Pulling Back the Curtain on Deep Networks
Fuente:
arXiv
Salvato in:
| Autori principali: | , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866918487232872448 |
|---|---|
| author | Satkiewicz, Maciej Corizzo, Roberto Pietroń, Marcin |
| author_facet | Satkiewicz, Maciej Corizzo, Roberto Pietroń, Marcin |
| contents | In linear models, visualizing a weight vector naturally reveals the model's preferred input direction, but extending this intuition to deep networks via gradients or gradient ascent often yields brittle or adversarial-looking features. We argue that deep networks are better understood as input-conditioned affine operators, whose natural adjoint action pulls a neuron's preferred direction back to input space. We further refine this representation by backward-only softening and iterative enhancement to reconstruct coherent local structures encoded by the target neuron. This provides a unifying perspective on previously disparate ideas such as SmoothGrad, B-cos-style alignment, and Feature Accentuation. The resulting Semantic Pullbacks (SP) generate perceptually aligned, class-conditional post-hoc explanations that emphasize semantically meaningful features, facilitate coherent counterfactual perturbations, and remain theoretically grounded. Across convolutional architectures (ResNet50, VGG) and transformer-based models (PVT), Semantic Pullbacks achieve the best overall trade-off across faithfulness, stability, and target-sensitivity benchmarks, while remaining general, computationally efficient, and readily integrable into existing deep learning pipelines. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2507_22832 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Pulling Back the Curtain on Deep Networks Satkiewicz, Maciej Corizzo, Roberto Pietroń, Marcin Machine Learning Computer Vision and Pattern Recognition Neural and Evolutionary Computing I.2.6; I.4.10 In linear models, visualizing a weight vector naturally reveals the model's preferred input direction, but extending this intuition to deep networks via gradients or gradient ascent often yields brittle or adversarial-looking features. We argue that deep networks are better understood as input-conditioned affine operators, whose natural adjoint action pulls a neuron's preferred direction back to input space. We further refine this representation by backward-only softening and iterative enhancement to reconstruct coherent local structures encoded by the target neuron. This provides a unifying perspective on previously disparate ideas such as SmoothGrad, B-cos-style alignment, and Feature Accentuation. The resulting Semantic Pullbacks (SP) generate perceptually aligned, class-conditional post-hoc explanations that emphasize semantically meaningful features, facilitate coherent counterfactual perturbations, and remain theoretically grounded. Across convolutional architectures (ResNet50, VGG) and transformer-based models (PVT), Semantic Pullbacks achieve the best overall trade-off across faithfulness, stability, and target-sensitivity benchmarks, while remaining general, computationally efficient, and readily integrable into existing deep learning pipelines. |
| title | Pulling Back the Curtain on Deep Networks |
| topic | Machine Learning Computer Vision and Pattern Recognition Neural and Evolutionary Computing I.2.6; I.4.10 |
| url | https://arxiv.org/abs/2507.22832 |