The SCKAN Natural Language Interface (SCKAN NLI, http://fdi-nlp.ucsd.edu/) is an AI-based query interface designed to retrieve high-level connectivity knowledge from SCKAN. It employs few-shot learning with the GPT-4 (GPT-4o mini) model to extract potential semantic frames related to neuronal parts from natural language expressions. Based on these extracted frames, the NLI formulates and executes queries against SCKAN using the SCKAN-Explorer API. It then applies zero-shot learning to generate answers in context with the user’s question. Leveraging a Retrieval-Augmented Generation (RAG) technique, the NLI synthesizes responses based on the retrieved data from SCKAN, thereby enhancing both relevance and accuracy.
Example of stimulation/virtual electrode kind of question
What anatomical structures can be stimulated by the inferior mesenteric ganglion (IMG)? In this case, IMG may act as the origin, destination, or via structure in the connections. Include all the structures for the list of structures categorized by the origin, via, and destinations.
