Improving the Domain Adaptation of Retrieval Augmented Generation (RAG) Models for Open Domain Question Answering

Retrieval Augment Generation (RAG) is a recent advancement in Open-Domain Question Answering (ODQA). RAG has only been trained and explored with a Wikipedia-based external knowledge base and is not optimized for use in other specialized domains such as healthcare and news. In this paper, we evaluate the impact of joint training of the retriever and generator components of RAG for the task of domain adaptation in ODQA. We propose RAG-end2end, an extension to RAG that can adapt to a domain-specific knowledge base by updating all components of the external knowledge base during training. In addition, we introduce an auxiliary training signal to inject more domain-specific knowledge. This auxiliary signal forces RAG-end2end to reconstruct a given sentence by accessing the relevant information from the external knowledge base. Our novel contribution is that, unlike RAG, RAG-end2end does joint training of the retriever and generator for the end QA task and domain adaptation. We evaluate our approach with datasets from three domains: COVID-19, News, and Conversations, and achieve sig-nificant performance improvements compared to the original RAG model. Our work has been open-sourced through the HuggingFace Transformers library, attesting to our work's credibility and technical consistency. © 2023 Association for Computational Linguistics. Distributed under a CC-BY 4.0 license.

Jammify: Interactive Multi-sensory System for Digital Art Jamming

As social distancing is becoming the new normal, technology holds the potential to bridge this societal gap through novel interaction modalities that allow multiple users to collaborate and create content together. We present Jammify, an interactive multi-sensory system that focuses on providing a unique digital art-jamming experience with a visual display and a wearable arm-sleeve. The ‘jamming-canvas’ visual display is a two-sided LED light wall (2 m × 6 m) where users can draw free-hand gestures on either side and switch between two view modes: own-view and shared-view. The arm-sleeve uses shape-memory-alloy integrated fabric to sense and re-create a subtle and natural touch sensation on each other’s hands. We describe the details of the design and interaction possibilities based on the diverse combinations of both input and output modalities of the system, as well as findings from a user study with ten participants. © 2021, IFIP International Federation for Information Processing.