top of page
TEAM
Improving RAG by Averaging
Qidu(Quentin) Fu, Gilyoung Cheong, Sixuan Lou, Junichi Koganemaru, Dapeng Shang, XINYUAN LAI
![clear.png](https://static.wixstatic.com/media/55f531_b9f3f13ce3aa4af78af2cc6d3563b81b~mv2.png/v1/fill/w_3,h_3,al_c,lg_1,q_85,enc_auto/clear.png)
We implement a specific pipeline of Retrieval-Augmented Generation (RAG) for a question answering machine using SBERT developed by Nils Reimers and Iryna Gurevych based on Google's BERT. Experimentally, we show that the one we implement (averaging RAG) is better than the other baseline one (naïve RAG) in retrieval based on two reasonable relative performance metrics. In the retrieval process, we also apply K-Means Clustering to reduce the runtime significantly.
![](https://static.wixstatic.com/media/a994932411404ef3bb797ba005125f5d.png/v1/fill/w_45,h_45,al_c,q_85,usm_0.66_1.00_0.01,enc_auto/a994932411404ef3bb797ba005125f5d.png)
![](https://static.wixstatic.com/media/a994932411404ef3bb797ba005125f5d.png/v1/fill/w_45,h_45,al_c,q_85,usm_0.66_1.00_0.01,enc_auto/a994932411404ef3bb797ba005125f5d.png)
![](https://static.wixstatic.com/media/a994932411404ef3bb797ba005125f5d.png/v1/fill/w_45,h_45,al_c,q_85,usm_0.66_1.00_0.01,enc_auto/a994932411404ef3bb797ba005125f5d.png)
bottom of page