Das sind Ihre Aufgaben
HITS is looking for a
Postdoctoral Researcher in Large Language Models and Knowledge Graphs (m/f/d)
to join the Machine Learning and Artificial Intelligence (MLI) Group to perform research at the intersection of large language models and knowledge graphs. A known disadvantage of large language models is their lack of factual correctness. Often it is not clear if generated content is factually correct, and even worse, often it is not possible to decide based on the generated output if the answer is correct: The generated content of a large language model sounds confident independently of its correctness.
The current state-of-the-art approach for making language models more factual is based on so called retrieval augmented generation, where additional input data is appended to a user’s query.
In this project, we want to develop an alternative approach that does not only add additional information to a user’s query, but instead combines inference within a language model with inference on structured data such as knowledge graphs and Bayesian networks, with the goal of making the language model more factual. To this end we develop a joint inference process that combines inference in a transformer model with inference on structured data and derive these inference algorithms in a common theoretical framework.
Postdoctoral Researcher in Large Language Models and Knowledge Graphs (m/f/d)
to join the Machine Learning and Artificial Intelligence (MLI) Group to perform research at the intersection of large language models and knowledge graphs. A known disadvantage of large language models is their lack of factual correctness. Often it is not clear if generated content is factually correct, and even worse, often it is not possible to decide based on the generated output if the answer is correct: The generated content of a large language model sounds confident independently of its correctness.
The current state-of-the-art approach for making language models more factual is based on so called retrieval augmented generation, where additional input data is appended to a user’s query.
In this project, we want to develop an alternative approach that does not only add additional information to a user’s query, but instead combines inference within a language model with inference on structured data such as knowledge graphs and Bayesian networks, with the goal of making the language model more factual. To this end we develop a joint inference process that combines inference in a transformer model with inference on structured data and derive these inference algorithms in a common theoretical framework.