{"version":1,"type":"story","url":"https://digestai.news/story/lora-boosts-sas-vision-transformer-auprc-to-0-679-in-underwater-target","json":"https://digestai.news/story/lora-boosts-sas-vision-transformer-auprc-to-0-679-in-underwater-target.json","markdown":"https://digestai.news/story/lora-boosts-sas-vision-transformer-auprc-to-0-679-in-underwater-target.md","slug":"lora-boosts-sas-vision-transformer-auprc-to-0-679-in-underwater-target","headline":"LoRA boosts SAS vision transformer AUPRC to 0.679 in underwater target recognition","summary":"Researchers adapted a DINOv3 Vision Transformer (ViT) for synthetic aperture sonar (SAS) automatic target recognition, a task limited by scarce imagery and noisy acoustic backgrounds. Their three‑stage, parameter‑efficient framework first applies Low‑Rank Adaptation (LoRA) while keeping the ViT backbone frozen, then adds hard‑negative mining, and finally uses supervised contrastive learning (SupCon).\\n\\nEvaluations on at‑sea SAS data with a geographic split showed that LoRA alone raised the area under the precision‑recall curve (AUPRC) from 0.300 to 0.679 ± 0.027, training only 0.26 % of the model’s weights (Rank 4). The subsequent hard‑negative mining stage altered AUPRC by –0.0045 ± 0.0119 compared with a random curriculum, and SupCon changed it by +0.0002 ± 0.0096 versus the preceding stage. These near‑zero effects suggest that a single efficient adaptation step captures most of the target‑clutter geometry, making additional refinement unnecessary.","keyPoints":["LoRA adaptation raised AUPRC from 0.300 to 0.679 ± 0.027 using a frozen ViT backbone.","Only 0.26 % of model weights (Rank 4) were trained during LoRA adaptation.","Hard‑negative mining and SupCon produced negligible AUPRC changes (‑0.0045 ± 0.0119 and +0.0002 ± 0.0096)."],"whyItMatters":"Demonstrates that lightweight fine‑tuning can dramatically improve sonar target detection, enabling naval AI systems to work with minimal data and compute.","category":{"slug":"research","name":"Research","url":"https://digestai.news/category/research"},"entities":{"companies":[],"models":["DINOv3 Vision Transformer"],"people":[]},"firstPublishedAt":"2026-09-21T04:00:00Z","updatedAt":"2026-09-21T04:00:00Z","sourceCount":1,"hasPrimarySource":true,"sources":[{"outlet":"arXiv cs.AI","title":"LoRA Enhanced Contrastive Learning with SAS Vision Transformers","url":"https://arxiv.org/abs/2609.21061","publishedAt":"2026-09-21T04:00:00Z","type":"primary","primary":true,"lead":true}],"sourceNotes":null,"discussions":[],"thread":null,"cite":{"text":"Digest AI, \"LoRA boosts SAS vision transformer AUPRC to 0.679 in underwater target recognition\", 21 September 2026, https://digestai.news/story/lora-boosts-sas-vision-transformer-auprc-to-0-679-in-underwater-target","publisher":"Digest AI","title":"LoRA boosts SAS vision transformer AUPRC to 0.679 in underwater target recognition","datePublished":"2026-09-21T04:00:00Z","url":"https://digestai.news/story/lora-boosts-sas-vision-transformer-auprc-to-0-679-in-underwater-target"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse"}