Research map / Architectures and efficiency / Attention and transformers
Transformers with Endogenous In-Context Learning: Bias Characterization and Mitigation
Haotian Wang, Hao Zou, Haoxuan Li, Haoang Chi, Yang Shi, Yuanxing Zhang, Wenjing Yang, Xinwang Liu, Zhouchen Lin
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 |
|---|---|---|---|
| Increase DDbias ICL examples from 15 to 60 | DDbias on LLaMA · Yelp RPI | MAE: 22.8 -> 16.7 | Improved |
| Increase DDbias ICL examples from 15 to 60 | DDbias on LLaMA · Yelp ROR | MAE: 0.36 -> 0.16 | Improved |
| Apply DDbias with 60 ICL examples instead of vanilla LLaMA | LLaMA · Yelp RPI | MAE: 23.8 -> 16.7 | Improved |
| Apply DDbias with 60 ICL examples instead of vanilla LLaMA | LLaMA · Yelp ROR | MAE: 0.46 -> 0.16 | Improved |
| Apply DDbias with only 15 ICL examples versus DMCEE baseline | LLaMA · Yelp RPI | MAE: 21.2 -> 22.8 | Hurt |
| Apply DDbias with 15 ICL examples versus SC baseline | LLaMA · Yelp ROR | MAE: 0.24 -> 0.36 | Hurt |
| Apply DDbias to 5-layer non-linear transformer | Transformer L=5 · Polynomial f, Conf@2.0 | prediction error: 0.600 -> 0.260 | Improved |
| Apply DDbias to 5-layer non-linear transformer | Transformer L=5 · Polynomial f, Conf@0.5 | prediction error: 0.280 -> 0.115 | Improved |
| Apply DDbias to 7-layer non-linear transformer | Transformer L=7 · Polynomial f, Conf@2.0 | prediction error: 0.750 -> 0.290 | Improved |
| Apply DDbias to 7-layer non-linear transformer | Transformer L=7 · Polynomial f, Conf@0.5 | prediction error: 0.320 -> 0.135 | Improved |
| Increase confounding strength from Conf@0.5 to Conf@2.0 (biased TF) (ablation) | Vanilla TF L=5 · Polynomial f | prediction error: 0.280 -> 0.600 | Hurt |
| Increase confounding strength from Conf@0.5 to Conf@2.0 (DDbias) (ablation) | DDbias L=5 · Polynomial f | prediction error: 0.115 -> 0.260 | Hurt |
Closest papers by meaning
- Technical Debt in In-Context Learning: Diminishing Efficiency in Long Context (NeurIPS 2025 · similarity 0.90)
- In-Context Learning through the Bayesian Prism (ICLR 2024 · similarity 0.89)
- How Do Transformers Learn In-Context Beyond Simple Functions? A Case Study on Learning with Representations (ICLR 2024 · similarity 0.89)
- Can In-context Learning Really Generalize to Out-of-distribution Tasks? (ICLR 2025 · similarity 0.89)
- Position: Do pretrained Transformers Learn In-Context by Gradient Descent? (ICML 2024 · similarity 0.89)
- Understanding In-Context Learning in Transformers and LLMs by Learning to Learn Discrete Functions (ICLR 2024 · similarity 0.89)
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.
