Observing Pre-Trained Convolutional Neural Network (CNN) Layers as Feature Extractor for Detecting Bias in Image Classification Data
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Detecting bias in data is crucial since it can pose serious problems when developing an AI algorithm. The research aims to propose a novel study design to detect bias in image classification data by using pretrained Convolutional Neural Network (CNN) layers as a feature extractor. There are three datasets used in the research with varying degrees of complexity, those are low, medium, and high complexity. There are Modified National Institute of Standards and Technology (MNIST) Digits, batik collections (Parang, Megamendung, and Kawung), and Canadian Institute for Advanced Research (CIFAR-10) datasets. Then, the researchers make a baseline workflow and substitute a step-in feature extraction with a convolution using the first pre-trained CNN layer and each of its kernels. Then, the researchers evaluate the effect of the experiments using accuracy. By observing the effect of the individual kernel, the research can better make sense of what happens inside a CNN layer. The research finds that color in the image is an essential factor when working with CNN. Furthermore, the proposed study design can detect bias in image classification data where it is related to the color of the image. Detecting this bias early is important in helping developers to improve AI algorithms.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.003 | 0.001 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.002 | 0.000 |
| Scholarly communication | 0.000 | 0.004 |
| Open science | 0.003 | 0.002 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it