Research map / Vision-language and multimodal / Human and embodied modeling
Performance-optimized deep neural networks are evolving into worse models of inferotemporal visual cortex
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| train DNNs with the neural harmonizer (align gradients to human feature importance maps) | ImageNet-trained DNNs (6 architectures) · IT recordings, Monkey 1 ML | neural prediction accuracy: T(5) = 5.30, p < 0.01 | Helped |
| train DNNs with the neural harmonizer | ImageNet-trained DNNs · IT recordings, Monkey 1 PL | neural prediction accuracy: T(5) = 6.11, p < 0.001 | Helped |
| train DNNs with the neural harmonizer | ImageNet-trained DNNs · IT recordings, Monkey 2 ML | neural prediction accuracy: T(5) = 6.89, p < 0.001 | Helped |
| increase ImageNet accuracy (ConvNext tiny 74.3 vs ResNetv2-152x4 84.9) | pretrained DNNs · IT recordings | neural prediction accuracy: 74.3 -> 84.9 ImageNet accuracy, equal neural fit | No clear effect |
| train with self-supervision instead of supervised ImageNet | DNNs · IT recordings | neural prediction accuracy | No clear effect |
| train for adversarial robustness | DNNs · IT recordings | neural prediction accuracy | No clear effect |
| train on extra internet data beyond ImageNet / newer architectures such as ViTs | DNNs · IT recordings | neural prediction accuracy | No clear effect |
| increase ImageNet accuracy across 104 DNNs | pretrained DNNs · Brain-Score IT datasets | neural prediction accuracy | Hurt |
Closest papers by meaning
- Harmonizing the object recognition strategies of deep neural networks with humans (NeurIPS 2022 · similarity 0.92)
- Towards robust vision by multi-task learning on monkey visual cortex (NeurIPS 2021 · similarity 0.91)
- Simulating a Primary Visual Cortex at the Front of CNNs Improves Robustness to Image Perturbations (NeurIPS 2020 · similarity 0.90)
- Brain-Like Object Recognition with High-Performing Shallow Recurrent ANNs (NeurIPS 2019 · similarity 0.90)
- Explicitly Modeling Subcortical Vision with a Neuro-Inspired Front-End Improves CNN Robustness (NeurIPS 2025 · similarity 0.89)
- Benchmarking Out-of-Distribution Generalization Capabilities of DNN-based Encoding Models for the Ventral Visual Cortex. (NeurIPS 2024 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.