Research map / Vision-language and multimodal / CLIP and zero-shot
VITA: Zero-Shot Value Functions via Test-Time Adaptation of Vision–Language Models
Christos Ziakas, Alessandra Russo
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 test-time adaptation with sequential updates to CLIP-based value function | VITA · ms_sweep (long-horizon) | VOC: 0.434 -> 0.490 | Improved |
| apply sequential per-step adaptation instead of trajectory-level update (ablation) | VITA · tk_pnp validation | VOC: 0.1917 -> 0.7822 | Improved |
| sequential per-step adaptation instead of memoryless reset each step (ablation) | VITA · tk_pnp validation | VOC: 0.1918 -> 0.7822 | Improved |
| sequential adaptation instead of explicit window memory (ablation) | VITA · tk_pnp validation | VOC: 0.1957 -> 0.7822 | Improved |
| use dissimilarity-based sampling instead of full-trajectory sampling (ablation) | VITA · Scripted datasets | Disc@VOC: 0.20 -> 1.0 | Improved |
| use dissimilarity-based sampling instead of random sampling (ablation) | VITA · Scripted datasets | Disc@VOC: 0.8 -> 1.0 | Improved |
| reduce sampled windows from k=8 to k=4 (ablation) | VITA · Scripted datasets | Disc@VOC: 1.0 -> 1.0 | No clear effect |
| reduce sub-trajectory length from 8 to 4 (ablation) | VITA · Scripted datasets | Disc@VOC: 1.0 -> 0.60 | Hurt |
| increase test-time steps from 1 to 10 at lr 0.1 (ablation) | VITA · Validation set | VOC: 0.783 -> 0.369 | Hurt |
| increase test-time steps from 1 to 10 at lr 0.01 (ablation) | VITA · Validation set | VOC: 0.454 -> 0.781 | Improved |
| use VITA instead of GVL autoregressive VLM | VITA · tk_pnp (in-distribution) | VOC: 0.269 -> 0.782 | Improved |
| use VITA instead of CLIP-GRU recurrent memory | VITA · 10 evaluation datasets VOC | VOC: 6 of 10 datasets better | Mixed |
Closest papers by meaning
- Vision Language Models are In-Context Value Learners (ICLR 2025 · similarity 0.90)
- Test-Time Adaptation with CLIP Reward for Zero-Shot Generalization in Vision-Language Models (ICLR 2024 · similarity 0.90)
- Vision-Language Models are Zero-Shot Reward Models for Reinforcement Learning (ICLR 2024 · similarity 0.89)
- PriorVLA: Prior-Preserving Adaptation for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.88)
- NS-VLA: Towards Neuro-Symbolic Vision-Language-Action Models (NeurIPS 2026 · similarity 0.88)
- Exploring the Limits of Vision-Language-Action Manipulation in Cross-task Generalization (NeurIPS 2025 · similarity 0.87)
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.
