Probability Consistency in Large Language Models: Theoretical Foundations Meet Empirical Discrepancies

• URL: http://arxiv.org/abs/2505.08739v1
• Date: Tue, 13 May 2025 16:52:19 GMT
• Title: Probability Consistency in Large Language Models: Theoretical Foundations Meet Empirical Discrepancies
• Authors: Xiaoliang Luo, Xinyi Xu, Michael Ramscar, Bradley C. Love,
• Abstract summary: We prove sequence perplexity is invariant under any factorization, including forward, backward, or arbitrary permutations.<n>We retrain GPT-2 models across forward, backward, and arbitrary permuted orders on scientific text.
• Score: 11.615194868239849
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Can autoregressive large language models (LLMs) learn consistent probability distributions when trained on sequences in different token orders? We prove formally that for any well-defined probability distribution, sequence perplexity is invariant under any factorization, including forward, backward, or arbitrary permutations. This result establishes a rigorous theoretical foundation for studying how LLMs learn from data and defines principled protocols for empirical evaluation. Applying these protocols, we show that prior studies examining ordering effects suffer from critical methodological flaws. We retrain GPT-2 models across forward, backward, and arbitrary permuted orders on scientific text. We find systematic deviations from theoretical invariance across all orderings with arbitrary permutations strongly deviating from both forward and backward models, which largely (but not completely) agreed with one another. Deviations were traceable to differences in self-attention, reflecting positional and locality biases in processing. Our theoretical and empirical results provide novel avenues for understanding positional biases in LLMs and suggest methods for detecting when LLMs' probability distributions are inconsistent and therefore untrustworthy.

Related papers

• Flipping Against All Odds: Reducing LLM Coin Flip Bias via Verbalized Rejection Sampling [59.133428586090226]
  Large language models (LLMs) can often accurately describe probability distributions using natural language.<n>This mismatch limits their use in tasks requiring reliableity, such as Monte Carlo methods, agent-based simulations, and randomized decision-making.<n>We introduce Verbalized Rejection Sampling (VRS), a natural-language adaptation of classical rejection sampling.
  arXiv  Detail & Related papers  (2025-06-11T17:59:58Z)
• Counterfactual Realizability [52.85109506684737]
  We introduce a formal definition of realizability, the ability to draw samples from a distribution, and then develop a complete algorithm to determine whether an arbitrary counterfactual distribution is realizable.<n>We illustrate the implications of this new framework for counterfactual data collection using motivating examples from causal fairness and causal reinforcement learning.
  arXiv  Detail & Related papers  (2025-03-14T20:54:27Z)
• Permutation-Based Rank Test in the Presence of Discretization and Application in Causal Discovery with Mixed Data [16.892960387325743]
  In psychometric studies, the continuous level of certain personality dimensions of a person can only be measured after discretization.<n>We propose Mixed data Permutation-based Rank Test (MPRT), which properly controls the statistical errors even when some variables are discretized.<n>MPRT can effectively control the Type I error in the presence of discretization while previous methods cannot.
  arXiv  Detail & Related papers  (2025-01-31T09:47:26Z)
• Transformation-Invariant Learning and Theoretical Guarantees for OOD Generalization [34.036655200677664]
  This paper focuses on a distribution shift setting where train and test distributions can be related by classes of (data) transformation maps. We establish learning rules and algorithmic reductions to Empirical Risk Minimization (ERM) We highlight that the learning rules we derive offer a game-theoretic viewpoint on distribution shift.
  arXiv  Detail & Related papers  (2024-10-30T20:59:57Z)
• Generalizing to any diverse distribution: uniformity, gentle finetuning and rebalancing [55.791818510796645]
  We aim to develop models that generalize well to any diverse test distribution, even if the latter deviates significantly from the training data. Various approaches like domain adaptation, domain generalization, and robust optimization attempt to address the out-of-distribution challenge. We adopt a more conservative perspective by accounting for the worst-case error across all sufficiently diverse test distributions within a known domain.
  arXiv  Detail & Related papers  (2024-10-08T12:26:48Z)
• Credal Learning Theory [4.64390130376307]
  We lay the foundations for a credal' theory of learning, using convex sets of probabilities to model the variability in the data-generating distribution. Bounds are derived for the case of finite hypotheses spaces, as well as infinite model spaces, which directly generalize classical results.
  arXiv  Detail & Related papers  (2024-02-01T19:25:58Z)
• Incoherent Probability Judgments in Large Language Models [4.307483901449801]
  We assess the coherence of probability judgments made by autoregressive Large Language Models (LLMs)<n>Our results show that the judgments produced by these models are often incoherent, displaying human-like systematic deviations from the rules of probability theory.
  arXiv  Detail & Related papers  (2024-01-30T00:40:49Z)
• Prototype-based Aleatoric Uncertainty Quantification for Cross-modal Retrieval [139.21955930418815]
  Cross-modal Retrieval methods build similarity relations between vision and language modalities by jointly learning a common representation space. However, the predictions are often unreliable due to the Aleatoric uncertainty, which is induced by low-quality data, e.g., corrupt images, fast-paced videos, and non-detailed texts. We propose a novel Prototype-based Aleatoric Uncertainty Quantification (PAU) framework to provide trustworthy predictions by quantifying the uncertainty arisen from the inherent data ambiguity.
  arXiv  Detail & Related papers  (2023-09-29T09:41:19Z)
• Correcting Underrepresentation and Intersectional Bias for Classification [49.1574468325115]
  We consider the problem of learning from data corrupted by underrepresentation bias. We show that with a small amount of unbiased data, we can efficiently estimate the group-wise drop-out rates. We show that our algorithm permits efficient learning for model classes of finite VC dimension.
  arXiv  Detail & Related papers  (2023-06-19T18:25:44Z)
• Causal Discovery in Heterogeneous Environments Under the Sparse Mechanism Shift Hypothesis [7.895866278697778]
  Machine learning approaches commonly rely on the assumption of independent and identically distributed (i.i.d.) data. In reality, this assumption is almost always violated due to distribution shifts between environments. We propose the Mechanism Shift Score (MSS), a score-based approach amenable to various empirical estimators.
  arXiv  Detail & Related papers  (2022-06-04T15:39:30Z)
• Distributional Reinforcement Learning via Moment Matching [54.16108052278444]
  We formulate a method that learns a finite set of statistics from each return distribution via neural networks. Our method can be interpreted as implicitly matching all orders of moments between a return distribution and its Bellman target. Experiments on the suite of Atari games show that our method outperforms the standard distributional RL baselines.
  arXiv  Detail & Related papers  (2020-07-24T05:18:17Z)

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.