Al-POWERED CONCEPT-DRIVEN VISUALIZATION AUTHORING

Smart overview of the Invention

The patent application introduces a method and system for generating visualizations using AI-driven tools. It focuses on transforming existing data concepts into new ones through a program synthesizer and a generative model. The aim is to simplify the process of creating visualizations by automating data transformations, thus reducing the need for users to manually adjust data formats to fit specific visualization requirements.

Background

Traditional visualization tools require users to format data sets to match the input requirements of specific visualizations. This often necessitates the use of specialized software or manual data processing, which can be cumbersome and time-consuming. The proposed system addresses this challenge by allowing users to define high-level visualization intents without needing to manage the low-level data transformation steps themselves.

Core Functionality

The system employs an AI-powered visualization paradigm that separates the visualization intent from data transformation. Users can define data concepts in natural language or based on existing data examples. The program synthesizer and generative multimodal model then automatically transform these inputs into new data concepts suitable for visualization, offering feedback to help users understand the transformation process.

Data Transformation Process

The data transformation process leverages AI to automate the conversion of data concepts. It uses a program synthesizer to generate code that transforms data based on user-defined intents. Additionally, a generative multimodal model processes various data types to propose candidate transformations. This approach eliminates the need for external data processing tools by integrating AI capabilities directly into the visualization authoring process.

Generative Model Capabilities

The generative multimodal model can handle multiple data modalities, such as natural language and images, to generate new data concepts. It utilizes transformer-based Large Language Models (LLMs) to understand and process data, offering a flexible and powerful tool for data transformation. Examples of LLMs used include BLOOM and GPT-4, which are capable of understanding complex data structures and generating accurate visualizations.