Introduction week
Learn the limitations of RAG, go through PubMedQA Dataset, test code for MedRAG and MIRAGE, develop solutions to optimize RAG and increase accuracy
Ran and tested code, preprocessed all 5 datasets of MIRAGE, tested Llama3 on MIRAGE datasets, jot down results and improve later on, test GPT4 and GPT3.5 on CoT and MedRAG
Preprocess PubMedQA labeled dataset for testing, test code on LLAMA3 with different retrievers and embedding models, implement MedCPT and BioBert
Continue testing, research effective solutions to optimize RAG; Simple Contrastive Sentence Embeddings and Contextual Document Embeddings, weaviate tutorials and how to implement into RAG system
Research other retriever models for better performance of LLAMA3. Test each retriever on how accurately it was retrieving the correct contexts
Preprocess data for training, begin training BERT model
Implement model into RAG pipeline and get results, compare with other retrievers
Outline ways model can be optimized, future work for RAG pipeline, begin working on research poster and final paper
Complete research poster and final paper, complete any unfinished testing