Iterative ANN Training with Small Data Samples
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Training classification engines using artificial neural networks (ANNs) with unevenly distributed data categories leads to poor performance due to imbalances in learning algorithms, requiring large training sets and significant computing resources.
Innovation Solution
An iterative training procedure that selects a small sample of training data, trains the model, and evaluates its performance, repeating cycles of sampling, training, and inference until specified criteria are met, allowing for efficient use of computing resources and overcoming instability issues.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If larger training sets are used to address category imbalance, then recognition performance across all categories improves, but computing resource requirements increase significantly
Solution Approach 1:
The patent applies partial action by using a small subset of training data (e.g., 100 images) rather than the full training set. This selective sampling approach achieves satisfactory model performance without requiring computationally expensive processing of large datasets, directly resolving the contradiction between recognition performance and computing resource consumption
Solution Approach 2:
The patent performs preliminary action by conducting iterative training cycles with small data samples before final model deployment. Through multiple rounds of training, validation, and error analysis on small subsets, the model is progressively improved to achieve high performance without needing to process large datasets in each training cycle
2Productivity
If iterative training with small samples is used, then computing resource efficiency improves, but training stability may deteriorate
Solution Approach 1:
The patent implements feedback mechanisms through iterative training cycles where the model is trained on small samples, evaluated on validation data, and refined based on performance metrics. Error analysis from each iteration provides feedback for adjusting training parameters and selecting problematic cases for retraining, ensuring stability while maintaining efficiency
Solution Approach 2:
The patent maintains continuity of useful action by performing multiple iterative training cycles continuously. Rather than single-pass training, the system repeatedly trains on small samples, validates performance, and refines the model, ensuring stable convergence while keeping computational costs low through continuous improvement
Data Source
AI summary
A method for generating a classification model using a training data set. An iterative procedure for training an ANN model, in which an iteration includes selecting a small sample of training data from a source of training data, training the model using the sample, using the model in inference mode over a larger sample of the training data, and reviewing the results of the inferencing. The results can be evaluated to determine whether the model is satisfactory, and if it does not meet specified criteria, then cycles of sampling, training, inferencing and reviewing results (STIR cycles) are repeated in an iterative process until the criteria are met. A classification engine trained as described herein is provided.