Novel Applications of Transformer Models in Data Interpretation and Visualization
- 0 Collaborators
This project explores the possibilities of using attention-based transformer models to aid in tasks related to data visualization and interpretation, through the use of various Intel oneAPI toolkits such as the DNN library, oneAPI AI Kit, as well as oneAPI Render Kit. ...learn more
Project status: Concept
oneAPI, Artificial Intelligence
Intel Technologies
DevCloud,
oneAPI,
AI DevCloud / Xeon,
Intel Opt ML/DL Framework
Overview / Usage
The problems being solved by this project are those related to data visualization and interpretation, which can often be very complex tasks. This work provides a new way to approach these tasks, using transformer models to help understand and interact with data more effectively. With its self-attention mechanism, the transformer model can learn complex patterns and correlations in data, which can be used to generate meaningful visualizations. This work can be used in production by data scientists and analysts who want to better understand and interact with their data.
Methodology / Approach
Transformer models have shown great promise in many fields, especially in Natural Language Processing tasks, such as machine translation, question answering, and generation. More recently, they have begun to be applied to other areas such as computer vision and time series analysis. In this project, we aim to explore these models' further potential in the data visualization and interpretation domain. The attention mechanism of the transformer will be used to provide a new way to understand complex data patterns and correlations. More specifically, the patterns and correlations learned in a self-supervised manner could be used to generate meaningful visualizations of the underlying correlations that the model has learned. This could be used to interactively highlight the most critical parts when visualizing large amounts of data from different perspectives.
Technologies Used
This project seeks to further explore these models' potential in data visualization and interpretation through the use of various Intel oneAPI toolkits. In particular, this project aims to use the DNN library and oneAPI AI Kit to develop and train a transformer model. The Intel DevCloud can be used to speed up the development through its many tools for DL training, inference, and data analytics as part of the AI Analytics Toolkit. Further, we will use the oneAPI Render Kit to develop a basic data visualization application for learned parameters such as the model's weights and then expand to other more complex 3D visualizations that can be interactively explored.