![[2602.22277] X-REFINE: XAI-based RElevance input-Filtering and archItecture fiNe-tuning for channel Estimation](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.22277] X-REFINE: XAI-based RElevance input-Filtering and archItecture fiNe-tuning for channel Estimation
Summary
The paper presents X-REFINE, an XAI-based framework for optimizing channel estimation in 6G wireless communications by combining input filtering and architecture fine-tuning.
Why It Matters
As AI plays a crucial role in the development of 6G technology, understanding and improving the interpretability and efficiency of deep learning models is essential. X-REFINE addresses the limitations of existing models by enhancing their performance while reducing computational complexity, making it significant for future wireless communication systems.
Key Takeaways
- X-REFINE integrates input filtering with architecture fine-tuning for better channel estimation.
- The framework utilizes a decomposition-based, sign-stabilized LRP epsilon rule for deriving relevance scores.
- Simulation results show improved interpretability and performance with reduced computational complexity.
- X-REFINE maintains robust bit error rate (BER) performance across various scenarios.
- This research contributes to the advancement of AI-native architectures for 6G communications.
Computer Science > Machine Learning arXiv:2602.22277 (cs) [Submitted on 25 Feb 2026] Title:X-REFINE: XAI-based RElevance input-Filtering and archItecture fiNe-tuning for channel Estimation Authors:Abdul Karim Gizzini, Yahia Medjahdi View a PDF of the paper titled X-REFINE: XAI-based RElevance input-Filtering and archItecture fiNe-tuning for channel Estimation, by Abdul Karim Gizzini and 1 other authors View PDF HTML (experimental) Abstract:AI-native architectures are vital for 6G wireless communications. The black-box nature and high complexity of deep learning models employed in critical applications, such as channel estimation, limit their practical deployment. While perturbation-based XAI solutions offer input filtering, they often neglect internal structural optimization. We propose X-REFINE, an XAI-based framework for joint input-filtering and architecture fine-tuning. By utilizing a decomposition-based, sign-stabilized LRP epsilon rule, X-REFINE backpropagates predictions to derive high-resolution relevance scores for both subcarriers and hidden neurons. This enables a holistic optimization that identifies the most faithful model components. Simulation results demonstrate that X-REFINE achieves a superior interpretability-performance-complexity trade-off, significantly reducing computational complexity while maintaining robust bit error rate (BER) performance across different scenarios. Comments: Subjects: Machine Learning (cs.LG); Signal Processing (eess.SP) Cite as...