ML Platform team. And which challenges are still difficult to solve.An experience report including (mis-)decisions that will help to take the right path with your own challenges. In addition, it gives an overview of the essential components of a functioning machine learning platform that focuses on solving business challenges through high automation and low vertical integration.

The current standard for monitoring patients’ health conditions and predicting complications is discontinuous and suffers from inter-observer variations, which can lead to over- or under-treatment. Therefore, it is critical to address the shortcomings of the current standard and develop continuous and more standardized tools. In this talk, I will introduce AI-based technologies that can be used to monitor health and predict complications of vulnerable patient populations including infants and minority groups. I will also discuss how these technologies are designed to be lightweight to enable its use in portable and wearable devices. I will also show how the performance of these technologies was comparable to that of trained health professionals. Finally, I will conclude by presenting pressing challenges and several future directions for AI-enabled wearable technology in healthcare.