The 38th International Conference on Machine Learning (ICML) runs Sunday, July 18 through Saturday, July 24, 2021 as a virtual only event. This premier conference features cutting-edge research in machine learning and related fields including robotics, machine vision, speech recognition, computational biology, data science, and, more generally, in Artificial Intelligence (AI). ICML is a fast growing event gathering participants with a broad range of backgrounds from academia and industry. IARAI is proud to take part in ICML 2021 as a Platinum sponsor.
At ICML 2021, IARAI is presenting in several scientific sessions:
1. The EXPO session “Institute of Advanced Research in Artificial Intelligence – Unique Research Opportunities in AI Algorithms, Health, Traffic, and Weather“. This 1 hour session will be held on July 18 at 16:00 CET. The session will introduce IARAI’s unique research environment with industrial scale real-world data openly shared with scientific community. We will showcase our datasets including high-resolution privacy preserving urban traffic trajectories, high-resolution multi-channel weather satellite data, and data from over 10 million health records. We will follow by highlighting our recent research in algorithm development, linking attention in Transformers with associative memory, achieving world-class cross-domain few-shot learning performance with a novel light-weight approach, and more. The session will be closed by outlining the latest Scientific Programmes, opportunities for collaboration and engaging at the institute, and a short panel discussion.
2. The Earth observed by the Machines. In this fireside chat on July 21 at 16:00 CET, Dr Pedram Ghamisi (IARAI and Helmholtz-Zentrum Dresden-Rossendorf), Dr Naoto Yokoya (the University of Tokyo and RIKEN), Dr Ronny Hänsch (German Aerospace Center), and Dr Fabio Pacifici (Maxar Technologies) will give a short introduction to the field of remote sensing and its challenges. A rapid increase in the number and heterogeneity of remote sensing data sources has been paralleled by an increase in computing power and data availability. Recent developments in signal processing and machine learning have led to impressive progress in the analysis of remote sensing data. However, the state-of-the-art machine learning approaches still cannot be used to their full potential for high dimensional, heterogeneous, and complex remote sensing data, mainly due to lack of high-quality training datasets. The panelist will discuss how a tight synergy between remote sensing and AI can offer solutions to these challenges.
3. Modern Hopfield Networks. This session on July 20 at 18:00 CET focuses on recent groundbreaking research by IARAI’s founding director Dr Sepp Hochreiter and his team. The authors propose a new paradigm for deep learning by equipping each layer of a deep-learning architecture with modern Hopfield networks. Recently, AI community saw a renaissance of Hopfield networks with tremendously increased storage capacity and convergence in one update step, while ensuring global convergence to a local energy minimum. The surprising discovery is that the Transformer attention mechanism is equal to modern Hopfield networks. In layers of deep learning architectures, they allow the storage of and the access to raw input data, intermediate results, reference data, or learned prototypes. These Hopfield layers enable new ways of deep learning and provide pooling, memory, nearest-neighbor, set association, and attention mechanisms. Deep networks with Hopfield layers are applied to various domains, where they show significant improvements on different tasks and benchmarks.
4. IARAI seminar entitled “Which one is more important: more parameters or more computation?” will be held on July 22 at 17:00 CET. The seminar is presented by Dr Jason Weston, a research scientist at Facebook, New York. In this presentation, the author argues that the amount of computation is an important performance metric of a deep learning model, independent of its size measured by the number of parameters. The seminar will be held on Zoom and live streamed on YouTube.
5. Ask Me Anything (AMA) session will be held on July 20 at 15:00 CET. In this session, IARAI members Daniel Springer, Moritz Neun, David Kreil and Michael Kopp will give a brief introduction to the institute, including our research directions, activities and career and collaboration opportunities and will then happily answer any question from the audience. Feel free to ask us anything!
All the above sessions will be held on Zoom and live streamed on YouTube. In order to join these events, please register on the given links. The Expo session recording will be made available after the conference.
6. IARAI Virtual Booth. Our booth is an interactive virtual place during all days of the ICML conference where guests can interact with IARAI researchers, academics and recruiters. The booth is structured through the Topia tool – a more human video chat in a customized virtual world. Our customized Topia IARAI Booth World represents a space with 4 areas:
- IARAI@ICML , a showcase of the agenda of activities and sessions at ICML21.
- Talk to us, where participants will be able to interact with academics and recruiters, in order to understand more about IARAI and about collaboration opportunities.
- Engage with us, the current landscape of our competitions and workshops (T4C, W4C, CDCEO).
- Our work, an area which gives participants the opportunity to check out IARAI publications, seminars, workshops and much more.
IARAI is happy to connect virtually with participants in a more human way through Topia. They will be able to easily move in and out of conversations similarly to a face-to-face conversation with Topia’s spatial video world. Participants of the ICML 2021 Conference are welcome to visit IARAI virtual booth anytime during the conference to chat with the IARAI team and browse and apply directly for the IARAI job openings.