72nd Vienna Deep Learning Meetup: TiRex + Responsible Annotation

Hi Deep Learners,
It took us a while, but our next Deep Learning Meetup is taking place on April 28 at Frequentis. Our topics this time are: Zero-Shot Forecasting with TiRex and Responsible Annotation Services.

***
Agenda:

• 18:15 Arrival
• 18:30 Introduction by the meetup organizers
• Welcome by the host: Frequentis
• 18:45 Talk 1: TiRex: Zero-Shot Forecasting Across Long and Short Horizons with Enhanced In-Context Learning by Sebastian Böck (NXAI GmbH)
• 19:30 Announcements
• Networking Break
• 20:00 Talk 2: Combining technological precision with social responsibility in training data preparation by Andreas Schachl (Responsible Annotation Services)
• 20:45 Networking
• ~21:30 Wrap up & End

***

Talk Details:

Talk 1: TiRex: Zero-Shot Forecasting Across Long and Short Horizons with Enhanced In-Context Learning
TiRex is a compact pre-trained forecasting model for zero-shot time series forecasting across both short and long horizons. Built on an xLSTM backbone, it combines recurrent state tracking and strong in-context learning with contiguous patch masking and masked inference, enabling robust forecasting and efficient uncertainty estimation. This talk outlines the key architectural and training choices behind the model, including data augmentation strategies and large-scale pre-training. Results on public benchmarks show strong zero-shot forecasting performance together with fast inference and a low memory footprint. The talk also highlights how TiRex representations transfer to downstream tasks such as time series classification, illustrating the promise of forecasting models as broader foundation models for time series.

About the speaker:
Sebastian Böck is Lead Researcher at NXAI, specializing in Deep Learning, Sequence Modeling, and Time Series Forecasting. He received his PhD in Computer Science from JKU in Linz and his diploma degree in Electrical Engineering and Information Technology from the Technical University in Munich. Long before his current focus on the efficiency and scalability of foundation models, he pioneered the application of recurrent neural networks (RNNs) in Music Information Retrieval—spanning onset detection, beat tracking, and polyphonic transcription. This early work cemented his deep expertise in state-tracking and recurrent architectures. Today, his research at NXAI centers on overcoming the fundamental scaling and inference bottlenecks of modern AI through the extended Long Short-Term Memory (xLSTM) paradigm and its application across diverse modalities.

Talk 2: Combining technological precision with social responsibility in training data preparation
This presentation introduces Responsible Annotation Services (RAS), a specialized All-Austrian provider of high-quality data processing. Founded as a spin-off of the non-profit Responsible Annotation, RAS leverages a unique workforce of neurodivergent experts to deliver exceptional attention to detail, pattern recognition, and perseverance - helping to create reliable, ethical, and effective AI systems.

Three distinct case studies will be presented demonstrating the versatility across modalities and industries:
Bank Foyer Safety: A strategic collaboration with KEBA to refine their AI assistant, KeBob. This involved the meticulous labeling of human postures and mobility aid usage to enhance real-time emergency detection in semi-public banking environments.
Text Annotation: A dive into our natural language processing project with Newsadoo, showcasing how our team structures complex textual data to train advanced language models with high semantic accuracy.
Air-Vehicle Surveillance: A large-scale project with Frequentis involving the annotation of approximately 5 million video frames for airport monitoring, transforming raw footage into precise training data.

About the Speaker:
Andreas Schachl is a seasoned tech leader and open source enthusiast who believes that creating a better future requires action — not just ideas. Together with his friend Hans‑Peter Kranewitter, Andreas co‑founded Responsible Annotation Services, an AI‑focused venture that streamlines the entire data pipeline. Their key to success is a neuro‑diverse team of annotation experts who consistently deliver top‑tier results at high speed, even over prolonged periods.

We are looking forward to welcoming you at our April Meetup.
Your VDLM organizer team.