Do Large Language Models Capture the Diversity in their Training Data?

• URL: http://arxiv.org/abs/2609.02275v1
• Date: Wed, 02 Sep 2026 08:21:55 GMT
• Title: Do Large Language Models Capture the Diversity in their Training Data?
• Abstract summary: We study whether large language models capture the full diversity of plausible outputs present in their training data.<n>We propose a post-hoc correction mechanism that generates multiple outputs for each input and reweights them through a matrix-entropy projection.<n>Our results reveal a systematic conditional diversity gap between modern generative models and their training data, and provide an information-theoretic framework for measuring and mitigating this gap.
• Score: 18.198091804236782
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large language models are trained to model conditional distributions over text, yet it remains inadequately understood whether they capture the full diversity of plausible outputs present in their training data. We study this question through an information-theoretic lens by comparing the conditional entropy of model-generated outputs with that of the corresponding training data. Given paired input-output samples, we use conditional entropy and its matrix-based analogue based on von Neumann entropy to measure output variability beyond what is explained by the conditioning input, without requiring multiple reference outputs for the same prompt. Across LLM families with publicly available training data, including OLMo, Pythia, and GPT-Neo, we consistently find that model-generated outputs exhibit lower conditional entropy than their training data, across different model scales, sequence lengths, and decoding strategies. We observe a similar conditional diversity gap beyond language modeling, including class-conditioned ImageNet generators and text-conditioned models trained on MS-COCO. To address this gap, we propose a post-hoc correction mechanism that generates multiple outputs for each input and reweights them through a matrix-entropy projection, increasing conditional diversity while remaining close to the original model distribution. We prove the concavity of the matrix-based conditional entropy functional, which makes the resulting entropy-constrained projection a convex optimization problem, and develop a scalable mirror-descent algorithm for its implementation. Our results reveal a systematic conditional diversity gap between modern generative models and their training data, and provide an information-theoretic framework for measuring and mitigating this gap.

Related papers

• Diffusion models for multivariate subsurface generation and efficient probabilistic inversion [0.0]
  Diffusion models offer stable training and state-of-the-art performance for deep generative modeling tasks.<n>We introduce a likelihood approximation accounting for the noise-contamination that is inherent in diffusion modeling.<n>Our tests show significantly improved statistical robustness, enhanced sampling of the posterior probability density function.
  arXiv  Detail & Related papers  (2025-07-21T17:10:16Z)
• Self-Refining Training for Amortized Density Functional Theory [5.5541132320126945]
  We propose a novel method that reduces the dependency of amortized DFT solvers on large pre-collected datasets by introducing a self-refining training strategy.<n>We derive our method as a minimization of the variational upper bound on the KL-divergence measuring the discrepancy between the generated samples and the target Boltzmann distribution defined by the ground state energy.
  arXiv  Detail & Related papers  (2025-06-02T00:32:32Z)
• On conditional diffusion models for PDE simulations [53.01911265639582]
  We study score-based diffusion models for forecasting and assimilation of sparse observations. We propose an autoregressive sampling approach that significantly improves performance in forecasting. We also propose a new training strategy for conditional score-based models that achieves stable performance over a range of history lengths.
  arXiv  Detail & Related papers  (2024-10-21T18:31:04Z)
• Steering Masked Discrete Diffusion Models via Discrete Denoising Posterior Prediction [88.65168366064061]
  We introduce Discrete Denoising Posterior Prediction (DDPP), a novel framework that casts the task of steering pre-trained MDMs as a problem of probabilistic inference. Our framework leads to a family of three novel objectives that are all simulation-free, and thus scalable. We substantiate our designs via wet-lab validation, where we observe transient expression of reward-optimized protein sequences.
  arXiv  Detail & Related papers  (2024-10-10T17:18:30Z)
• Masked Language Modeling Becomes Conditional Density Estimation for Tabular Data Synthesis [0.74454067778951]
  We introduce MaCoDE by redefining the consecutive multi-class classification task of Masked Modeling (MLM) as histogram-based conditional density estimation. Our approach enables the estimation of conditional densities across arbitrary combinations of target and conditional variables. To validate our proposed model, we evaluate its performance in synthetic data generation across 10 real-world datasets.
  arXiv  Detail & Related papers  (2024-05-31T03:26:42Z)
• Unveil Conditional Diffusion Models with Classifier-free Guidance: A Sharp Statistical Theory [87.00653989457834]
  Conditional diffusion models serve as the foundation of modern image synthesis and find extensive application in fields like computational biology and reinforcement learning. Despite the empirical success, theory of conditional diffusion models is largely missing. This paper bridges the gap by presenting a sharp statistical theory of distribution estimation using conditional diffusion models.
  arXiv  Detail & Related papers  (2024-03-18T17:08:24Z)
• CoCoGen: Physically-Consistent and Conditioned Score-based Generative Models for Forward and Inverse Problems [1.0923877073891446]
  This work extends the reach of generative models into physical problem domains. We present an efficient approach to promote consistency with the underlying PDE. We showcase the potential and versatility of score-based generative models in various physics tasks.
  arXiv  Detail & Related papers  (2023-12-16T19:56:10Z)
• DIFFormer: Scalable (Graph) Transformers Induced by Energy Constrained Diffusion [66.21290235237808]
  We introduce an energy constrained diffusion model which encodes a batch of instances from a dataset into evolutionary states. We provide rigorous theory that implies closed-form optimal estimates for the pairwise diffusion strength among arbitrary instance pairs. Experiments highlight the wide applicability of our model as a general-purpose encoder backbone with superior performance in various tasks.
  arXiv  Detail & Related papers  (2023-01-23T15:18:54Z)
• Learning from aggregated data with a maximum entropy model [73.63512438583375]
  We show how a new model, similar to a logistic regression, may be learned from aggregated data only by approximating the unobserved feature distribution with a maximum entropy hypothesis. We present empirical evidence on several public datasets that the model learned this way can achieve performances comparable to those of a logistic model trained with the full unaggregated data.
  arXiv  Detail & Related papers  (2022-10-05T09:17:27Z)
• A Graphical Model for Fusing Diverse Microbiome Data [2.385985842958366]
  We introduce a flexible multinomial-Gaussian generative model for jointly modeling such count data. We present a computationally scalable variational Expectation-Maximization (EM) algorithm for inferring the latent variables and the parameters of the model.
  arXiv  Detail & Related papers  (2022-08-21T17:54:39Z)
• Autoregressive Score Matching [113.4502004812927]
  We propose autoregressive conditional score models (AR-CSM) where we parameterize the joint distribution in terms of the derivatives of univariable log-conditionals (scores) For AR-CSM models, this divergence between data and model distributions can be computed and optimized efficiently, requiring no expensive sampling or adversarial training. We show with extensive experimental results that it can be applied to density estimation on synthetic data, image generation, image denoising, and training latent variable models with implicit encoders.
  arXiv  Detail & Related papers  (2020-10-24T07:01:24Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.