The KI Service Center @HPI is honored to announce the first KISZ talk by Matei Zaharia. Matei is an associate professor at Stanford CS, where he works on computer systems and machine learning as part of the Stanford DAWN Project. He also co-founded and is the Chief Technologist of Databricks, a data and AI platform startup.

Background: The effort to democratize large language models aims to expand access and usage of these complex neural networks to a wider audience. Dolly 2.0 is the first open-source, instruction-following LLM, fine-tuned on a human-generated instruction dataset licensed for research and commercial use. Databricks are open-sourcing Dolly 2.0 in its entirety, including the training code, the dataset, and the model weights; its licence is suitable for commercial use. Hence, any organisation can create, own, and customise powerful LLMs that can talk to people, without paying for API access or sharing data with third parties.

Ramin Khalili, Huawei Munich Research Center, presents a new method in the field of Federated Learning (FL) that allows successive freezing and training of the parameters of a FL model, thus reducing the resource requirements of the training on the devices while still allowing sufficient co-adaptation between parameters.

Max Sommerfeld, a Lead Data Scientist at Deutsche Bank, explains, why he sees generative AI (genAI) as a huge opportunity for the banking industry and the banking industry in turn as the ideal place to apply generative AI. To this end, he presents established and emerging use-cases of generative AI in banking. In addition, Max Sommerfeld discusses the technical challenges and how to solve them as well as regulatory challenges and the resulting question of how generative AI in banking can be made safe and trustworthy.

Dive into profitable AI with Elizabeth Press, founder of D3M Labs and data leader, strategist and cybersecurity expert with 20 years of experience in digitalization, data and AI. Despite the hype and activity, German companies are struggling to operationalize and deploy AI, let alone use it to drive profitable growth. This talk delves into the profitability of AI, focusing on the business case and productization – aspects that are often underestimated. The talk also addresses the challenges of implementing and executing AI, exploring key topics such as MLOps and data readiness, including quality and governance.

AI systems often struggle with data limitations, changes in data distribution over time, and poor user experience. Human-in-the-loop design offers a solution by placing users at the centre of AI systems and using human feedback for continuous improvement. In this talk, Adrian Loy of Merantix Momentum dives deep into a large-scale automated rodent behaviour analysis tool. The system allows users to identify and annotate relevant segments within a large corpus of video data.