DS@GT ARC at ImageCLEFmedical 2026: Architectural Diversity for Concept Detection and Foundation-Model Scaling for Caption Prediction in Medical Image Analysis

We describe the DS@GT submissions to the ImageCLEFmedical Caption 2026 challenge, which continues a long-running benchmark on the ROCOv2 dataset with two tracks: Concept Detection (Task 1), assigning UMLS Concept Unique Identifiers (CUIs) to radiology images, and Caption Prediction (Task 2), generating natural-language captions. For Task 1, our primary submission was a three-way late-fusion ensemble of ConvNeXt-V2, BiomedCLIP ViT-B/16, and DenseNet-169 with a regularized ''Honest Threshold Tuning'' procedure designed to avoid validation overfitting on rare concepts; this submission ranked first on the official submission with a primary $F_1$ of $0.5790$ and a secondary $F_1$ of $0.9657$. In parallel, we submitted a training-free KNN retrieval pipeline over frozen BiomedCLIP embeddings, which reached a primary $F_1$ of $0.5780$ and a secondary $F_1$ of $0.9599$-essentially matching the fine-tuned ensemble on the primary track at a fraction of the cost. For Task 2, our submissions included a fine-tuned Gemma-3 27B model (overall $0.3571$, ranking third in the official submission), a fully fine-tuned BLIP pipeline with custom Vizwins merging ($0.3564$), and a zero-shot MedGemma-4B run with a PubMed-style prompt ($0.3186$), spanning a wide range of model scales and training costs. Code: https://github.com/dsgt-arc/imageclef-caption-2026.

Similarity Is Not Logic: Factored Inference for Dual-Encoder Vision-Language Models

Dual-encoder vision-language models (VLMs) expose a similarity interface that enables zero-shot retrieval but fails compositional constraints: queries like "umbrella and no person" retrieve images containing both, even when concept detection is reliable. We trace this to an interface-level Bag-of-Concepts effect, where similarity scores approximate mean pooling of concept evidence regardless of operators. Although operator-dependent signals exist in text embeddings, they are too weak or misaligned to affect rankings. Fine-tuning does not reliably resolve this failure because the dominant bottleneck is how similarity aggregates evidence rather than what encoders represent. We propose factored inference, which separates evidence extraction from constraint execution, and introduce LCSE (Logic-Constrained Score Editing), a training-free method that executes constraints externally using concept scores from frozen encoders. We also introduce FACTOR-Bench, where LCSE achieves 85.5% accuracy versus 73.2% for the best fine-tuned baseline, 90.7% when applied to SigLIP 2, and improves NegBench COCO MCQ accuracy from 27.2% to 65.2% while preserving retrieval performance.

Assessing Reliability of Symbol Detection in Concept Bottleneck Models

Concept Bottleneck Models (CBMs) are a relevant tool for explainable Artificial Intelligence because they make their predictions through human-interpretable symbols. However, high task accuracy does not guarantee that these symbols are detected faithfully: jointly trained CBMs may encode task-specific shortcuts in the bottleneck, making their explanations unreliable. In this paper, we study concept-detection reliability by swapping independently trained concept detectors and classification heads that share the same symbolic vocabulary. We use the resulting performance degradation, concept-level metrics, and symbol-wise uncertainty estimates to identify concepts that are especially prone to spurious firing. Finally, we propose a reliability-aware training strategy in which a shared concept detector is optimized with multiple classification heads and penalized for relying on globally or instance-wise unreliable symbols. On CUB-200-2011 with full concept supervision, detectors and heads are almost freely interchangeable (swap drop below one accuracy point, relative retention above $99\%$, and no concept detected below chance), whereas on a controlled synthetic task we show that, as the concept-supervision weight is reduced, models keep near-perfect task accuracy while swapped accuracy and agreement with the ground-truth concepts collapse to chance. Our reliability-aware training substantially mitigates this leakage, roughly doubling swap accuracy in the leaky regime.

The Signs Were Always There: Training-Free Concept Detection and Steering in Raw Transformer Dimensions

The standard basis of transformer hidden states is a training-free, architecture-general feature basis for detecting concepts and, in language models, steering them; with no learned dictionary. Individual dimensions act as binary registers read one at a time: their signs (+/-1) encode content, their magnitudes strength. A feature is just a subset of dimensions with a consistent sign pattern, read by counting sign agreements. We validate this Bag of Dims (BoD) framework across seven models spanning language, vision, and audio; reading dimensions one at a time loses nothing, as a full-capacity MLP adds zero AUC over per-dim reading. The same per-dimension signs appear in every modality, so they reflect transformer training itself, not the language objective. Sign alone carries predictive content: setting all magnitudes to unity preserves 60-93% top-5 next-token accuracy through the LM head. From a single-token cache (one forward pass per token, no labels) we detect 175 categories at AUC 0.97-0.99 by counting sign agreements, and from random seeds alone discovery scales to 1500 features per model. A trained probe adds only +0.018 AUC and converges to axis-aligned weights: the rotation dictionaries learn buys little. Signs are causally operative: they survive the attention projections, and flipping a concept's sign pattern in the live forward pass suppresses it. Reading and steering are separate roles in the same basis: a concept's reader dimensions are not its writer dimensions. The writer target is just as cheap, the sign of the summed unembedding rows over a few seeds, no training. Injected through the attention output pathway under closed-loop control, it steers concepts into fluent text on four language models (62-92% of twelve concepts). The signs were in the standard basis all along; the open problem is no longer finding the right rotation but cataloging what each dimension encodes.