Papers by Daniel Dajun Zeng
POSITION BIAS MITIGATES POSITION BIAS: Mitigate Position Bias Through Inter-Position Knowledge Distillation (2025.emnlp-main)
Copied to clipboard
| Challenge: | Positional bias (PB) manifests as non-uniform sensitivity across contextual locations . previous studies have addressed PB by modifying the underlying architectures or employing extensive contextual awareness training. |
| Approach: | They propose a position-to-position knowledge distillation framework that leverages position-induced disparities to counteract PB. |
| Outcome: | The proposed framework reduces positional bias and improves performance on retrieval and reasoning tasks. |
Entropy Scheduling in Reinforcement Learning for Large Language Models (2026.findings-acl)
Copied to clipboard
| Challenge: | entropy in reinforcement learning functions analogously to the learning rate in LLMs. |
| Approach: | They propose an entropy scheduling system that optimizes different pre-set goals by controlling and scheduling entropicy at each step of the RL process. |
| Outcome: | The proposed method improves AIME2024 from 50.9 to 54.9 within 40 training steps. |
Reinforcement Learning–Guided Adaptive Tuning for Out-of-Distribution Harmful Text Detection (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for testing harmful information on social media rely on fixed parameters that fail to handle substantial semantic discrepancies . RLAT can be used to adapt to semantic variations while preventing overfitting from continuous tuning. |
| Approach: | They propose a reinforcement learning-guided adaptive tuning method for harmful text detection that optimizes consistency loss and applies word-level attention constraints to reduce over-reliance on local words. |
| Outcome: | The proposed method outperforms state-of-the-art models in cross-platform and cross-temporal scenarios across multiple public datasets. |
Uncertainty Unveiled: Can Exposure to More In-context Examples Mitigate Uncertainty for Large Language Models? (2025.findings-acl)
Copied to clipboard
| Challenge: | Recent advances in handling long sequences have unlocked new possibilities for long-context in-contact learning (ICL). |
| Approach: | They investigate how increased examples influence predictive uncertainty . they quantify uncertainty across different “shot” configurations and focus on EU . |
| Outcome: | The proposed model reduces uncertainty in simple and complex tasks by injecting task-specific knowledge. |