The image created by Midjourney in patient collaboration with ChatGPT4.o: “A portrait-format collage in sea colours, on the left side with a human figure from science stylised into a geometric prism. A single white ray of light enters the prism and splits into seven rays that spread out to the right. Each ray ends in shapes such as sparks, overlapping circles, waves, balanced scales, branching trees and faceted crystals. The background is bright with overlays of scientific and social motifs from the fields of mobility, medicine, production and education.’

Research at L3S

Goals, guidelines, examples

Excellent research is the foundation of scientific innovation and progress that has a lasting impact on our society. What does research at L3S aim to achieve? What are our objectives and methods? What are examples of L3S research excellence?

One of our early mentors, Gio Wiederhold, Professor of Databases at Stanford University until 2001 and an important member of the Scientific Advisory Board in the early years of the L3S, gave us some questions in the early years of the L3S that George Heilmeyer, former director of the US research agency ARPA, had formulated about 50 years ago as evaluation criteria for project proposals, and which were subsequently used in many other contexts and became known as “Heilmeyer’s Catechism”.

We reproduce them as formulated by Gio Wiederhold. They are suitable as evaluation criteria for research programs and project proposals, as well as (somewhat modified) as guidelines for doctoral theses, scientific papers, and scientific publications.

Heilmeyer’s Catechism
  1. What is the problem, why is it hard?
  2. How is it solved today?
  3. What is the new technical idea; why can we succeed now?
  4. What is the impact if successful?
  5. How will the program be organized?
  6. How will intermediate results be generated?
  7. How will you measure progress?
  8. What will it cost?

A simple but very effective set of questions that should be answered in every research proposal. Let’s take a closer look at some of these aspects and our research philosophy below.

Originality and Relevance

Excellent research starts with innovative questions that have not yet been answered and are also highly relevant to science and society. We aim to link economic and societal challenges with technological innovations that solve problems in new ways, better than before.

Methodology, transparency and procedures

Methodology and research results must be verifiable. Especially in AI, where models are often complex and difficult to understand, transparency is particularly important; not only the result, but also the traceable path to its derivation is part of excellent research. At a time when all available data is being used to train large AI models, we face the additional challenge of being able to evaluate them in a scientifically sound manner. The L3S Research Center attaches great importance to the open availability of data, algorithms and results in order to ensure replicability and verifiability by the scientific community. To this end, planning with concrete milestones and measurable progress in all phases of the project is helpful and important before work begins.

Interdisciplinary collaboration

AI as a new fundamental discipline (remember the recent Nobel Prizes in Physics and Chemistry awarded to basic researchers in AI) often requires collaboration with scientists from other disciplines: Computer science, ethics and law, for example, in the development of algorithms and new approaches; computer science and important application areas, such as mechanical engineering, medicine or education, in the development of efficient and tailored solutions.

Impact on the scientific, economic and social environment

Excellent research is not only important in science, but also has a clear impact on the economy and society. AI has the potential to transform a wide range of fields, from healthcare to mobility. A model developed at L3S to analyze health data could help make better treatment decisions and ultimately improve healthcare (page 21). The LaMMOn study (page 24) shows how innovative technologies such as language models and neural graph networks can pave the way for more efficient transportation systems. L3S research on digital epidemic control (page 20) shows that social networks are not only a place for communication, but also a valuable source of data for crisis management. L3S scientists are also working to use large language models to soften the sometimes overly harsh tone of social networks, allowing for more productive discussions (page 18). Some machine learning methods, such as deep reinforcement learning, hold great promise for embodied AI. But before it can be fully deployed in critical applications such as industrial systems or autonomous vehicles, issues of efficiency, generalizability, and robustness need to be addressed. The L3S is also working on these issues (page 16).

Ethical responsibility

Ethical responsibility is a key factor in the field of AI. AI-based systems are increasingly influencing important decisions in our daily lives. However, there is still much research to be done before generative AI systems can be widely used in critical areas such as medical practice. A recent L3S study showed that while GPT-4 has impressive capabilities in answering medical questions, it still sometimes makes incorrect recommendations and answers with great conviction (page 22).

The L3S Research Center is committed to responsible research and emphasizes the importance of fair and transparent algorithms. We also address the ethical challenges of AI and develop approaches to minimize bias in models while ensuring that AI systems remain transparent and understandable. This is critical to building societal trust in these technologies and ensuring their fair use. The L3S studies on hybrid systems against bias in learning models (page 12) and on FairTrade (page 14) demonstrate that it is possible to design AI systems that are both fair and efficient.

Sustainable knowledge transfer and education

Excellent research goes beyond publishing articles – it also involves transferring knowledge to the next generation of scientists. The L3S Research Center makes an important contribution to the training of young researchers by offering programs for the advancement of young scientists. To promote exchange, L3S organizes workshops, seminars, and international conferences, such as the ACM International Conference on Web Search and Data Mining, which will take place in Hannover, Germany, in March 2025. This helps to connect and inspire our young researchers with top researchers from around the world and accelerate innovation in all fields.

L3S Best Publications: l3s.de/category/best-publications/

Contact

Prof. Dr. Wolfgang Nejdl

Wolfgang Nejdl is the managing director of the L3S Research Center.

Prof. Dr. Marius Lindauer

L3S member Marius Lindauer heads the Institute for Artificial Intelligence at Leibniz Universität Hannover.