Advancements in Vision-Language Models and Tabular Data Classification

The field of vision-language models and tabular data classification is experiencing significant growth, with a focus on improving model performance, robustness, and fairness. Researchers are exploring innovative methods to enhance surrogate models, fine-grained bias exploration, and mitigation techniques. The development of new frameworks, such as the Human-Data-Model Interaction Canvas, is providing fresh perspectives on visual analytics. Furthermore, studies are investigating the capabilities of large language models in reasoning over tabular data and their limitations. Noteworthy papers include Harnessing LLMs Explanations to Boost Surrogate Models in Tabular Data Classification, which proposes a novel in-context learning framework, and Fine-Grained Bias Exploration and Mitigation for Group-Robust Classification, which introduces a method for capturing distributions as a mixture of latent groups. Additionally, the paper Towards Fair In-Context Learning with Tabular Foundation Models explores the fairness implications of tabular in-context learning and proposes preprocessing strategies to address bias.

Sources

Harnessing LLMs Explanations to Boost Surrogate Models in Tabular Data Classification

Do Language Model Agents Align with Humans in Rating Visualizations? An Empirical Study

Fine-Grained Bias Exploration and Mitigation for Group-Robust Classification

How well do LLMs reason over tabular data, really?

The Human-Data-Model Interaction Canvas for Visual Analytics

Arrow-Guided VLM: Enhancing Flowchart Understanding via Arrow Direction Encoding

Measuring and predicting variation in the difficulty of questions about data visualizations

Judging the Judges: Can Large Vision-Language Models Fairly Evaluate Chart Comprehension and Reasoning?

Position: Restructuring of Categories and Implementation of Guidelines Essential for VLM Adoption in Healthcare

Towards Fair In-Context Learning with Tabular Foundation Models

Boosting Text-to-Chart Retrieval through Training with Synthesized Semantic Insights

Built with on top of