Beyond the Final Layer: Intermediate Representations for Better Multilingual Calibration in Large Language Models

2025-10-03 · Ej Zhou, Caiqi Zhang, Tiancheng Hu, Chengzu Li, Nigel Collier, Ivan Vulić, Anna Korhonen arxiv

Confidence calibration, the alignment of a model's predicted confidence with its actual accuracy, is crucial for the reliable deployment of Large Language Models (LLMs). However, this critical property remains largely under-explored in multilingual contexts. In this work, we conduct the first large-scale, systematic studies of multilingual calibration across six model families and over 100 languages, revealing that non-English languages suffer from systematically worse calibration. To diagnose this, we investigate the model's internal representations and find that the final layer, biased by English-centric training, provides a poor signal for multilingual confidence. In contrast, our layer-wise analysis uncovers a key insight that late-intermediate layers consistently offer a more reliable and better-calibrated signal. Building on this, we introduce a suite of training-free methods, including Language-Aware Confidence Ensemble (LACE), which adaptively selects an optimal ensemble of layers for each specific language. Our study highlights the hidden costs of English-centric alignment and offer a new path toward building more globally equitable and trustworthy LLMs by looking beyond the final layer.

📄 PDF Abstract BibTeX arXiv:2510.03136

Code (0)

등록된 구현이 없습니다.

Self-Distilled Self-Supervised Representation Learning

2021-11-25 · Jiho Jang, Seonhoon Kim, KiYoon Yoo, Chaerin Kong 외

State-of-the-art frameworks in self-supervised learning have recently shown that fully utilizing transformer-based models can lead to performance boost compared to conventional CNN models. Striving to maximize the mutual…

Linear evaluation Representation Learning Self-Supervised Learning

Beyond the final layer: Attentive multilayer fusion for vision transformers

2026-01-14 · Laure Ciernik, Marco Morik, Lukas Thede, Luca Eyring 외 arxiv

With the rise of large-scale foundation models, efficiently adapting them to downstream tasks remains a central challenge. Linear probing, which freezes the backbone and trains a lightweight head, is computationally effi…

Online Knowledge Distillation via Mutual Contrastive Learning for Visual Recognition

2022-07-23 · Chuanguang Yang, Zhulin An, Helong Zhou, Fuzhen Zhuang 외

The teacher-free online Knowledge Distillation (KD) aims to train an ensemble of multiple student models collaboratively and distill knowledge from each other. Although existing online KD methods achieve desirable perfor…

Contrastive Learning image-classification Image Classification Knowledge Distillation+3

When Less Is More? Diagnosing ASR Predictions in Sardinian via Layer-Wise Decoding

2026-02-10 · Domenico De Cristofaro, Alessandro Vietti, Marianne Pouplier, Aleese Block arxiv

Recent studies have shown that intermediate layers in multilingual speech models often encode more phonetically accurate representations than the final output layer. In this work, we apply a layer-wise decoding strategy …

Beyond Compression: Quantifying Spectral Accessibility in Vision Representations

2026-06-02 · Akayou A. Kitessa, Yijun Zhao arxiv

Vision-language models map visual features into a shared embedding space through learned projection layers, yet it remains unclear how these transformations alter the structure of visual information. This study examines …

Dimensionality Reduction