{"slug": "structured-token-attention-for-fine-grained-mmwave-liquid-identification", "title": "Structured-Token Attention for Fine-Grained mmWave Liquid Identification", "summary": "A new arXiv paper (2610.27512v1) reports that the StructAttn model, which encodes radar feature tensors as eleven structured tokens across antenna, feature, branch and global views and processes them with a three-layer transformer, reached 96.70% mean best-epoch accuracy over five seeds on the public FG-LiquID dataset covering 30 liquids and 3,000 samples. StructAttn matched the original-recipe MMNet baseline and beat a scalar-token attention model by 1.50 percentage points, while a five-seed probability ensemble hit 97%, or 18 errors among 600 test samples, and StructAttn recorded 95.30% out-of-fold accuracy in five-fold testing. Residual confusions concentrated in the water, soft-drink and edible-oil classes, and the authors conclude structured tokenization is a practical design choice for attention-based liquid recognition.", "body_md": "arXiv:2610.27512v1 Announce Type: new \nAbstract: Fine-grained liquid identification requires a classifier to distinguish similar substances from compact millimeter-wave measurements. We investigate structured-token attention, which represents a radar feature tensor through complementary antenna, feature, branch, and global views. The proposed StructAttn model embeds these views into eleven tokens and combines them with a three-layer transformer. Experiments on the public FG-LiquID data cover 30 liquids and 3,000 samples. Over five seeds on the released split, StructAttn achieves 96.70\\% mean best-epoch accuracy, matching the original-recipe MMNet baseline and exceeding a scalar-token attention model by 1.50 percentage points. A five-seed probability ensemble reaches 97\\%, corresponding to 18 errors among 600 test samples. Final-epoch and five-fold results complement the checkpoint-selected comparison, with StructAttn obtaining 95.30\\% out-of-fold accuracy. Token and feature analyses show that structured views are useful for prediction, while residual confusions concentrate in water, soft-drink, and edible-oil classes. These results support structured tokenization as a practical design choice for attention-based liquid recognition.", "url": "https://wpnews.pro/news/structured-token-attention-for-fine-grained-mmwave-liquid-identification", "canonical_source": "https://www.machinebrief.com/news/structured-token-attention-for-fine-grained-mmwave-liquid-id-50n3", "published_at": "2026-10-21 04:00:00+00:00", "updated_at": "2026-10-21 05:50:06.838308+00:00", "lang": "en", "topics": ["machine-learning", "ai-research", "neural-networks"], "entities": ["StructAttn", "FG-LiquID", "MMNet", "arXiv"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/structured-token-attention-for-fine-grained-mmwave-liquid-identification", "markdown": "https://wpnews.pro/news/structured-token-attention-for-fine-grained-mmwave-liquid-identification.md", "text": "https://wpnews.pro/news/structured-token-attention-for-fine-grained-mmwave-liquid-identification.txt", "jsonld": "https://wpnews.pro/news/structured-token-attention-for-fine-grained-mmwave-liquid-identification.jsonld"}}