Recurrent graph network lowers QC-MDPC decoding failures in McEliece
Original title: Recurrent Bipartite Graph Neural Networks for QC-MDPC Error Decoding in McEliece Cryptosystems
On 20 unseen matrix topologies, R-BGNN cut the decoding failure rate 63.7% at noise level t=10.
The paper proposes a Recurrent Bipartite Graph Neural Network as a proof of concept for decoding QC-MDPC codes used in McEliece post-quantum cryptosystems. Classical belief propagation can stall on short cycles, and those failures can leak secret key information. Gated recurrent units and multi-topology curriculum training are used to break limit-cycle oscillations. The model beat Bit-Flipping and Min-Sum baselines and converged in fewer iterations.
Unconfirmed: The paper does not show a deployed break or measured key recovery from the lower failure rate.
- Evaluation used 20 matrix topologies the model had not seen.
- Decoding failure rate fell 63.7% at moderate noise t=10.
- Gated recurrent units sit inside the message-passing framework.
- The work is a proof of concept, not a fielded decoder.
From the original article
1 min read at arxiv.orgAs the advent of large-scale quantum computing threatens classical encryption, Quasi-Cyclic Moderate Density Parity-Check (QC-MDPC) codes have emerged as a foundational element of Post-Quantum Cryptography (PQC). However, decoding these relatively dense codes using classical Belief Propagation algorithms often leads to failures caused by short graph cycles that stall the decoder.
By Parth Rajeshkumar Tandel · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.