Hierarchical Data Labeling With Graph Embedding for Multi-Level ML
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Supervised machine learning requires large amounts of labeled data, which becomes challenging as data sets grow in size and complexity, especially when multiple labels are needed, leading to inefficiencies in computational resources and energy consumption.
Innovation Solution
A semi-supervised, multi-level labeling framework using a hybrid model of random forest and graph embedding models to automatically generate multiple labels for data samples, reducing computational complexity and resource consumption by iteratively applying label features to generate additional labels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If supervised training with large amounts of labeled data is used, then machine learning model performance is improved, but computational resources and energy consumption increase
Solution Approach 1:
The system pre-processes and labels data in advance using automated labeling frameworks, creating ready-to-use training datasets before model training begins. This preliminary labeling action reduces the need for extensive manual labeling during model development cycles, thereby reducing overall computational resource consumption while maintaining model performance.
Solution Approach 2:
The patent introduces an intermediary automated labeling system that acts as a mediator between raw data and machine learning models. This intermediary framework uses semi-supervised learning and active learning techniques to generate high-quality labels with minimal human intervention, reducing the computational burden compared to traditional supervised labeling approaches.
2Adaptability or versatility
If data sets grow in size and complexity with multiple labels, then machine learning model capabilities are improved, but data labeling task difficulty increases
Solution Approach 1:
The patent segments the complex multi-level labeling task into hierarchical levels, where each level addresses specific labeling requirements. The system divides data samples into different categories and applies appropriate labeling strategies at each level, making the overall complex labeling process more manageable and systematic while maintaining versatility in handling diverse data types.
Solution Approach 2:
The system adds a hierarchical dimension to the labeling process by creating multiple levels of labels (e.g., coarse-grained categories at higher levels and fine-grained classifications at lower levels). This dimensional approach to labeling simplifies the complexity of multi-label tasks by organizing them in a structured hierarchy, making it easier to manage and process complex datasets.
3Measurement precision
If manual data labeling is performed to ensure data quality, then labeling accuracy is improved, but time and resource efficiency decrease
Solution Approach 1:
The patent implements feedback mechanisms where the automated labeling system continuously learns from and adjusts based on labeled data quality metrics. The system uses feedback loops to refine labeling accuracy over time, incorporating human-in-the-loop validation for uncertain cases while maintaining high overall efficiency. This feedback-driven approach ensures labeling accuracy comparable to manual methods while significantly improving productivity.
Solution Approach 2:
The system employs self-service automated labeling techniques where algorithms independently generate labels without requiring extensive human intervention. The framework uses self-supervised learning and consistency checks to maintain labeling accuracy while dramatically reducing the time and resources compared to manual labeling processes.
Data Source
AI summary
Implementations are directed to receiving a plurality of data samples comprising a first set of data samples associated with respective labels and a second set of data samples to be labeled; generating a random forest structure comprising a set of decisions trees, each decision tree including nodes corresponding to the first set of data samples; adding the second set of data samples into each decision tree as additional nodes of each decision tree; merging the set of decision trees to obtain a universal graph, wherein each node corresponds to a data sample; extracting, using a graph embedding algorithm, an embedding feature for each data sample that corresponds to each node included in the universal graph; determining a distance between any pair of two data samples using respective embedding features of the two data samples; and determining a label for each of the second set of data samples using the distance.