Research map / Vision-language and multimodal / Cross-modal learning
Training Transitive and Commutative Multimodal Transformers with LoReTTa
Manuel Tran, Yashin Dicente Cid, Amal Lahiani, Fabian J Theis, Tingying Peng, Eldad Klaiman
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 transitive modeling (LoReTTa) on top of C2M3, pretrained on (T,I),(T,A) | C2M3 · SVL-MNIST unseen pair (I,A) | perplexity: 104.27 -> 5.04 | Improved |
| add transitive modeling, pretrained on (A,I),(A,T) | C2M3 · SVL-MNIST unseen pair (I,T) | perplexity: 9.70 -> 3.13 | Improved |
| add transitive modeling, pretrained on (I,T),(I,A) | C2M3 · SVL-MNIST unseen pair (T,A) | perplexity: 30.88 -> 11.21 | Improved |
| add transitive modeling, pretrained on (T,I),(T,A) | C2M3 · SVL-MNIST probe accuracy, unseen (I,A) | accuracy: 72.9 -> 80.9 | Improved |
| add transitive modeling, pretrained on (I,T),(I,A) | C2M3 · SVL-MNIST probe accuracy, unseen (T,A) | accuracy: 71.2 -> 84.0 | Improved |
| add transitive modeling, pretrained on (A,I),(A,T) | C2M3 · SVL-MNIST probe accuracy, unseen (I,T,A) | accuracy: 89.9 -> 91.6 | Improved |
| compare LoReTTa with full trimodal C2M3 upper baseline | LoReTTa · SVL-MNIST probe accuracy (I,T,A) | accuracy: 88.5 -> 91.6 | Improved |
| add commutative bimodal pretraining (C2M3 (I,T) vs CM3 (I)) | CM3 · SVL-MNIST probe accuracy, modality I | accuracy: 79.7 -> 82.9 | Improved |
| add transitive modeling to BERT (T-BERT) | BERT · TCGA-OMICS unseen (I,R) | c-index: 0.594 -> 0.623 | Mixed |
| add transitive modeling to GPT (T-GPT) | GPT · TCGA-OMICS unseen (M,I,R) | c-index: 0.588 -> 0.622 | Improved |
| apply LoReTTa instead of C2M3 | C2M3 · TCGA-OMICS unseen (I,R) | c-index: 0.571 -> 0.623 | Improved |
| apply LoReTTa instead of C2M3 | C2M3 · TCGA-OMICS modality R | c-index: 0.599 -> 0.563 | Hurt |
Closest papers by meaning
- Multimodal Pathway: Improve Transformers with Irrelevant Data from Other Modalities (CVPR 2024 · similarity 0.87)
- MSP: Modality Self-Play for Training Multimodal LLMs Without Paired Cross-Modal Data (NeurIPS 2026 · similarity 0.87)
- UNIALIGN: Scaling Multimodal Alignment within One Unified Model (CVPR 2025 · similarity 0.86)
- The Design Space of Tri-Modal Masked Diffusion Models (NeurIPS 2026 · similarity 0.86)
- A TRIANGLE Enables Multimodal Alignment Beyond Cosine Similarity (NeurIPS 2025 · similarity 0.86)
- Dissecting Multimodality in VideoQA Transformer Models by Impairing Modality Fusion (ICML 2024 · 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.
