Neural Network Topology Learning via Unlabeled Data

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods for learning topologies of deep learning neural networks are inefficient, relying on heuristic approaches and requiring labeled data, which limits their applicability and accuracy.

Innovation Solution

A novel technique for learning efficient and hardware-friendly topologies of deep learning networks using unlabeled data, which involves deep latent variable models, automatic topology learning, and on-the-fly structure learning, allowing for optimal solutions and flexible network designs.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If conventional heuristic trial-and-error approaches are used to learn network topology, then some topology can be obtained, but the process is inefficient and requires labeled data

Engineering Contradiction:
Improvetopology learning efficiencyVSAvoidrequirement for labeled data
Core Design Contradiction:
ProductivityVSLoss of information

Solution Approach 1:

The system performs self-service by automatically learning network topology structures from unlabeled data through unsupervised learning algorithms. The deep belief network automatically adjusts its own architecture and parameters without requiring external labeled data or manual heuristic tuning, enabling the network to serve its own topology learning needs independently.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent replaces the mechanical trial-and-error approach with an automated computational system based on deep belief networks. Instead of manually adjusting topology through heuristic methods, the system uses unsupervised learning algorithms that automatically discover optimal network structures from data, substituting manual mechanical processes with intelligent automated systems.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Measurement precision

If deep neural networks with complex topologies are used to improve accuracy, then task performance increases, but computational power requirements and hardware costs increase

Engineering Contradiction:
Improvetask accuracyVSAvoidcomputational power consumption
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The system employs dynamic topology learning where the network architecture is not fixed but adapts based on the specific task requirements and data characteristics. The deep belief network dynamically determines the optimal number of hidden layers, neurons per layer, and connection structures during the learning process, allowing the system to achieve high accuracy with minimal computational resources by avoiding unnecessary network complexity.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent changes key parameters of the neural network including the number of hidden layers, neurons per layer, and connection densities based on unsupervised learning from unlabeled data. By automatically optimizing these parameters rather than using fixed complex architectures, the system achieves task-specific accuracy while minimizing computational power consumption and hardware requirements.

Inventive Principle:
Principle #35Parameter changes

3Adaptability or versatility

If manual hand-crafting of network topology is used, then intuition-based designs can be created, but the process is time-consuming and lacks systematic optimization

Engineering Contradiction:
Improvetopology design flexibilityVSAvoidtopology development time
Core Design Contradiction:
Adaptability or versatilityVSLoss of time

Solution Approach 1:

The system performs preliminary action by pre-training deep belief networks on large volumes of unlabeled data to learn optimal topology structures before actual task execution. This pre-learning phase automatically discovers effective network architectures that can be directly applied to specific tasks, eliminating the need for time-consuming manual design and trial-and-error optimization for each new application.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent implements feedback mechanisms where the deep belief network continuously evaluates its own performance and automatically adjusts its topology structure based on learning outcomes. The system uses unsupervised learning feedback from unlabeled data to refine network architecture, enabling systematic optimization that maintains design flexibility while dramatically reducing development time compared to manual methods.

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP3399471B1Efficient learning and using of topologies of neural networks in machine learning
Publication Date: 2025.03.05 INTEL CORP
  • EP3399471B1 patent drawingFigure 1
  • EP3399471B1 patent drawingFigure 2A
  • EP3399471B1 patent drawingFigure 2B

AI summary

A mechanism is described for facilitating learning and application of neural network topologies in machine learning at autonomous machines. A method of embodiments, as described herein, includes monitoring and detecting structure learning of neural networks relating to machine learning operations at a computing device having a processor, and generating a recursive generative model based on one or more topologies of one or more of the neural networks. The method may further include converting the generative model into a discriminative model.