Getting the most out of classification images
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
The classification image technique is a method of estimating an observer's internal template on a detection or discrimination task. Originally used in the context of Vernier acuity (Ahumada 1996), this approach has recently been adapted to more complex tasks, including disparity processing (Neri et al. 1999), illusory contour completion (Gold et al. 2000) and face recognition (Sekular et al. 2004). The nature of the procedure limits the number of stimulus dimensions that can be probed, as well as the resolution. We therefore sought an analysis procedure that would maximize the efficiency of the classification image technique. A widely-used approach to statistical testing of classification images is to apply a global threshold, along with a Bonferroni correction, to individual image components, a method which ignores correlations between adjacent image components. More efficient methods are available. For instance, hard thresholding of image components in overcomplete tight frames yields efficient image denoising (Yu et al. 1996). False discovery rate (FDR) testing has been shown to be as conservative as the Bonferroni correction in terms of global type I error, yet less prone to type II errors (Benjamini & Hochberg 1995). We adapted these two methods to the statistical testing of classification images. The hybrid FDR/tight frame method was applied to classification images from a simulated LAM observer, using a variety of idealized observer templates from previously published classification image experiments. The number of trials required to reach a desired Pearson's correlation (0.5) between estimated and true template was typically an order of magnitude lower with the hybrid technique than with Bonferroni thresholding. Improvements were greatest in templates with complex, oriented features, such as faces. These results suggest that the hybrid method improves the efficiency of classification image measurements, particularly in experiments with high-resolution or high-dimensional stimuli.
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 | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| 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