Deep Probabilistic Logic Module for Automated Self-Supervision
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning methods require large amounts of labeled data, which can be expensive or impractical to obtain, especially in domains like healthcare, and self-supervised learning with noisy or incomplete labels is inefficient due to the need for manual creation of self-supervisions.
Innovation Solution
A method using a deep probabilistic logic module that automatically determines and adds new virtual evidence to expand the training data, allowing for iterative self-supervision and reducing the need for human input through structured self-training and feature-based active learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If self-supervised learning uses manually crafted self-supervisions, then model training can proceed with limited labeled data, but the process becomes tedious and challenging
Solution Approach 1:
The system enables automated self-supervision generation where the deep probabilistic logic module automatically creates virtual evidence and factors without manual intervention. The model learns to generate its own training signals by leveraging prior knowledge and automatically determining new factors from output probabilities, eliminating the tedious manual crafting process while maintaining the benefit of training with limited labeled data
Solution Approach 2:
The system dynamically adjusts the complexity and types of virtual evidence based on the model's learning progress. By changing the parameters of the probabilistic logic module and automatically determining new factors from output probabilities, the system adapts the self-supervision process to improve efficiency without requiring manual intervention at each stage
2Ease of manufacture
If deep probabilistic logic uses pre-specified self-supervision, then training can be performed via variational EM, but manually crafting self-supervisions remains tedious and challenging
Solution Approach 1:
The system automatically generates virtual evidence and determines new factors without manual intervention. The deep probabilistic logic module uses its output probabilities to self-generate training signals, eliminating the time-consuming manual self-supervision creation process while maintaining the structured training approach through variational EM
Solution Approach 2:
The system pre-generates virtual evidence and factors before formal model training begins. By automatically determining new factors from output probabilities in advance and iteratively adding them to the training process, the system prepares comprehensive self-supervision materials that streamline the subsequent variational EM training without manual time investment
3Extent of automation
If automated self-supervision generation is implemented, then manual crafting is reduced, but new virtual evidence must be automatically determined and added
Solution Approach 1:
The system uses feedback loops where the deep probabilistic logic module's output probabilities are fed back into the factor determination process. This automatic feedback mechanism allows the system to iteratively determine new factors from its own output and add them as virtual evidence, achieving high automation while managing complexity through structured iterative refinement rather than complex one-shot processing
Data Source
AI summary
The present disclosure relates to devices and methods for determining new virtual evidence to use with a deep probabilistic logic module. The devices and methods may receive output from a deep probabilistic logic module in response to running an initial set of virtual evidence through the deep probabilistic logic module. The devices and methods may use the output to automatically propose at least one factor as new virtual evidence for use with the deep probabilistic logic module. The devices and methods may add the new virtual evidence to the deep probabilistic logic module.