Skip to content

Opening book details…

Can I read Bridging the Gap Between Multimodal Foundation Models and World Models on EtoBox?

Bridging the Gap Between Multimodal Foundation Models and World Models by Xuehai He is a book available to read on EtoBox.

What is Bridging the Gap Between Multimodal Foundation Models and World Models about?

List of Tables2.1 Parameter count in Adapter-tuning, LoRA, Compacter, and KAdaptation. L is the number of layers in the Transformer. k is the sizeof the input dimension to the Adapter layer. d is the bottleneckdimension in the Adapter layer. dmodel is the Transformer hiddensize. r denotes the rank in the low-rank decomposition step. n isthe number of Kronecker products usually very small. . . . . . . . 172.2 The averaged 5-shot experimental result comparison on 20 datasetsfrom ELEVATER benchmark [241] in terms of accuracy (%) andnumber of trainable parameters (#Params) across random seedsof {0, 1, 2}. The vision transformer (ViT-B-224/32) via CLIPpretraining is evaluated. Our method achieves the best tradeoffbetween accuracy and parameter efficiency: it obtains the best average accuracy among all efficient model adaptation methods, whileupdating only 0.09% of the model parameters in CLIP. We coloreach accuracy value as the best and second best: the same hereinafter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192.3 Experimental result comparison on CIFAR10 [223], CIFAR100 [223],SUN397 [486], DTD [70], STL10 [72], FGVCAircraft [308], andFER2013 [123] dataset

Author
Xuehai He
Published
2025
Language
EN

More by Xuehai He

Browse all works by Xuehai He