Multi-head attention debiasing and contrastive learning for mitigating Dataset Artifacts in Natural Language Inference

While Natural Language Inference (NLI) models have achieved high performances on benchmark datasets, there are still concerns whether they truly capture the intended task, or largely exploit dataset artifacts. Through detailed analysis of the Stanford Natural Language Inference (SNLI) dataset, we have uncovered complex patterns of various types of artifacts and their interactions, leading to the development of our novel structural debiasing approach. Our fine-grained analysis of 9,782 validation examples reveals four major categories of artifacts: length-based patterns, lexical overlap, subset relationships, and negation patterns. Our multi-head debiasing architecture achieves substantial improvements across all bias categories: length bias accuracy improved from 86.03% to 90.06%, overlap bias from 91.88% to 93.13%, subset bias from 95.43% to 96.49%, and negation bias from 88.69% to 94.64%. Overall, our approach reduces the error rate from 14.19% to 10.42% while maintaining high performance on unbiased examples. Analysis of 1,026 error cases shows significant improvement in handling neutral relationships, traditionally one of the most challenging areas for NLI systems.

Vehicle Detection and Classification for Toll collection using YOLOv11 and Ensemble OCR

Traditional automated toll collection systems depend on complex hardware configurations, that require huge investments in installation and maintenance. This research paper presents an innovative approach to revolutionize automated toll collection by using a single camera per plaza with the YOLOv11 computer vision architecture combined with an ensemble OCR technique. Our system has achieved a Mean Average Precision (mAP) of 0.895 over a wide range of conditions, demonstrating 98.5% accuracy in license plate recognition, 94.2% accuracy in axle detection, and 99.7% OCR confidence scoring. The architecture incorporates intelligent vehicle tracking across IOU regions, automatic axle counting by way of spatial wheel detection patterns, and real-time monitoring through an extended dashboard interface. Extensive training using 2,500 images under various environmental conditions, our solution shows improved performance while drastically reducing hardware resources compared to conventional systems. This research contributes toward intelligent transportation systems by introducing a scalable, precision-centric solution that improves operational efficiency and user experience in modern toll collections.