arXiv:2510.01038v2 Announce Type: replace Abstract: Perturbation-based explainability methods face criticism due to their reliance on out-of-distribution mutants. This raises doubts about the quality of the explanations. In this paper, we introduce a novel forward pass paradigm, Activation-Deactivation (AD), which obviates the need for perturbation of the input. AD replaces perturbation of input features with switching off parts of the model corresponding to to the intended perturbations. We implement ConvAD, an AD approximation algorithm for CNNs. ConvAD is a drop-in mechanism that can be eas
The increasing scrutiny and regulatory focus on AI explainability, driven by the desire for trustworthy and accountable AI, is pushing for advancements in robust interpretability methods.
Improved explainable AI (XAI) methods can enhance trust, facilitate debugging, and accelerate adoption of advanced AI systems in critical applications where transparency is paramount.
This new paradigm shifts explainability away from input perturbation and towards model-internal deactivation, offering a potentially more robust and reliable approach to understanding AI decisions.
- · AI developers and researchers
- · Enterprises adopting AI in regulated industries
- · AI explainability platforms
- · AI auditing and compliance firms
- · Companies relying solely on perturbation-based XAI methods
The adoption of Activation-Deactivation (AD) could lead to more reliable and less 'hackable' AI explanations.
Increased trust in AI systems could accelerate their deployment in high-stakes domains like healthcare, finance, and autonomous systems.
Greater transparency might expose unexpected biases or vulnerabilities in widely used AI models, leading to a new wave of model refinement and ethical AI development.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.AI