![[2510.00664] Batch-CAM: Introduction to better reasoning in convolutional deep learning models](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2510.00664] Batch-CAM: Introduction to better reasoning in convolutional deep learning models
Summary
The paper introduces Batch-CAM, a training framework for convolutional deep learning models that enhances interpretability by aligning model focus with class-representative features without pixel-level annotations.
Why It Matters
As deep learning models become prevalent in high-stakes applications, their opacity poses challenges for deployment. Batch-CAM addresses this issue by improving model interpretability, which is crucial for trust and accountability in AI systems. This framework could lead to broader adoption of AI in sensitive domains by ensuring models are both accurate and explainable.
Key Takeaways
- Batch-CAM integrates into the training loop with minimal computational overhead.
- It uses two regularization terms to enhance model focus on relevant features.
- The method produces more coherent saliency maps compared to existing techniques.
- Maintains competitive classification accuracy while reducing spurious feature activation.
- Offers a scalable approach to training interpretable models in deep learning.
Computer Science > Artificial Intelligence arXiv:2510.00664 (cs) [Submitted on 1 Oct 2025 (v1), last revised 13 Feb 2026 (this version, v2)] Title:Batch-CAM: Introduction to better reasoning in convolutional deep learning models Authors:Giacomo Ignesti, Davide Moroni, Massimo Martinelli View a PDF of the paper titled Batch-CAM: Introduction to better reasoning in convolutional deep learning models, by Giacomo Ignesti and 1 other authors View PDF HTML (experimental) Abstract:Deep learning opacity often impedes deployment in high-stakes domains. We propose a training framework that aligns model focus with class-representative features without requiring pixel-level annotations. To this end, we introduce Batch-CAM, a vectorised implementation of Gradient-weighted Class Activation Mapping that integrates directly into the training loop with minimal computational overhead. We propose two regularisation terms: a Prototype Loss, which aligns individual-sample attention with the global class average, and a Batch-CAM Loss, which enforces consistency within a training batch. These are evaluated using L1, L2, and SSIM metrics. Validated on MNIST and Fashion-MNIST using ResNet18 and ConvNeXt-V2, our method generates significantly more coherent and human-interpretable saliency maps compared to baselines. While maintaining competitive classification accuracy, the framework successfully suppresses spurious feature activation, as evidenced by qualitative reconstruction analysis. Batch-CAM ...
