Training Data Outlier Removal for Image Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning algorithms do not adequately optimize training data sets, leading to low-representative examples that can introduce ambiguity or errors during model training, affecting the precision of the resulting model.
Innovation Solution
A method involving a machine learning system that includes a training module to iteratively reduce the training data set by eliminating outliers using clustering and distance learning algorithms, such as a Siamese neural network, to improve data separability and classification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If the full training data set is used for model training, then the quantity of training data is maximized, but low-representative examples introduce ambiguity and reduce classification precision
Solution Approach 1:
The patent extracts and removes outlier data points from the training data set using clustering algorithms and distance metrics. By identifying data points that fall outside normal clusters or have excessive distance from cluster centers, the system extracts only the representative portions of the data for training, thereby eliminating ambiguous examples while preserving the beneficial quantity of training data.
Solution Approach 2:
The patent applies different quality standards to different portions of the training data. Instead of treating all data points uniformly, the system evaluates each data point's representativeness locally through clustering analysis and distance metrics. This allows the system to retain high-quality representative examples while discarding low-quality outliers, thereby improving overall classification precision without unnecessarily reducing the total training data quantity.
2Measurement precision
If outlier removal processes are applied to the training data set, then classification precision is improved, but the complexity of the training process increases
Solution Approach 1:
The patent performs outlier removal as a preliminary action before the main model training process. By pre-processing the training data to eliminate outliers using clustering algorithms and distance metrics, the system prepares a cleaned data set that requires less complex training procedures. This preliminary action reduces the complexity of subsequent training by preventing ambiguous examples from interfering with the learning process.
Solution Approach 2:
The patent introduces clustering algorithms and distance metrics as intermediary tools between the raw training data and the model training process. These intermediaries automatically identify and flag outlier data points based on their spatial relationships in the feature space, providing an objective criterion for outlier removal that reduces the need for complex manual intervention or heuristic-based filtering during training.
Data Source
AI summary
A method for machine learning-based classification may include training a machine learning model with a full training data set, the full training data set comprising a plurality of data points, to generate a first model state of the machine learning model, generating respective embeddings for the data points in the full training data set with the first model state of the machine learning model, applying a clustering algorithm to the respective embeddings to generate one or more clusters of the embeddings, identifying outlier embeddings from the one or more clusters of the embeddings, generating a reduced training data set comprising the full training data set less the data points associated with the outlier embeddings, training the machine learning model with the reduced training data set to a second model state, and applying the second model state to one or more data sets to classify the one or more data sets.