2025· Neural Information Processing Systems· pp. 33908-33937· 4 citations· ⚡ 1 influential· 46 references
Computer Science
TL;DR
This work proposes a novel training framework that explicitly aligns the LLM-generated judgment distribution with human evaluation distributions, and incorporates adversarial training to ensure a robust alignment with this true distribution, rather than overfitting to its imperfect approximation.
Abstract
LLMs have emerged as powerful evaluators in the LLM-as-a-Judge paradigm, offering significant efficiency and flexibility compared to human judgments. However, previous methods primarily rely on single-point evaluations, overlooking the inherent diversity and uncertainty in human evaluations. This approach leads to information loss and decreases the reliability of evaluations. To address this limitation, we propose a novel training framework that explicitly aligns the LLM-generated judgment distribution with human evaluation distributions. Specifically, we propose a distributional alignment objective based on KL divergence, combined with an auxiliary cross-entropy regularization to stabilize the training process. Furthermore, due to limited human annotations, empirical human distributions are merely noisy estimates of the true underlying distribution. We therefore incorporate adversarial training to ensure a robust alignment with this true distribution, rather than overfitting to its imperfect approximation. Extensive experiments across various LLM backbones and evaluation tasks demonstrate that our framework significantly outperforms existing closed-source LLMs and conventional single-point alignment methods, with superior alignment quality, strong robustness, and competitive evaluation accuracy.
NAPHA (eNtropy-Aware Post-Hoc Alignment), a simple yet effective lightweight post-hoc alignment method that matches the LLM distribution to the HJD by first assigning an instance to a discrete entropy class and then routing it to specialized, trained alignment models is proposed.
Sebastian Steindl, Nikos Voskarides, Alberto Gasparin et al.· 0 citations
This work proposes a Localize-Then-Decide framework, which restores the monotonic relationship between confidence and disagreement risk and enables high-probability agreement guarantees in large language models.
Xinyue Li, Yi Zhou, Guanqun Cao et al.· 0 citations
Deployment of Large Language Models (LLMs) on memory-constrained edge devices relies heavily on aggressive post-training quantization. However, evaluating these models is largely based on zero-shot task accuracy, which depends solely on argmax predictions and is insensitive to changes in the underlying predictive distribution. Consequently, accuracy can exhibit unstable, non-monotonic behavior under progressive quantization, masking substantial fidelity loss relative to the BFloat16 (BF16) uncompressed base model and providing misleading deployment signals. We introduce a distribution-sensitive evaluation framework quantifying information loss in quantized LLMs as the divergence between full-vocabulary predictive distributions at the token decision boundary. We compute statistical distances, including Jensen-Shannon Divergence and Total Variation Distance, between outputs of full-precision and quantized models, enabling a fine-grained analysis of distributional shift. Using this framework, we quantify probability mass displacement and distributional drift relative to the BF16 reference, capturing predictive distribution changes not reflected in top-1 accuracy. We conduct a 120-run experimental matrix across five foundation architectures and four reasoning benchmarks under progressive quantization regimes, from uncompressed BF16 to Q2_K, providing a systematic fidelity analysis. Our results show divergence metrics generally increase under stronger quantization, complementing task accuracy with a fidelity signal. Across tested llama-cpp schemes, mixed-precision Q4_K generally yields lower divergence than uniform Q4_0 at similar memory footprints. These findings motivate distribution-aware evaluation as a practical diagnostic complement to task accuracy; they do not directly establish correctness, calibration, safety, or user-perceived quality.
Shahzeb Qamar, L. Sparrenberg, Christian Bauckhage et al.· 0 citations
It is shown that prior scores, even when included only as context metadata, anchor judgments and systematically shift ratings toward their values, and effective mitigation must be validated for the intended model and task or domain.
A. Kapetanović, Kemal Altwlkany, Andro Merćep et al.· 0 citations
LigBench is proposed, an automated evaluation benchmark that enables fine-grained and reliable evaluation of AI research ideas, consistently applicable across different generation distributions, and PAIR-IQ, a dataset tailored for training pairwise idea judgment models and serving as an auxiliary reference to support more objective comparative evaluation.
Chen-Run Wang, Mingxuan Zhu, Tiancheng Huang et al.· 0 citations
Results demonstrate that the proposed method outperforms the baselines, including an LLM without debiasing and previous calibration methods, and it is confirmed that scoring bias varies across LLMs, tasks, and score ranges, indicating the importance of measuring latent number bias as the case may be.
