1317-GRP GenFEC: Generative Forward Error Correction for Natural-Language Packet Erasures
Location
https://campus.kennesaw.edu/colleges-departments/ccse/events/computing-showcase/fa26-cday-program.php
Document Type
Event
Start Date
20-11-2026 2:00 PM
Description
GenFEC uses transmitted token labels and language-model inference to recover missing tokens from lost packets while preserving received tokens. The labels narrow the possible values of missing text, and the model resolves the remaining ambiguity. We evaluate 60 WikiText-2 messages under three packet-loss channels and approximately matched transmitted overhead. Conventional references achieve stronger complete-message recovery. At 8 and 16 metadata bits per token, selected GenFEC configurations improve exact erased-token recovery (ETR) by 13.2 and 9.2 percentage points. Frequency-aware token labels improve recovery at unchanged bit width. These findings motivate further study alongside conventional forward error correction for delay-tolerant text or log recovery.
1317-GRP GenFEC: Generative Forward Error Correction for Natural-Language Packet Erasures
https://campus.kennesaw.edu/colleges-departments/ccse/events/computing-showcase/fa26-cday-program.php
GenFEC uses transmitted token labels and language-model inference to recover missing tokens from lost packets while preserving received tokens. The labels narrow the possible values of missing text, and the model resolves the remaining ambiguity. We evaluate 60 WikiText-2 messages under three packet-loss channels and approximately matched transmitted overhead. Conventional references achieve stronger complete-message recovery. At 8 and 16 metadata bits per token, selected GenFEC configurations improve exact erased-token recovery (ETR) by 13.2 and 9.2 percentage points. Frequency-aware token labels improve recovery at unchanged bit width. These findings motivate further study alongside conventional forward error correction for delay-tolerant text or log recovery.