AI Dataset Management via Active Learning and Distributed Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models require large amounts of labeled data for training, but existing active learning techniques are limited when applied to high-dimensional data, leading to inefficiencies in labeling and resource consumption.
Innovation Solution
A system and method that combines active learning and deep learning to efficiently train AI models by creating a data mask describing the labeling status of dataset items, allowing parallel training on multiple nodes, and optimizing hyperparameters to reduce the need for extensive labeling and improve model performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If massive datasets are used to train AI models, then model quality and performance are improved, but resource consumption and labeling costs increase
Solution Approach 1:
The patent segments the massive dataset into multiple smaller subsets that can be processed in parallel across distributed computing nodes. Each node receives a portion of the data and trains local model instances, enabling scalable processing without requiring all data to be loaded into a single system, thus maintaining model quality while reducing per-node resource consumption.
Solution Approach 2:
The system performs preliminary actions by pre-processing and preparing data subsets before distribution to computing nodes. Data is organized, validated, and formatted in advance, allowing parallel training to proceed efficiently without bottlenecks during the actual training phase, thereby reducing overall resource consumption while maintaining model performance.
2Manufacturing precision
If more data is labeled by human annotators, then training data quality improves, but time and cost increase
Solution Approach 1:
The patent applies partial action by labeling only a subset of data that is sufficient to achieve the desired model performance threshold. Rather than labeling all available data, the system identifies and labels the most informative portions, reducing labeling time and cost while maintaining adequate training data quality through iterative model improvement.
Solution Approach 2:
The system enables continuous useful action by implementing an iterative process where labeled data is used to train models, which then generate predictions that are refined through additional selective labeling. This continuous cycle allows the model to improve progressively without requiring complete labeling of all data upfront, reducing overall labeling time while maintaining quality.
3Reliability
If complex AI models are trained thoroughly, then model performance improves, but computational resources and training time increase
Solution Approach 1:
The patent segments the complex training process into multiple smaller training jobs distributed across numerous computing nodes. Each node processes a portion of the data with a simplified model instance, and results are aggregated to achieve the performance of a comprehensive complex model. This segmentation reduces the computational energy burden on any single node while maintaining overall model performance.
Solution Approach 2:
The system creates multiple copies of simplified model instances that run in parallel across distributed nodes. Each copy processes a subset of data independently, and their collective results approximate the outcome of a single complex model. This copying approach distributes computational energy consumption across many nodes, reducing the energy burden on individual components while achieving comparable performance.
4Loss of energy
If datasets are processed sequentially, then resource requirements are reduced, but processing speed decreases
Solution Approach 1:
The patent segments the dataset and processing workload into independent chunks that can be handled by multiple computing nodes simultaneously. Each node processes its assigned segment independently, enabling parallel execution that dramatically increases processing speed while keeping per-node resource consumption manageable through proper workload distribution.
Solution Approach 2:
The system transitions from sequential single-dimensional processing to parallel multi-dimensional processing by distributing data across spatial dimensions (multiple nodes). This dimensional change allows simultaneous processing of multiple data segments, increasing processing speed while maintaining controlled resource usage through distributed architecture.
Data Source
AI summary
A method and system for managing a dataset. An artificial intelligence (AI) model is to be used on the dataset. A data mask describes a labeling status of the data items. A loop is repeated until patience parameters are satisfied. The loop comprises receiving trusted labels provided by trusted labelers; updating the data mask; from a labelled data items subset, training the AI model; cloning the trained AI model into a local AI model on processing nodes; creating and chunking a randomized unlabeled subset into data subsets for dispatching to the processing nodes; receiving an indication that predicted label answers have been inferred by the processing nodes using the local AI model; computing a model uncertainty measurement from statistical analysis of the predicted label answers. The patience parameters include one or more of a threshold value on the model uncertainty measurement and information gain between different training cycles.