Characterizing the Predictive Impact of Modalities with Supervised Latent-Variable Modeling
Divyam Madaan, Sumit Chopra, Kyunghyun Cho · 2026
Read Characterizing the Predictive Impact of Modalities with Supervised Latent-Variable Modeling by Divyam Madaan, Sumit Chopra, Kyunghyun Cho (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-16979.txt.
Description
Divyam Madaan, Sumit Chopra, and Kyunghyun Cho propose PRIMO, a supervised latent-variable model gauging the predictive impact of modes. Despite recent success of multimodal large language models, existing approaches predominantly assume multiple modes are on hand during training and inference. In practice, multimodal data is often incomplete because modes may be missing, collected asynchronously, or available only for a subset of examples.
PRIMO quantifies the predictive impact of any missing modality within the multimodal learning setting. It enables use of all available training examples, whether modalities are complete or partial. It models the missing mode through a latent variable capturing its relationship with the observed mode in the context of prediction. Tests on a synthetic XOR dataset, Audio-Vision MNIST, and MIMIC-III for mortality and ICD-9 prediction show scores comparable to unimodal baselines when a mode is missing.
Catalog record
| CREATOR | Divyam Madaan, Sumit Chopra, Kyunghyun Cho |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 51 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-16979