Targeted Fine-Tuning of DNN-Based Receivers via Influence Functions

2025-09-19

AI Summary

This work addresses the suboptimal bit-prediction performance of deep neural network (DNN)-based wireless receivers under specific channel conditions. To this end, it introduces influence functions—previously applied primarily in classification and regression—to the domain of receiver fine-tuning for the first time. The proposed method comprises: (i) a sample importance analysis based on relative loss influence, and (ii) a second-order alignment update strategy. Integrated with a fully convolutional DeepRx architecture, capacity-constrained binary cross-entropy loss, and self-influence correction, it enables efficient, interpretable, sample-level targeted adaptation. Experiments demonstrate substantial BER reduction in single-target scenarios, achieving performance close to that of an ideal assisted receiver and significantly outperforming random fine-tuning. The key contribution lies in the novel extension of influence functions to communication signal processing, balancing computational efficiency and model interpretability. Generalization to multi-target settings remains an open challenge.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Search and Optimization: Learning to Search
• Natural Language Processing: Learning & Optimization for NLP

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

We present the first use of influence functions for deep learning-based wireless receivers. Applied to DeepRx, a fully convolutional receiver, influence analysis reveals which training samples drive bit predictions, enabling targeted fine-tuning of poorly performing cases. We show that loss-relative influence with capacity-like binary cross-entropy loss and first-order updates on beneficial samples most consistently improves bit error rate toward genie-aided performance, outperforming random fine-tuning in single-target scenarios. Multi-target adaptation proved less effective, underscoring open challenges. Beyond experiments, we connect influence to self-influence corrections and propose a second-order, influence-aligned update strategy. Our results establish influence functions as both an interpretability tool and a basis for efficient receiver adaptation.

Problem

• Targeted fine-tuning of DNN-based wireless receivers
• Identifying influential training samples via influence functions
• Improving bit error rate through selective sample updates

Innovation

• Influence functions for deep learning receivers
• Targeted fine-tuning using beneficial samples
• First-order updates with capacity-like loss