Bibliographic record
Abstract
Background: Deep convolutional neural networks (DCNNs) trained to classify objects have reached remarkable levels of performance and are predictive of brain response in both human and non-human primates. However, DCNNs rely more on texture than shape relative to humans (Baker, Lu, Erlikhman & Kellman, 2018; Geirhos et al., 2018), and also appear to be biased toward local shape features (Baker et al., 2020, but see Keshvari et al. 2019). Here we employ a novel method to test for DCNN selectivity for the global shape of an object. Method: We used a dataset of animal silhouettes from 10 animal categories. To assess selectivity for global shape, we created two variants of these stimuli that disrupt the global configuration of the object while largely preserving local features. In the first variant, we flipped the top portion of the object left-to-right but maintained its smooth connection with the bottom of the object, thus disrupting global shape but preserving object coherence. In the second variant we also shifted the top portion laterally so that both global shape and global coherence were disrupted. We then analyzed the classification performance of the Resnet50 DCNN (He et al., 2016) on these stimuli, using two different training curricula: ImageNet alone, and ImageNet + Stylized Imagenet, which has been reported to improve performance on silhouettes (Geirhos et al., 2018). Results: We found that disrupting global shape while maintaining local shape and object coherence induced a ~60% drop in classification performance, while also disrupting coherence induced an additional ~80% drop. Interestingly, co-training on Stylized Imagenet did not mitigate these impacts and reduced performance on silhouettes overall. Implications: While prior work suggests that DCNNs are biased toward texture and local shape features, our findings suggest that at least some ImageNet-trained DCNNs are profoundly selective for global shape and object coherence.
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.
How this classification was reachedexpand
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 | 0.000 | 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 itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".