Back to skills
SKILL.md
Attribution Neuron Utility
ASecurityGradient-based eXplained Deviation (GXD) neuron utility metric for plasticity restoration in deep networks. Combines attribution-guided neuron importance scoring with critical bottleneck period (CBP) detection and targeted reset. Use when: catastrophic forgetting, plasticity loss, neuron-level analysis, continual learning reset methods, attribution-guided training.
- 3 stars
- 0 votes
- 0 copies
- 2 views
- Added September 11, 2026
Security analysis
100/100npx -y skills add hiyenwong/ai_collection --skill attribution-neuron-utility --agent claude-codeAre you the author of Attribution Neuron Utility?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/hiyenwong-attribution-neuron-utility)---
name: attribution-neuron-utility
description: "Gradient-based eXplained Deviation (GXD) neuron utility metric for plasticity restoration in deep networks. Combines attribution-guided neuron importance scoring with critical bottleneck period (CBP) detection and targeted reset. Use when: catastrophic forgetting, plasticity loss, neuron-level analysis, continual learning reset methods, attribution-guided training."
---
# Attribution-Based Neuron Utility for Plasticity Restoration
## Core Method
**GXD (Gradient-based eXplained Deviation)**: Attribution-based neuron utility metric that identifies which neurons contribute to prediction vs. which are dead.
## CBP Detection
Critical Bottleneck Periods (CBPs) are phases where gradient norms systematically decrease, indicating plasticity loss.
## Reset Strategy
1. Compute GXD scores per neuron
2. Identify low-utility (dead) neurons during CBPs
3. Selective reset of dead neurons while preserving functional ones
4. Outperforms L2 regularization, dropout, and L3 methods
## Paper
- Elisii et al., arXiv:2605.06834, 2026
Attribution
Related skills
View all in tools →Comments
Loading comments…
Newsletter
Get the weekly list of safe new skills, and the ones we pulled.