AI & softwareResearch findings
Multimodal Learning for Real-Time Software Defect Prediction in Autonomous Driving
Original published 2025.02.28 16:45 KSTAdded to Blackdeck 2026.11.10 21:39 KST
What has changed?
A multimodal transformer-based model that integrates text, numerical, and categorical data modalities for just-in-time defect prediction.
Key points
- The reliability of autonomous driving software is critically dependent on real-time defect prediction.
- A multimodal transformer model combining code, change metrics, and contextual data is proposed.
- Experiments on Apollo, Carla, and Donkeycar show significant performance improvement over state-of-the-art models.
Connect it to your work
The model uses cross-modal attention to enhance defect prediction accuracy in autonomous driving software.
What remains unproven
This research note is based on an abstract; check the original for test conditions and application scope.
Originals used for this note
Links open the original publisher in a new tab.