Neural Graphical Model for Mixed Data Types

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Traditional probabilistic graphical models face limitations in handling complex distributions and are restricted by assumptions on data types, leading to high computational complexities and insufficient representation of relationships between features.

Innovation Solution

The development of neural graphical models that generate a neural view using a dependency structure, allowing for the representation of complex distributions without restrictions on data types, using a deep learning architecture with projection modules to handle mixed data types and support various graph structures.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If traditional probabilistic graphical models are used to model relationships between features, then the model structure and computational complexity are reduced, but the ability to handle complex distributions and mixed data types is limited

Engineering Contradiction:

Improveability to handle complex distributions and mixed data typesVSAvoidcomputational complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent introduces projection modules as intermediary components that transform different data types (images, text, tables) into a common embedding space. These projection modules act as mediators between the diverse input data and the graphical model, enabling the model to handle mixed data types without increasing overall computational complexity.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The neural graphical model is designed with universal components that can process multiple data types through the same framework. The graphical model structure serves multiple functions: representing dependencies, handling mixed data types, and performing inference, thereby increasing adaptability without proportionally increasing complexity.

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Measurement precision

If simplifying assumptions are made in probabilistic graphical models, then the computational complexity is reduced, but the representation accuracy of relationships between features is insufficient

Engineering Contradiction:

Improverepresentation accuracy of relationshipsVSAvoidcomputational complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent changes the parameters of the graphical model by using neural networks to represent the conditional probability distributions. Instead of using simple parametric forms, the neural networks learn complex non-linear relationships from data, significantly improving representation accuracy while the parameters are optimized through efficient training procedures.

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent replaces traditional mechanical probabilistic graphical model computations with neural network-based representations. The neural networks substitute for the complex mathematical computations required in traditional PGMs, enabling more accurate representation of relationships while maintaining computational efficiency through modern deep learning optimizations.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Adaptability or versatility

If neural graphical models are used to represent complex distributions, then the representation capability is improved, but the computational costs increase

Engineering Contradiction:

Improverepresentation capability of complex distributionsVSAvoidcomputational costs

Core Design Contradiction:

Adaptability or versatilityVSUse of energy by moving object

Solution Approach 1:

The patent segments the computational tasks by separating the data type-specific processing (projection modules) from the universal graphical model processing. This segmentation allows each component to be optimized independently, reducing overall computational costs while maintaining the ability to represent complex distributions.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The projection modules perform preliminary transformations of the input data into embedded representations before the data enters the graphical model. This preliminary action pre-processes and standardizes the data, reducing the computational burden on subsequent processing stages and enabling efficient handling of complex distributions.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS20240111988A1Neural graphical models for generic data types

Publication Date: 2024.04.04 MICROSOFT TECHNOLOGY LICENSING LLC

AI summary

The present disclosure relates to methods and systems for providing a neural graphical model. The methods and systems generate a neural view of the neural graphical model for a domain. The input data is generated from the domain and includes generic input data. The input data also includes a combination of different data types of input data. The neural view of the neural graphical model represents the functions of the different features of the domain using a neural network. The functions are learned for the features of the domain using a dependency structure of an input graph for the input data and the neural network. The methods and systems use the neural graphical model to perform inference tasks. The methods and systems also use the neural graphical model to perform sampling tasks.