Research map / Reinforcement learning / Visual and representation RL
TiCRL: Textual Image Classification with Reinforcement Learning-Based Curriculum Learning
Gayoung KIM, Yuncheol Kang
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 RL-based curriculum (TiCRL) | CNN · RVL-CDIP CNN from scratch | accuracy: 80.43 -> 83.67 | Improved |
| add RL-based curriculum (TiCRL) vs equal-data baseline | CNN · RVL-CDIP CNN from scratch | accuracy: 78.69 -> 83.67 | Improved |
| replace linear/step/root predefined schedule with RL scheduler | TiCRL · RVL-CDIP CNN from scratch | accuracy: 76.96 -> 83.67 | Improved |
| replace SPL/ACL with RL scheduler | TiCRL · RVL-CDIP CNN from scratch | accuracy: 75.84 -> 83.67 | Improved |
| add RL agent instead of EMA threshold heuristic (ablation) | TiCRL · RVL-CDIP CNN from scratch | accuracy: 81.49 -> 83.67 | Improved |
| restrict action to 1D threshold adjustment instead of 2D action (ablation) | TiCRL · RVL-CDIP CNN from scratch | accuracy: 72.30 -> 83.67 | Improved |
| add Canny edge feature to difficulty measurer (ablation) | TiCRL · RVL-CDIP CNN (ablation seed) | accuracy: 81.40 -> 85.05 | Improved |
| add HOG feature to difficulty measurer (ablation) | TiCRL · RVL-CDIP CNN (ablation seed) | accuracy: 81.63 -> 85.05 | Improved |
| add GLCM contrast feature to difficulty measurer (ablation) | TiCRL · RVL-CDIP CNN (ablation seed) | accuracy: 81.07 -> 85.05 | Improved |
| add GLCM homogeneity feature to difficulty measurer (ablation) | TiCRL · RVL-CDIP CNN (ablation seed) | accuracy: 82.35 -> 85.05 | Improved |
| add cosine similarity feature to difficulty measurer (ablation) | TiCRL · RVL-CDIP CNN (ablation seed) | accuracy: 81.23 -> 85.05 | Improved |
| apply TiCRL to DiT fine-tuning | DiT · RVL-CDIP | accuracy: 91.07 -> 93.02 | Improved |
Closest papers by meaning
- BalCapRL : A Balanced Framework for RL-Based MLLM Image Captioning (NeurIPS 2026 · similarity 0.84)
- CtrlSynth: Controllable Image Text Synthesis for Data-Efficient Multimodal Learning (ICML 2025 · similarity 0.83)
- Learning Textual Prompts for Open-World Semi-Supervised Learning (CVPR 2025 · similarity 0.83)
- Policy Optimized Text-to-Image Pipeline Design (NeurIPS 2025 · similarity 0.83)
- Compound Text-Guided Prompt Tuning via Image-Adaptive Cues (AAAI 2024 · similarity 0.83)
- T-MARS: Improving Visual Representations by Circumventing Text Feature Learning (ICLR 2024 · similarity 0.83)
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.
