Research map / Vision-language and multimodal / Cross-modal learning
CAMP: Coherent Alignment of Multimodal Prototypes for Explainable Complementary Learning
Alvaro Lopez Pellicer, Eoin M. Kenny, Simran Lamba, Shubham Sharma, Plamen P Angelov, Saumitra Mishra
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add geometry loss (ablation) | Proto-Classic · Suite average (frozen encoders) | score: 0.7445 -> 0.7538 | Improved |
| add allocation loss (ablation) | Proto-Classic + geometry · Suite average (frozen encoders) | score: 0.7538 -> 0.7658 | Improved |
| add allocation loss (ablation) | Proto-Classic + geometry · Prototype health | dead prototypes: 26.6% -> 5.2% | Improved |
| add anchor loss (ablation) | Proto-Classic + geometry + allocation · Suite average (frozen encoders) | score: 0.7658 -> 0.7852 | Improved |
| add class-wise OT multimodal loss (ablation) | CAMP-F without OT · Suite average (frozen encoders) | score: 0.7852 -> 0.8139 | Improved |
| add class-wise OT multimodal loss (ablation) | CAMP-F · Cross-modal disagreement | Wε: 0.23 -> 0.12 | Improved |
| add all structural losses to Proto-Classic (ablation) | Proto-Classic · Suite average (frozen encoders) | score: 0.7445 -> 0.8139 | Improved |
| remove OT from full model (leave-one-out) (ablation) | CAMP-F · Suite average (frozen encoders) | score: -0.029 | Hurt |
| vary prototype capacity K (ablation) | CAMP · Suite average | score: varies by only ±0.001 | No clear effect |
| use gated fusion instead of concatenation or attention (ablation) | CAMP · 16 datasets | score: better on 15 of 16 datasets | Improved |
| use CAMP-F over cross-entropy MLP on same features | frozen Swin-L + DeBERTa-v3 · 16 benchmarks | MRR: 0.174 -> 0.448 | Improved |
| end-to-end fine-tuning (CAMP-E) instead of frozen head | CAMP · PTech | AUC: 0.744 -> 0.800 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Proto-Base — ProTo: Program-Guided Transformer for Program-Guided Tasks (NeurIPS 2021) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- With Limited Data for Multimodal Alignment, Let the STRUCTURE Guide You (NeurIPS 2025 · similarity 0.88)
- On the Value of Cross-Modal Misalignment in Multimodal Representation Learning (NeurIPS 2025 · similarity 0.87)
- TokenSwap: Benchmarking and Reducing the Modality Gap in Multimodal LLMs (NeurIPS 2026 · similarity 0.87)
- CovMatch: Cross-Covariance Guided Multimodal Dataset Distillation with Trainable Text Encoder (NeurIPS 2025 · similarity 0.87)
- When to Align, When to Predict: A Phase Diagram for Multimodal Learning (NeurIPS 2026 · similarity 0.87)
- A TRIANGLE Enables Multimodal Alignment Beyond Cosine Similarity (NeurIPS 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
