notesum.ai
Published at October 31An Information Criterion for Controlled Disentanglement of Multimodal Data
cs.LG
cs.AI
cs.IT
math.IT
Released Date: October 31, 2024
Authors: Chenyu Wang1, Sharut Gupta1, Xinyi Zhang1, Sana Tonekaboni2, Stefanie Jegelka3, Tommi Jaakkola4, Caroline Uhler1
Aff.: 1MIT, Broad Institute of MIT and Harvard; 2Broad Institute of MIT and Harvard; 3MIT, TU Munich; 4MIT

| Dataset | MIMIC | MOSEI | MOSI | UR-FUNNY | MUSTARD |
| CLIP | 64.97 (0.60) | 76.87 (0.45) | 64.24 (0.88) | 62.73 (0.92) | 56.04 (4.19) |
| FactorCL-emb | 65.25 (0.45) | 71.80 (0.64) | 62.97 (0.81) | 63.29 (2.07) | 56.76 (4.66) |
| FactorCL-proj | 59.43 (1.70) | 74.61 (1.65) | 56.02 (1.26) | 61.25 (0.47) | 55.80 (2.18) |
| FOCAL | 64.42 (0.34) | 76.77 (0.51) | 63.65 (1.09) | 62.98 (1.52) | 54.35 (0.00) |
| JointOpt | 66.11 (0.64) | 76.71 (0.14) | 64.24 (1.75) | 63.58 (1.45) | 56.52 (2.61) |
| DisentangledSSL (shared) | 63.16 (0.48) | 76.94 (0.22) | 65.16 (0.81) | 64.14 (1.53) | 54.11 (1.51) |
| DisentangledSSL (specific) | 65.73 (0.09) | 75.99 (0.60) | 51.70 (0.72) | 60.27 (1.28) | 61.60 (2.61) |
| DisentangledSSL (both) | 66.44 (0.31) | 77.45 (0.06) | 65.11 (0.80) | 64.24 (1.54) | 56.52 (2.18) |