Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Zero-Forget Preservation of Semantic Communication Alignment in Distributed AI Networks

Jingzhi Hu, Geoffrey Ye Li

TL;DR

This paper tackles the problem of semantic communication alignment between distributed AIs when each node undergoes local domain adaptation. It introduces zero-forget domain adaptation (ZFDA), which preserves SC alignment by applying sparse additive modifications (SAM) to pre-trained encoder/decoder parameters, decoupled into binary masks and continuous deltas and optimized via a score-based, straight-through estimator approach alongside layer-wise sparsity regularization. The method reformulates the alignment problem from an intractable joint adaptation task into tractable SAM optimization and demonstrates, on a CIFAR-100 autoencoder for image transmission, that SC alignment can be retained with zero PSNR loss and memory overhead under 1% of the model size, even as domains vary (VA, VP, VC, VH). The results indicate that SAM-enabled ZFDA can support scalable, zero-forget semantic networking across distributed AI systems with diverse local preferences, offering a practical path to robust SC in large-scale AI ecosystems.

Abstract

Future communication networks are expected to connect massive distributed artificial intelligence (AI). Exploiting aligned priori knowledge of AI pairs, it is promising to convert high-dimensional data transmission into highly-compressed semantic communications (SC). However, to accommodate the local data distribution and user preferences, AIs generally adapt to different domains, which fundamentally distorts the SC alignment. In this paper, we propose a zero-forget domain adaptation (ZFDA) framework to preserve SC alignment. To prevent the DA from changing substantial neural parameters of AI, we design sparse additive modifications (SAM) to the parameters, which can be efficiently stored and switched-off to restore the SC alignment. To optimize the SAM, we decouple it into tractable continuous variables and a binary mask, and then handle the binary mask by a score-based optimization. Experimental evaluations on a SC system for image transmissions validate that the proposed framework perfectly preserves the SC alignment with almost no loss of DA performance, even improved in some cases, at a cost of less than 1% of additional memory.

Zero-Forget Preservation of Semantic Communication Alignment in Distributed AI Networks

TL;DR

This paper tackles the problem of semantic communication alignment between distributed AIs when each node undergoes local domain adaptation. It introduces zero-forget domain adaptation (ZFDA), which preserves SC alignment by applying sparse additive modifications (SAM) to pre-trained encoder/decoder parameters, decoupled into binary masks and continuous deltas and optimized via a score-based, straight-through estimator approach alongside layer-wise sparsity regularization. The method reformulates the alignment problem from an intractable joint adaptation task into tractable SAM optimization and demonstrates, on a CIFAR-100 autoencoder for image transmission, that SC alignment can be retained with zero PSNR loss and memory overhead under 1% of the model size, even as domains vary (VA, VP, VC, VH). The results indicate that SAM-enabled ZFDA can support scalable, zero-forget semantic networking across distributed AI systems with diverse local preferences, offering a practical path to robust SC in large-scale AI ecosystems.

Abstract

Future communication networks are expected to connect massive distributed artificial intelligence (AI). Exploiting aligned priori knowledge of AI pairs, it is promising to convert high-dimensional data transmission into highly-compressed semantic communications (SC). However, to accommodate the local data distribution and user preferences, AIs generally adapt to different domains, which fundamentally distorts the SC alignment. In this paper, we propose a zero-forget domain adaptation (ZFDA) framework to preserve SC alignment. To prevent the DA from changing substantial neural parameters of AI, we design sparse additive modifications (SAM) to the parameters, which can be efficiently stored and switched-off to restore the SC alignment. To optimize the SAM, we decouple it into tractable continuous variables and a binary mask, and then handle the binary mask by a score-based optimization. Experimental evaluations on a SC system for image transmissions validate that the proposed framework perfectly preserves the SC alignment with almost no loss of DA performance, even improved in some cases, at a cost of less than 1% of additional memory.

Paper Structure

This paper contains 10 sections, 16 equations, 5 figures, 1 algorithm.

Table of Contents

  1. Introduction
  2. System Model
  3. ZFDA Framework to Preserve SC Alignment
  4. SAM Optimization Algorithm
  5. Sparsity-Aware SAM Decomposition
  6. Score-Aided SAM Optimization
  7. Sparsity Distribution Regularization
  8. Experimental Evaluations
  9. Conclusion
  10. Proof of Proposition \ref{['thm_loss_decrease']}

Figures (5)

  • Figure 1: SC alignment between Tx and Rx AIs using pre-trained parameters (left); Distorted SC alignment due to respective DA of AIs (right).
  • Figure 2: Samples from the original domain and the other varied local domains.
  • Figure 3: PSNR for the SC given Tx AI (left) or Rx AI (right) adapted to each varied domain with different alignment methods being employed.
  • Figure 4: PSNR for the four local domains versus the sparsity rate of the SAM, given different cases of sparsity distribution and mask optimization.
  • Figure : SAM Optimization Algorithm