This ongoing project aims to develop a novel hybrid architecture called LLM-ACTR designed to address the limitations of Large Language Models (LLMs) in decision-making.
The repository is organized into the following main directories:
/VSM-ACTR: This directory hosts all resources related to the VSM-ACTR cognitive model./Notebooks: Contains notebooks to support results of aaai submission.
/model: Core model file./raw data: output sample from cognitive model./processed data: preprocessed dataset
/feature extraction: Notebooks for feature extraction for behavior prediction./finetuning: Notebook for fine-tuning for knowledge transfer./baseline: Notebooks with two baselines./salient embedding: Notebooks for VSM-ACTR traces semantic extraction and dimension reduction./VSM-ACTR analysis: Notebook to analysis VSM-ACTR.