Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Reinforcement-Learned Unequal Error Protection for Quantized Semantic Embeddings

Moirangthem Tiken Singh, Adnan Arif

TL;DR

This work addresses preserving semantic meaning under bandwidth constraints by proposing a reinforcement-learning framework for per-dimension unequal error protection via discrete repetition coding. It introduces a composite semantic distortion $D_S$ that balances global embedding similarity and entity-level correctness, and uses an A2C policy to allocate repetition counts under a fixed budget, enabling fine-grained protection not possible with traditional block codes. Empirical results on AG News show significant gains in chrF and entity preservation at low SNRs, with robust transfer to coarser quantization and resilience to channel mismatch. The findings highlight that code structure should align with semantic granularity, supporting edge- and IoT-oriented semantic networks while opening avenues for joint embedder-policy optimization and multimodal extensions.

Abstract

This paper tackles the pressing challenge of preserving semantic meaning in communication systems constrained by limited bandwidth. We introduce a novel reinforcement learning framework that achieves per-dimension unequal error protection via adaptive repetition coding. Central to our approach is a composite semantic distortion metric that balances global embedding similarity with entity-level preservation, empowering the reinforcement learning agent to allocate protection in a context-aware manner. Experiments show statistically significant gains over uniform protection, achieving 6.8% higher chrF scores and 9.3% better entity preservation at 1 dB SNR. The key innovation of our framework is the demonstration that simple, intelligently allocated repetition coding enables fine-grained semantic protection -- an advantage unattainable with conventional codes such as LDPC or Reed-Solomon. Our findings challenge traditional channel coding paradigms by establishing that code structure must align with semantic granularity. This approach is particularly suited to edge computing and IoT scenarios, where bandwidth is scarce, but semantic fidelity is critical, providing a practical pathway for next-generation semantic-aware networks.

Reinforcement-Learned Unequal Error Protection for Quantized Semantic Embeddings

TL;DR

This work addresses preserving semantic meaning under bandwidth constraints by proposing a reinforcement-learning framework for per-dimension unequal error protection via discrete repetition coding. It introduces a composite semantic distortion that balances global embedding similarity and entity-level correctness, and uses an A2C policy to allocate repetition counts under a fixed budget, enabling fine-grained protection not possible with traditional block codes. Empirical results on AG News show significant gains in chrF and entity preservation at low SNRs, with robust transfer to coarser quantization and resilience to channel mismatch. The findings highlight that code structure should align with semantic granularity, supporting edge- and IoT-oriented semantic networks while opening avenues for joint embedder-policy optimization and multimodal extensions.

Abstract

This paper tackles the pressing challenge of preserving semantic meaning in communication systems constrained by limited bandwidth. We introduce a novel reinforcement learning framework that achieves per-dimension unequal error protection via adaptive repetition coding. Central to our approach is a composite semantic distortion metric that balances global embedding similarity with entity-level preservation, empowering the reinforcement learning agent to allocate protection in a context-aware manner. Experiments show statistically significant gains over uniform protection, achieving 6.8% higher chrF scores and 9.3% better entity preservation at 1 dB SNR. The key innovation of our framework is the demonstration that simple, intelligently allocated repetition coding enables fine-grained semantic protection -- an advantage unattainable with conventional codes such as LDPC or Reed-Solomon. Our findings challenge traditional channel coding paradigms by establishing that code structure must align with semantic granularity. This approach is particularly suited to edge computing and IoT scenarios, where bandwidth is scarce, but semantic fidelity is critical, providing a practical pathway for next-generation semantic-aware networks.
Paper Structure (15 sections, 1 theorem, 13 equations, 2 figures, 8 tables)

This paper contains 15 sections, 1 theorem, 13 equations, 2 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Results and Analysis
  5. Experimental Setup
  6. Evaluation Metrics
  7. Training Convergence and Stability
  8. Ablation: Distortion Balancing ($\alpha$)
  9. Transferability Across Quantization Regimes
  10. Comparison with Alternative Allocation Strategies
  11. Interaction with Error-Correction Coding
  12. Robustness under Channel Mismatch
  13. Benchmarking Against State-of-the-Art
  14. Discussion
  15. Conclusion

Key Result

Theorem 1

Assume: (i) rewards are uniformly bounded, (ii) policy and value approximators are Lipschitz-continuous, (iii) gradient estimators have bounded second moments, (iv) step sizes $\{\alpha_k\}$ (critic) and $\{\gamma_k\}$ (actor) satisfy Robbins–Monro conditions with $\sum_k\alpha_k=\sum_k\gamma_k=\inf

Figures (2)

  • Figure 1: End-to-end semantic communication pipeline. The policy $\pi_\theta$ assigns per-dimension repetition counts $\mathbf{t}$ to quantized embeddings, optimized to minimize semantic distortion $D_S$ under a fixed channel-use budget $B$.
  • Figure 2: Performance comparison across $\alpha$ values under matched AWGN conditions. The balanced objective ($\alpha=0.5$, cyan) consistently outperforms both entity-only ($\alpha=0.0$, purple) and embedding-only ($\alpha=1.0$, yellow) objectives in the critical 1--2 dB SNR regime.

Theorems & Definitions (2)

  • Theorem 1: Convergence of Actor–Critic Policy
  • proof : Sketch