AI Paper Atlas

Research map / Vision-language and multimodal / CLIP and zero-shot

Do-Prompt: Causal Interventions Meet Variational Prompt Bottlenecks

Xueting Chen, Jun-Jie Huang, Yan Yan, Long Lan, Yuhua Tang, Wenjing Yang

ICML 2026 · Vision-language and multimodal › CLIP and zero-shot · approach: Prompt learning for CLIP

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 changedWhereMeasuredEffect
add variational prompt bottleneck (ablation)MaPLe* · Base-to-novel averageHM: 78.37 -> 78.93Improved
add do-intervention consistency (ablation)MaPLe* · Base-to-novel averageHM: 78.37 -> 79.02Improved
add both bottleneck and intervention to MaPLe (ablation)MaPLe* · Base-to-novel averageHM: 78.37 -> 80.12Improved
add variational prompt bottleneck (ablation)MMRL* · Base-to-novel averageHM: 80.59 -> 81.32Improved
add do-intervention consistency (ablation)MMRL* · Base-to-novel averageHM: 80.59 -> 81.43Improved
add both components to MMRL (ablation)MMRL* · Base-to-novel averageHM: 80.59 -> 82.56Improved
use all three intervention operators instead of mask only (ablation)MaPLe* + Do-Prompt · Base-to-novel averageHM: 79.01 -> 80.12Improved
use all three operators instead of swap only (ablation)MaPLe* + Do-Prompt · Base-to-novel averageHM: 78.94 -> 80.12Improved
core-environment split 2:3 instead of 3:2 (ablation)MMRL + Do-Prompt · Base-to-novel averageHM: 82.56 -> 81.27Hurt
core-environment split 4:1 instead of 3:2 (ablation)MaPLe + Do-Prompt · Base-to-novel averageHM: 80.12 -> 79.31Hurt
add Do-Prompt to MMRLMMRL* · EuroSAT base-to-novelHM: 85.50 -> 91.28Improved
add Do-Prompt to MMRLMMRL* · StanfordCars base-to-novelHM: 77.99 -> 78.17No clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most 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

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy