Research map / Architectures and efficiency / RNNs and normalization
Communication Algorithms via Deep Learning
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 |
|---|---|---|---|
| Use two bidirectional GRU layers instead of one (ablation) | N-RSC · Rate-1/2 RSC, block length 100, 4dB | BER: 0.01400 -> 0.00128 | Helped |
| Use bidirectional GRU layers instead of unidirectional GRU layers (ablation) | N-RSC · Rate-1/2 RSC, block length 100, 4dB | BER: 0.01787 -> 0.00128 | Helped |
| Use two bidirectional GRU layers instead of two bidirectional vanilla RNN layers (ablation) | N-RSC · Rate-1/2 RSC, block length 100, 4dB | BER: 0.05814 -> 0.00128 | Helped |
| Increase bidirectional GRU depth from two layers to three (ablation) | N-RSC · Rate-1/2 RSC, block length 100, 4dB | BER: 0.00128 -> 0.00127 | No clear effect |
| Increase bidirectional GRU depth from two layers to four (ablation) | N-RSC · Rate-1/2 RSC, block length 100, 4dB | BER: 0.00128 -> 0.00128 | No clear effect |
| Increase bidirectional GRU depth from two layers to five (ablation) | N-RSC · Rate-1/2 RSC, block length 100, 4dB | BER: 0.00128 -> 0.00132 | No clear effect |
| Use two bidirectional GRU layers instead of one bidirectional LSTM layer (ablation) | N-RSC · Rate-1/2 RSC, block length 100, 4dB | BER: 0.01376 -> 0.00128 | Helped |
| Use two bidirectional GRU layers instead of one bidirectional GRU layer (ablation) | N-BCJR · BCJR-like RNN | BCJR Val MSE: 0.0035 -> 1.1532e-04 | Helped |
| Use bidirectional GRU layers instead of unidirectional GRU layers (ablation) | N-BCJR · BCJR-like RNN | BCJR Val MSE: 0.0026 -> 1.1532e-04 | Helped |
| Use two bidirectional GRU layers instead of two bidirectional vanilla RNN layers (ablation) | N-BCJR · BCJR-like RNN | BCJR Val MSE: 0.0010 -> 1.1532e-04 | Helped |
| Use N-RSC on AWGN instead of the MAP decoder | N-RSC · Rate-1/2 RSC on AWGN, BER across test SNRs and block lengths | BER | No clear effect |
| Apply the length-100-trained N-RSC to length-10,000 codes | N-RSC · Rate-1/2 RSC on AWGN, block length 10,000 | BER | No clear effect |
Closest papers by meaning
- KO codes: inventing nonlinear encoding and decoding for reliable wireless communication via deep-learning (ICML 2021 · similarity 0.90)
- DeepPolar: Inventing Nonlinear Large-Kernel Polar Codes via Deep Learning (ICML 2024 · similarity 0.85)
- Learning Linear Block Error Correction Codes (ICML 2024 · similarity 0.85)
- AdaDecode: Accelerating LLM Decoding with Adaptive Layer Parallelism (ICML 2025 · similarity 0.84)
- Exploiting Symmetric Temporally Sparse BPTT for Efficient RNN Training (AAAI 2024 · similarity 0.84)
- Benchmarking Predictive Coding Networks -- Made Simple (ICLR 2025 · similarity 0.83)
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.