Research map / Architectures and efficiency / Attention and transformers
Error Correction Code Transformer
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add code-aware masked self-attention (ablation) | ECCT (N=6,d=64) · BCH(63,36) training loss | training loss: 76% loss reduction | Helped |
| add code-aware masked self-attention (ablation) | ECCT (N=6,d=64) · POLAR(64,32) training loss | training loss: 69% loss reduction | Helped |
| add code-aware masked self-attention (ablation) | ECCT (N=6,d=64) · LDPC(49,24) training loss | training loss: 66% loss reduction | Helped |
| replace BP-based neural decoder (AR BP, 50 iter) with ECCT N=6 d=128 | neural decoder · Polar(64,32) 6dB | -ln BER: 9.82 -> 12.32 | Helped |
| replace AR BP with ECCT N=10 d=128 | neural decoder · LDPC(121,70) 6dB | -ln BER: 14.77 -> 16.85 | Helped |
| use shallow ECCT N=2 d=32 instead of AR BP on LDPC(121,60) | neural decoder · LDPC(121,60) 6dB | -ln BER: 13.07 -> 8.06 | Hurt |
| increase ECCT size from N=2,d=32 to N=6,d=128 (ablation) | ECCT · BCH(63,51) 6dB | -ln BER: 9.03 -> 11.01 | Helped |
| increase ECCT depth to N=10 for LDPC codes (ablation) | ECCT · LDPC(49,24) 6dB | -ln BER: 11.40 -> 12.43 | Helped |
Closest papers by meaning
- Learning Linear Block Error Correction Codes (ICML 2024 · similarity 0.92)
- A Foundation Model for Error Correction Codes (ICLR 2024 · similarity 0.89)
- Drop-in Circulant Structural Priors for Transformer Decoding of Cyclic Codes (ICML 2026 · similarity 0.87)
- Denoising Diffusion Error Correction Codes (ICLR 2023 · similarity 0.86)
- Friendly Attacks to Improve Channel Coding Reliability (AAAI 2024 · similarity 0.84)
- CrossMPT: Cross-attention Message-passing Transformer for Error Correcting Codes (ICLR 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.