DeepImbalancedRegressionForMLLMs: Four Multimodal Deep Imbalanced Regression Benchmarks

Four multimodal deep imbalanced regression benchmarks released for evaluating the CCC-GRPO reinforcement learning method. The dataset was created by Yao Du, Shanshan Song, and Xiaomeng Li for a paper accepted at ICML 2026. The training splits follow naturally long-tailed target distributions.

Use Cases

• Evaluating multimodal regression models based on the described long-tailed training distributions.
• Benchmarking reinforcement learning methods like CCC-GRPO for distributional awareness in MLLMs.
• Studying model performance on imbalanced multimodal regression tasks.
• Training models on naturally long-tailed data splits as described in the overview.

Strengths

• Contains four distinct benchmarks for evaluating multimodal regression models.
• Training splits follow naturally long-tailed target distributions as described.
• Created for a peer-reviewed ICML 2026 paper, suggesting academic rigor.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

ChanganYao on Hugging Face, associated with the paper 'CCC-GRPO: Injecting Distributional Awareness into MLLMs via Reinforcement Learning for Deep Imbalanced Regression'.

Collection Method

Created as benchmarks for evaluating the CCC-GRPO method; method of data collection is not specified.

Freshness

Last updated 2026-06-12 07:51:06