Image data bias detection with explainability in machine learning
View Patent ↗Bias in Machine Learning (ML) is when an ML algorithm tends to incompletely learn relevant and important patterns from a dataset, or learns the patterns from data incorrectly. Such inaccuracy can cause the algorithm to miss important relationships between patterns and features in data, resulting in inaccurate algorithm predictions. Systems and methods for detecting potential ML bias in input image datasets are described herein. After a target image is received, a subset of images related to the target image is extracted. The target image and subset of images are analyzed under an imbalance assessment and data bias assessment to determine the presence of any potential data bias in a ML training pipeline. If any data bias is determined, one or more messages summarizing the assessments and including explanations to enable more accurate predictions in image assessments are sent to the user.
1. A computer-implemented method of a computing system comprising a server and a database, the method comprising:
receiving a target image, wherein the target image comprises one or more attributes;
extracting a subset of images from an image database based on the one or more attributes;
analyzing the subset to determine an imbalance assessment;
performing mean shift clustering on each image in the subset and the target image, wherein the mean shift clustering comprises:
computing a Hopkins statistic on data points of a first image;
determining a quantile value based on the computed Hopkin statistic;
determining a kernel bandwidth based on a second quantity of the data points and the determined quantile value; and
performing the mean shift clustering based on the determined kernel bandwidth;
determining one or more clusters in each image in the subset and the target image based on the performed mean shift clustering;
performing data bias assessment on the one or more clusters of the target image, wherein the data bias assessment is performed according to a machine learning (ML) algorithm;
sending a first message of data bias according to a determination of an imbalance based on the imbalance assessment; and
sending a second message of data bias according to a determination of an inconsistency based on the performed data bias assessment.
2. The computer-implemented method of claim 1 , wherein, prior to the receiving the target image, assigning the one or more attributes to the target image.
3. The computer-implemented method of claim 1 , wherein, after the receiving the target image, determining the one or more attributes based on an attribute detection algorithm.
4. The computer-implemented method of claim 1 , wherein the determined imbalance assessment comprises:
determining one or more quantities of images in one or more image groups in the image database;
determining if a first quantity of the subset is below an imbalance threshold according to the one or more quantities of images in the one or more image groups; and
if the first quantity of the subset is below the imbalance threshold, determining the imbalance in the subset.
5. The computer-implemented method of claim 1 , wherein the performed data bias assessment comprises:
comparing the one or more clusters in the target image to the one or more clusters in each image in the subset;
determining a first difference between the one or more clusters in the target image to the one or more clusters in a first image in the subset;
determining a total difference wherein the total difference is a calculated balance of each difference for each of the images in the subset; and
if the total difference between the one or more clusters is greater than a consistency threshold, determining the inconsistency between the target image and the subset.
6. A computing system comprising:
one or more processors; and
a memory storing instructions that, when executed by the one or more processors, cause the one or more processors to:
receive a target image, wherein the target image comprises one or more attributes;
extract a subset of images from an image database based on the one or more attributes;
analyze the subset to determine an imbalance assessment;
perform mean shift clustering on each image in the subset and the target image;
determine one or more clusters in each image in the subset and the target image based on the performed mean shift clustering, wherein the mean shift clustering comprises:
computing a Hopkins statistic on data points of a first image;
determining a quantile value based on the computed Hopkin statistic
determining a darnel bandwidth based on a second quantity of the data points and the determined quantile value; and
performing the mean shift clustering based on the determined kernel bandwidth;
perform data bias assessment on the one or more clusters of the target image, wherein the data bias assessment is performed according to a machine learning (ML) algorithm;
send a first message of data bias according to a determination of an imbalance based on the imbalance assessment; and
send a second message of data bias according to a determination of an inconsistency based on the performed data bias assessment.
7. The computing system of claim 6 , wherein, prior to the receive the target image, assign the one or more attributes to the target image.
8. The computing system of claim 6 , wherein, after the receive the target image, determine the one or more attributes based on an attribute detection algorithm.
9. The computing system of claim 6 , wherein the determined imbalance assessment comprises:
determine one or more quantities of images in one or more image groups in the image database;
determine if a first quantity of the subset is below an imbalance threshold according to the one or more quantities of images in the one or more image groups; and
if the first quantity of the subset is below the imbalance threshold, determine the imbalance in the subset.
10. The computing system of claim 6 , wherein the performed data bias assessment comprises:
compare the one or more clusters in the target image to the one or more clusters in each image in the subset;
determine a first difference between the one or more clusters in the target image to the one or more clusters in a first image in the subset;
determine a total difference wherein the total difference is a calculated balance of each difference for each of the images in the subset; and
if the total difference between the one or more clusters is greater than a consistency threshold, determine the inconsistency between the target image and the subset.
11. A non-transitory storage medium storing instructions that, when executed by at least one processor of a computing system, cause the computing system to perform a method comprising:
receiving a target image, wherein the target image comprises one or more attributes;
extracting a subset of images from an image database based on the one or more attributes;
analyzing the subset to determine an imbalance assessment;
performing mean shift clustering on each image in the subset and the target image, wherein the mean shift clustering comprises:
computing a Hopkins statistic on data points of a first image;
determining a quantile value based on the computed Hopkin statistic;
determining a kernel bandwidth based on a second quantity of the data points and the determined quantile value; and
performing the mean shift clustering based on the determined kernel bandwidth;
determining one or more clusters in each image in the subset and the target image based on the performed mean shift clustering;
performing data bias assessment on the one or more clusters of the target image, wherein the data bias assessment is performed according to a machine learning (ML) algorithm;
sending a first message of data bias according to a determination of an imbalance based on the imbalance assessment; and
sending a second message of data bias according to a determination of an inconsistency based on the performed data bias assessment.
12. The non-transitory storage medium of claim 11 , wherein, prior to the receiving the target image, assigning the one or more attributes to the target image.
13. The non-transitory storage medium of claim 11 , wherein, after the receiving the target image, determining the one or more attributes based on an attribute detection algorithm.
14. The non-transitory storage medium of claim 11 , wherein the determined imbalance assessment comprises:
determining one or more quantities of images in one or more image groups in the image database;
determining if a first quantity of the subset is below an imbalance threshold according to the one or more quantities of images in the one or more image groups; and
if the first quantity of the subset is below the imbalance threshold, determining the imbalance in the subset.
15. The non-transitory storage medium of claim 11 , wherein the performed data bias assessment comprises:
comparing the one or more clusters in the target image to the one or more clusters in each image in the subset;
determining a first difference between the one or more clusters in the target image to the one or more clusters in a first image in the subset;
determining a total difference wherein the total difference is a calculated balance of each difference for each of the images in the subset; and
if the total difference between the one or more clusters is greater than a consistency threshold, determining the inconsistency between the target image and the subset.