While deep learning models became more and more sophisticated, it simultaneously became apparent that there are other bottlenecks slowing down progress on important applications. On the one end, sifting through petabytes of data per day and choosing the right subsets for annotation requires complex software and tools. And on the other end, verifying the quality of the annotations becomes difficult to scale.
The new bottleneck has become choosing both data distribution and data quality that new models are being trained with. The most common questions revolve around three main topics:
How can I verify the correctness of my individual annotations and/or the whole dataset?
Is my dataset diverse enough, sampling the data space to touch all decision boundaries with the right weighting?
Is my taxonomy defined such that learning becomes easier for my model?
Watch our virtual panel discussion and learn about the challenges that ML faces in image-based dataset quality. Industry leaders in data annotation from from Bosch, Hyundai MOBIS, Zeiss Meditec, and the MIT-IBM Watson Lab, discuss and share ideas about the current state of the art.
The conversation offers insights and explores how new software applications could help verify dataset quality and allow for annotation scalability.
PhD on Unsupervised Learning of Convolution Neural Networks
(Honda Research and TU Darmstadt)
Prof. Dr. Hilde Kuehne is Head of Computer Vision and Machine Learning at the Computational Vision & Artificial Intelligence Group at the Goethe University Frankfurt and affiliated professor at the MIT-IBM Watson AI Lab. Her research focuses on weakly and unsupervised recognition and understanding of video data. She obtained her doctoral degree in engineering from the Karlsruhe Institute of Technology (KIT) in 2014.
Her experience includes projects with various European and US universities and international technology companies with a focus on image and video understanding processing. She has published various high-impact works, including the HMDB action classification dataset, which got recently awarded the ICCV Helmholtz Prize, known as the Test of Time Award, recognizing papers from ten or more years earlier that had a significant impact on computer vision research. She has organized various workshops in the field, is running a renown website for benchmarks in the field (actionrecogntion.net), and served as area chair for CVPR, ICCV and WACV in 2021.
Paul is a Project Director at Robert Bosch GmbH with 15 years of experience in the field of video-based driver assistance systems and systems for automated driving. He has studied Computer Engineering and has a PhD in Computer Vision from Mannheim University. He specializes in Artificial Intelligence and Data Driven Development and has driven the change towards Deep Learning based technologies for Bosch.
Currently, he is in charge of the development of video-based perception algorithms in a big OEM customer project that spans from Driver Assistance up to Level 3 Automated Driving. Paul believes that the biggest influencing factor for optimizing algorithms in the field of automated driving is the provision of high-quality data at scale.
As Head of Artificial Intelligence at Carl Zeiss Meditec Georgy is currently responsible for supporting the Artificial Intelligence Enablement. Before joining Zeiss Georgy was working at Royal Philips as clinical researcher, product and project manager and as global R&D leader being responsible for designing and developing Philips AI Platform and Philips Marketplace. Georgy holds a Master Degree in Applied Mathematics and Computer Science and a PhD in Biomedical Engineering.
Daniel received his habilitation at Heidelberg University in the field of machine learning and data science. His startup, Pallas Ludens, enabled automotive and medical imaging companies to collect large machine learning training datasets. After about three years, in 2016, Daniel and his team joined Apple, where he worked for three years on dataset design, annotation and data quality for computer vision tasks.
In 2019, Daniel co-founded Quality Match, which he is leading as managing director. He has also co-founded a social networking start-up, and works as business angel with a number of investments in gaming and machine learning companies.