# Understanding Individual Decisions of CNNs via Contrastive Backpropagation

arXiv: https://arxiv.org/abs/1812.02100 · HTML: https://mlsanity.com/paper/1812.02100/

## Key facts

- Cited by 40 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 5 Dec 2018.

## Built on by (most cited first, 30 of 40)

- [Generic Attention-model Explainability for Interpreting Bi-Modal and Encoder-Decoder Transformers](https://mlsanity.com/paper/2103.15679/) (2021)
- [Interpretable Deep Learning: Interpretation, Interpretability, Trustworthiness, and Beyond](https://mlsanity.com/paper/2103.10689/) (2021)
- [Are Vision Transformers Robust to Patch Perturbations?](https://mlsanity.com/paper/2111.10659/) (2021)
- [Gradient based Feature Attribution in Explainable AI: A Technical Review](https://mlsanity.com/paper/2403.10415/) (2024)
- [Explainable Reinforcement Learning for Broad-XAI: A Conceptual Framework and Survey](https://mlsanity.com/paper/2108.09003/) (2021)
- [Towards Evaluating Explanations of Vision Transformers for Medical Imaging](https://mlsanity.com/paper/2304.06133/) (2023)
- [Vision Transformer using Low-level Chest X-ray Feature Corpus for COVID-19 Diagnosis and Severity Quantification](https://mlsanity.com/paper/2104.07235/) (2021)
- [Token Transformation Matters: Towards Faithful Post-hoc Explanation for Vision Transformer](https://mlsanity.com/paper/2403.14552/) (2024)
- [HINT: Hierarchical Neuron Concept Explainer](https://mlsanity.com/paper/2203.14196/) (2022)
- [Latent Space Explanation by Intervention](https://mlsanity.com/paper/2112.04895/) (2021)
- [ViTOL: Vision Transformer for Weakly Supervised Object Localization](https://mlsanity.com/paper/2204.06772/) (2022)
- [Rather a Nurse than a Physician -- Contrastive Explanations under Investigation](https://mlsanity.com/paper/2310.11906/) (2023)
- [New Definitions and Evaluations for Saliency Methods: Staying Intrinsic, Complete and Sound](https://mlsanity.com/paper/2211.02912/) (2022)
- [LRP-QViT: Mixed-Precision Vision Transformer Quantization via Layer-wise Relevance Propagation](https://mlsanity.com/paper/2401.11243/) (2024)
- [Grad-ECLIP: Gradient-based Visual and Textual Explanations for CLIP](https://mlsanity.com/paper/2502.18816/) (2025)
- [Do DALL-E and Flamingo Understand Each Other?](https://mlsanity.com/paper/2212.12249/) (2022)
- [EventRPG: Event Data Augmentation with Relevance Propagation Guidance](https://mlsanity.com/paper/2403.09274/) (2024)
- [Empowering CAM-Based Methods with Capability to Generate Fine-Grained and High-Faithfulness Explanations](https://mlsanity.com/paper/2303.09171/) (2023)
- [Decoding Layer Saliency in Language Transformers](https://mlsanity.com/paper/2308.05219/) (2023)
- [Explaining Object Detectors via Collective Contribution of Pixels](https://mlsanity.com/paper/2412.00666/) (2024)
- [Quantifying Predictive Uncertainty in Medical Image Analysis with Deep Kernel Learning](https://mlsanity.com/paper/2106.00638/) (2021)
- [BSED: Baseline Shapley-Based Explainable Detector](https://mlsanity.com/paper/2308.07490/) (2023)
- [Gradient Frequency Modulation for Visually Explaining Video Understanding Models](https://mlsanity.com/paper/2111.01215/) (2021)
- [PAMI: partition input and aggregate outputs for model interpretation](https://mlsanity.com/paper/2302.03318/) (2023)
- [Hidden Conflicts in Neural Networks and Their Implications for Explainability](https://mlsanity.com/paper/2310.20363/) (2023)
- [On the Control of Attentional Processes in Vision](https://mlsanity.com/paper/2101.01533/) (2021)
- [Temporal Relevance Analysis for Video Action Models](https://mlsanity.com/paper/2204.11929/) (2022)
- [Illuminating Salient Contributions in Neuron Activation with Attribution Equilibrium](https://mlsanity.com/paper/2205.11109/) (2022)
- [On Model Explanations with Transferable Neural Pathways](https://mlsanity.com/paper/2309.09887/) (2023)
- [Towards Better Visualizing the Decision Basis of Networks via Unfold and Conquer Attribution Guidance](https://mlsanity.com/paper/2312.14201/) (2023)

## Abstract (arXiv)

A number of backpropagation-based approaches such as DeConvNets, vanilla Gradient Visualization and Guided Backpropagation have been proposed to better understand individual decisions of deep convolutional neural networks. The saliency maps produced by them are proven to be non-discriminative. Recently, the Layer-wise Relevance Propagation (LRP) approach was proposed to explain the classification decisions of rectifier neural networks. In this work, we evaluate the discriminativeness of the generated explanations and analyze the theoretical foundation of LRP, i.e. Deep Taylor Decomposition. The experiments and analysis conclude that the explanations generated by LRP are not class-discriminative. Based on LRP, we propose Contrastive Layer-wise Relevance Propagation (CLRP), which is capable of producing instance-specific, class-discriminative, pixel-wise explanations. In the experiments, we use the CLRP to explain the decisions and understand the difference between neurons in individual classification decisions. We also evaluate the explanations quantitatively with a Pointing Game and an ablation study. Both qualitative and quantitative evaluations show that the CLRP generates better explanations than the LRP. The code is available.