Munich Datageeks December (Christmas) Edition

We are thrilled to announce our next (Christmas Edition) Meetup on December 19th at Adesso.

Format:

• 2 talks (each ca. 40 min incl. discussion)
• Time for networking + food + drinks before, in between and after the presentations
• Talks are held in English
• We will be taking photos and film footage at the event. These will be used to share news about our meetups and to publicize upcoming events.

The lineup:

Talk 1: Dr. Ahmed Fahmy - Causal Inference

Abstract:
This talk is a causal inference primer for data scientists. I will clarify the difference between prediction vs. causality. I will explain the basic concepts of causal inference such as potential outcomes, randomized control trials (RCT), DAGs, and average treatment effect (ATE) with examples and Python code. I will discuss when deploying causal models in production, what metrics you should monitor. Will share some references for further reading.

Bio:
Ahmed Fahmy, I have a PhD in Information Technology and applied Math, Stochastic Process, and Mathematical Optimization from INRIA/France. I am a Sr. Data Scientist at Generali focusing on pricing using causal inference. I have over ten years of experience solving complex business problems using advanced analytics and deploying solutions.

Talk 2: Walter Zimmer - TUMTraff

Abstract:
Intelligent Transportation Systems (ITS) allow a drastic expansion of the visibility range and decrease occlusions for autonomous driving. The "A9 Digital Testbed for Autonomous Driving" is located near Garching and is 3 km long. Overhead signs and sensor masts on the A9 freeway, the B 471 highway, and at an intersection in Garching-Hochbrück are equipped with 78 cameras, LiDAR, and radar sensors that monitor traffic to create a digital twin. This digital twin can then be sent to any networked vehicle, thus increasing the range of the sensors in the vehicle. Apps running in the vehicle can receive this fused data to make lane recommendations and provide warnings of traffic jams and accidents. Detailed labeled sensor data is required to obtain accurate detections of traffic participants. Unfortunately, high-quality 3D labels of LiDAR point clouds and cameras from the roadside perspective of an intersection are still rare. Therefore, we provide the TUM Traffic Dataset, which consists of labeled time-synchronized LiDAR point clouds and anonymized camera images. Here, we recorded the sensor output from roadside cameras and LiDARs mounted on gantry bridges. Our dataset comprises 16k images and point clouds with more than 125k labeled 3D boxes. With ten classes, it has a high diversity of traffic participants and scenarios, including accidents, traffic violation events, overtaking events, and U-turn maneuvers. In experiments, we show that the fusion of multiple sensors outperforms perception models using a single sensor. Our dataset is a valuable contribution to the scientific community to perform complex 3D camera-LiDAR roadside perception tasks. Labeled datasets from the "A9 Digital Testbed for Autonomous Driving" are available to the general public. https://innovation-mobility.com/tumtraf-dataset.

Bio:
Walter Zimmer (M.Sc.) is currently a Ph.D. candidate and research assistant at the Chair of Robotics, Artificial Intelligence, and Real-time Systems of the Technical University of Munich (TUM), where he has also been working as a research assistant since March 2020. He received his M.Sc. degree in Computer Science from the Technical University of Munich (TUM) in 2018. During his studies, he stayed abroad at the Technical University of Delft (Netherlands) and the University of California, San Diego (UCSD), where he was involved in developing perception and autonomous driving algorithms. His current research interests are mainly 3D object detection and simulation for autonomous driving.