Most research studies on deep learning (DL) applied to the physical layer of wireless communication do not put forward the critical role of the accuracy-generalization trade-off in developing and evaluating practical algorithms. To highlight the disadvantage of this common practice, we revisit a data decoding example from one of the first papers introducing DL-based end-to-end wireless communication systems to the research community and promoting the use of artificial intelligence (AI)/DL for the wireless physical layer. We then put forward two key trade-offs in designing DL models for communication, namely, accuracy versus generalization and compression versus latency. We discuss their relevance in the context of wireless communications use cases using emerging DL models including large language models (LLMs). Finally, we summarize our proposed evaluation guidelines to enhance the research impact of DL on wireless communications. These guidelines are an attempt to reconcile the empirical nature of DL research with the rigorous requirement metrics of wireless communications systems.

翻译：大多数关于深度学习应用于无线通信物理层的研究，并未突出准确性与泛化性权衡在开发和评估实用算法中的关键作用。为揭示这一常见做法的弊端，我们重新审视了最早向学界介绍基于深度学习的端到端无线通信系统、并推动人工智能/深度学习用于无线物理层的论文中的一个数据解码实例。随后，我们提出设计通信用深度学习模型时的两个关键权衡：准确性与泛化性、压缩与延迟。我们结合新兴的深度学习模型（包括大型语言模型）来讨论它们与无线通信用例的相关性。最后，我们总结所提出的评估指导原则，以增强深度学习对无线通信的研究影响力。这些原则旨在调和深度学习研究的实证性本质与无线通信系统的严格指标要求之间的矛盾。