Calibration, Entropy Rates, and Memory in Language Models

2019-06-11 · ICML 2020 1 · Mark Braverman, Xinyi Chen, Sham M. Kakade, Karthik Narasimhan, Cyril Zhang, Yi Zhang

Building accurate language models that capture meaningful long-term dependencies is a core challenge in natural language processing. Towards this end, we present a calibration-based approach to measure long-term discrepancies between a generative sequence model and the true distribution, and use these discrepancies to improve the model. Empirically, we show that state-of-the-art language models, including LSTMs and Transformers, are \emph{miscalibrated}: the entropy rates of their generations drift dramatically upward over time. We then provide provable methods to mitigate this phenomenon. Furthermore, we show how this calibration-based approach can also be used to measure the amount of memory that language models use for prediction.

📄 PDF Abstract BibTeX arXiv:1906.05664

Code (0)

등록된 구현이 없습니다.

PAEC: Position-Aware Entropy Calibration for LLM Reasoning in RLVR

2026-06-07 · Shumeng Yang, Yisu Liu, Jiayi Zheng, Zhaohui Yang 외 arxiv

Reinforcement learning with verifiable rewards (RLVR) improves large language model reasoning but often suffers from rapid policy-entropy collapse, where the policy prematurely concentrates on narrow high-probability rea…

Reinforcement Learning Mathematical Reasoning

Frontier Models Can Take Actions at Low Probabilities

2026-03-02 · Alex Serrano, Wen Xing, David Lindner, Erik Jenner arxiv

Pre-deployment evaluations inspect only a limited sample of model actions. A malicious model seeking to evade oversight could exploit this by randomizing when to "defect": misbehaving so rarely that no malicious actions …

Entropy-Tree: Tree-Based Decoding with Entropy-Guided Exploration

2026-01-02 · Longxuan Wei, Yubo Zhang, Zijiao Zhang, Zhihu Wang 외 arxiv

Large language models achieve strong reasoning performance, yet existing decoding strategies either explore blindly (random sampling) or redundantly (independent multi-sampling). We propose Entropy-Tree, a tree-based dec…

Respect Your Zero-Shot Uncertainty: Conservative Calibration for Test-Time-Adapted Vision-Language Models

2026-08-06 · Jingyan Jiang, Yaru Sun, Xiao Chen, Jiazhen Huang 외 arxiv

Test-time adaptation (TTA) can improve the recognition accuracy of vision-language models under distribution shift, but often degrades calibration, making predictive confidence unreliable for downstream decision-making. …

Test-time Adaptation

On the Entropy Calibration of Language Models

2025-11-15 · Steven Cao, Gregory Valiant, Percy Liang arxiv

We study the problem of entropy calibration, which asks whether a language model's entropy over generations matches its log loss on human text. Past work found that models are miscalibrated, with entropy per step increas…