mlsanityData to 14 Oct 2026

Subspace Alignment for Vision-Language Model Test-time Adaptation

arXiv 2601.08139 ↗First posted 13 Jan 2026cs.CV cs.AIcited by 5

On the map: Multimodal LLMs / vision-language models → Prompt learning and adaptation of CLIP

Abstract (arXiv)

Vision-language models (VLMs), despite their extraordinary zero-shot capabilities, are vulnerable to distribution shifts. Test-time adaptation (TTA) emerges as a predominant strategy to adapt VLMs to unlabeled test data on the fly. However, existing TTA methods heavily rely on zero-shot predictions as pseudo-labels for self-training, which can be unreliable under distribution shifts and misguide adaptation due to two fundamental limitations.First (Modality Gap), distribution shifts induce gaps between visual and textual modalities, making cross-modal relations inaccurate. Second (Visual Nuisance), visual embeddings encode rich but task-irrelevant noise that often overwhelms task-specific semantics under distribution shifts. To address these limitations, we propose SubTTA, which aligns the semantic subspaces of both modalities to enhance zero-shot predictions to better guide the TTA process. To bridge the modality gap, SubTTA extracts the principal subspaces of both modalities and aligns the visual manifold to the textual semantic anchor by minimizing their chordal distance. To eliminate visual nuisance, SubTTA projects the aligned visual features onto the task-specific textual subspace, which filters out task-irrelevant noise by constraining visual embeddings within the valid semantic span, and standard TTA is further performed on the purified space to refine the decision boundaries. Extensive experiments on various benchmarks and VLM architectures demonstrate the effectiveness of SubTTA, yielding an average improvement of 2.24% over state-of-the-art TTA methods. Our code is available at https://github.com/zhichenz98/SubTTA_EMNLP26.

Key facts. Cited by 5 papers in the map (data through 14 Oct 2026). Placed in Prompt learning and adaptation of CLIP, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 27 of its 106 references that are in the map. First posted on arXiv 13 Jan 2026.

Builds on

27 of 106 references.

Closest ancestors · 2

Same direction, newest first

  1. 2025Mint: A Simple Test-Time Adaptation of Vision-Language Models against Common Corruptions25 Oct
  2. 2025Complementary Subspace Low-Rank Adaptation of Vision-Language Models for Few-Shot Classification25 Jan

Other references · 25

  1. 2023GPT-4 Technical Reportcited by 18,396
  2. 2023LLaMA: Open and Efficient Foundation Language Modelscited by 14,132
  3. 2019Benchmarking Neural Network Robustness to Common Corruptions and Perturbationscited by 2,877
  4. 2022CoCa: Contrastive Captioners are Image-Text Foundation Modelscited by 1,274
  5. 2022PaLI: A Jointly-Scaled Multilingual Language-Image Modelcited by 766
  6. 2020Tent: Fully Test-time Adaptation by Entropy Minimizationcited by 69
  7. 2025Improving LLM General Preference Alignment via Optimistic Online Mirror Descentcited by 24
  8. 2025Quantization Meets dLLMs: A Systematic Study of Post-training Quantization for Diffusion LLMscited by 23
  9. 2025Visual question answering: from early developments to recent advances -- a surveycited by 19
  10. 2025Pave Your Own Path: Graph Gradual Domain Adaptation on Fused Gromov-Wasserstein Geodesicscited by 17
  11. 2025ResMoE: Space-efficient Compression of Mixture of Experts LLMs via Residual Restorationcited by 12
  12. 2014Subspace Alignment For Domain Adaptationcited by 10
13 more
  1. 2024Matcha: Mitigating Graph Structure Shifts with Test-Time Adaptationcited by 10
  2. 2024A Collaborative Ensemble Framework for CTR Predictioncited by 10
  3. 2025MORALISE: A Structured Benchmark for Moral Alignment in Visual Language Modelscited by 10
  4. 2025PLANETALIGN: A Comprehensive Python Library for Benchmarking Network Alignmentcited by 10
  5. 2025Hierarchical LoRA MoE for Efficient CTR Model Scalingcited by 10
  6. 2025Harnessing Consistency for Robust Test-Time LLM Ensemblecited by 10
  7. 2025CATS: Mitigating Correlation Shift for Multivariate Time Series Classificationcited by 7
  8. 2024THeGCN: Temporal Heterophilic Graph Convolutional Networkcited by 6
  9. 2025Panda: Test-Time Adaptation with Negative Data Augmentationcited by 5
  10. 2024SimCE: Simplifying Cross-Entropy Loss for Collaborative Filteringcited by 4
  11. 2026AdaFuse: Adaptive Ensemble Decoding with Test-Time Scaling for LLMscited by 4
  12. 2026ALERT: Zero-shot LLM Jailbreak Detection via Internal Discrepancy Amplificationcited by 3
  13. 2023Noisy Positive-Unlabeled Learning with Self-Training for Speculative Knowledge Graph Reasoningcited by 2

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 5.

5 papers.

  1. 2026Mixture of Sequence: Theme-Aware Mixture-of-Experts for Long-Sequence Recommendationcited by 4
  2. 2026Graph homophily booster: Reimagining the role of discrete features in heterophilic graph learningcited by 1
  3. 2026ReMix: Reinforcement routing for mixtures of LoRAs in LLM finetuningcited by 1
  4. 2026AvAtar: Learning to Align via Active Optimal Transportcited by 1
  5. 2026Ramen: Robust Test-Time Adaptation of Vision-Language Models with Active Sample Selectioncited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Realistic Test-Time Adaptation of Vision-Language Models
  2. 2025The Illusion of Progress? A Critical Look at Test-Time Adaptation for Vision-Language Models
  3. 2026ProtoDCS: Towards Robust and Efficient Open-Set Test-Time Adaptation for Vision-Language Models
  4. 2026From Inference to Adaptation: A Unified Optimal Transport View of Vision Language Model
  5. 2025Test-Time Consistency in Vision Language Models
  6. 2024Frustratingly Easy Test-Time Adaptation of Vision-Language Models
  7. 2026T-VSS: Test-Time Visual Subspace Steering for Adversarial Robustness of Vision-Language Models
  8. 2025Adapting Vision-Language Models Without Labels: A Comprehensive Survey