mlsanityData to 14 Oct 2026

Can VLM Pseudo-Labels Train a Time-Series QA Model That Outperforms the VLM?

arXiv 2509.25696 ↗First posted 30 Sep 2025cs.LG cs.CL eess.SPcited by 0

Not on the map. Why?

Abstract (arXiv)

Time-series question answering (TSQA) tasks face significant challenges due to the lack of labeled data. Alternatively, with recent advancements in large-scale models, vision-language models (VLMs) have demonstrated the potential to analyze time-series signals in a zero-shot manner. In this paper, we propose a training approach that uses pseudo labels generated by a VLM. Although VLMs can produce incorrect labels, TSQA models can still be effectively trained based on the property that deep neural networks are inherently robust to such noisy labels. Our experimental results demonstrate that TSQA models are not only successfully trained with pseudo labels, but also surpass the performance of the VLM itself by leveraging a large amount of unlabeled data.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 19 of its 26 references that are in the map. First posted on arXiv 30 Sep 2025.

Builds on

19 of 26 references.

References · 19

  1. 2020Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecastingcited by 2,169
  2. 2021Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecastingcited by 1,676
  3. 2019MixMatch: A Holistic Approach to Semi-Supervised Learningcited by 1,391
  4. 2023Large Language Models Are Zero-Shot Time Series Forecasterscited by 500
  5. 2020Early-Learning Regularization Prevents Memorization of Noisy Labelscited by 416
  6. 2020Time Series Data Augmentation for Deep Learning: A Surveycited by 226
  7. 2022PromptCast: A New Prompt-based Learning Paradigm for Time Series Forecastingcited by 172
  8. 2017Deep Learning is Robust to Massive Label Noisecited by 165
  9. 2023Large Language Models are Few-Shot Health Learnerscited by 118
  10. 2024ChatTS: Aligning Time Series with LLMs via Synthetic Data for Enhanced Understanding and Reasoningcited by 103
  11. 2024Language Models Still Struggle to Zero-shot Reason about Time Seriescited by 88
  12. 2025Time-MQA: Time Series Multi-Task Question Answering with Context Enhancementcited by 82
7 more
  1. 2023ECG-QA: A Comprehensive Question Answering Dataset Combined With Electrocardiogramcited by 54
  2. 2024TimeSeriesExam: A time series understanding examcited by 49
  3. 2024Towards Time Series Reasoning with LLMscited by 37
  4. 2025ITFormer: Bridging Time Series and Natural Language for Multi-Modal QA with Large-Scale Multitask Datasetcited by 37
  5. 2025TSPulse: Tiny Pre-Trained Models with Disentangled Representations for Rapid Time-Series Analysiscited by 12
  6. 2024Domain-Independent Automatic Generation of Descriptive Texts for Time-Series Datacited by 6
  7. 2024CLaSP: Learning Concepts for Time-Series Signals from Natural Language Supervisioncited by 6

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2024Can Visual Language Models Replace OCR-Based Visual Question Answering Pipelines in Production? A Case Study in Retail
  2. 2023Q: How to Specialize Large Vision-Language Models to Data-Scarce VQA Tasks? A: Self-Train on Unlabeled Images!
  3. 2026TS-Skill: A Benchmark for Evaluating Analytical Skills in Time-Series Question Answering
  4. 2023Generative Visual Question Answering
  5. 2024Right this way: Can VLMs Guide Us to See More to Answer Questions?
  6. 2024Natural Language Understanding and Inference with MLLM in Visual Question Answering: A Survey
  7. 2025When Big Models Train Small Ones: Label-Free Model Parity Alignment for Efficient Visual Question Answering using Small VLMs
  8. 2025The Quest for Visual Understanding: A Journey Through the Evolution of Visual Question Answering