Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Manifold-based Sampling for In-Context Hallucination Detection in Large Language Models

Bodla Krishna Vamshi, Rohan Bhatnagar, Haizhao Yang

TL;DR

MB-ICL is proposed, a manifold-based demonstration sampling framework for selecting in-context demonstrations that leverages latent representations extracted from frozen LLMs that outperforms standard ICL selection baselines in the majority of evaluated settings, with particularly strong gains on dialogue and summarization tasks.

Abstract

Large language models (LLMs) frequently generate factually incorrect or unsupported content, commonly referred to as hallucinations. Prior work has explored decoding strategies, retrieval augmentation, and supervised fine-tuning for hallucination detection, while recent studies show that in-context learning (ICL) can substantially influence factual reliability. However, existing ICL demonstration selection methods often rely on surface-level similarity heuristics and exhibit limited robustness across tasks and models. We propose MB-ICL, a manifold-based demonstration sampling framework for selecting in-context demonstrations that leverages latent representations extracted from frozen LLMs. By jointly modeling local manifold structure and class-aware prototype geometry, MB-ICL selects demonstrations based on their proximity to learned prototypes rather than lexical or embedding similarity alone. Across factual verification (FEVER) and hallucination detection (HaluEval) benchmarks, MB-ICL outperforms standard ICL selection baselines in the majority of evaluated settings, with particularly strong gains on dialogue and summarization tasks. The method remains robust under temperature perturbations and model variation, indicating improved stability compared to heuristic retrieval strategies. While lexical retrieval can remain competitive in certain question-answering regimes, our results demonstrate that manifold-based prototype selection provides a reliable and training light approach for hallucination detection without modifying LLM parameters, offering a principled direction for improved ICL demonstration selection.

Manifold-based Sampling for In-Context Hallucination Detection in Large Language Models

TL;DR

MB-ICL is proposed, a manifold-based demonstration sampling framework for selecting in-context demonstrations that leverages latent representations extracted from frozen LLMs that outperforms standard ICL selection baselines in the majority of evaluated settings, with particularly strong gains on dialogue and summarization tasks.

Abstract

Large language models (LLMs) frequently generate factually incorrect or unsupported content, commonly referred to as hallucinations. Prior work has explored decoding strategies, retrieval augmentation, and supervised fine-tuning for hallucination detection, while recent studies show that in-context learning (ICL) can substantially influence factual reliability. However, existing ICL demonstration selection methods often rely on surface-level similarity heuristics and exhibit limited robustness across tasks and models. We propose MB-ICL, a manifold-based demonstration sampling framework for selecting in-context demonstrations that leverages latent representations extracted from frozen LLMs. By jointly modeling local manifold structure and class-aware prototype geometry, MB-ICL selects demonstrations based on their proximity to learned prototypes rather than lexical or embedding similarity alone. Across factual verification (FEVER) and hallucination detection (HaluEval) benchmarks, MB-ICL outperforms standard ICL selection baselines in the majority of evaluated settings, with particularly strong gains on dialogue and summarization tasks. The method remains robust under temperature perturbations and model variation, indicating improved stability compared to heuristic retrieval strategies. While lexical retrieval can remain competitive in certain question-answering regimes, our results demonstrate that manifold-based prototype selection provides a reliable and training light approach for hallucination detection without modifying LLM parameters, offering a principled direction for improved ICL demonstration selection.
Paper Structure (24 sections, 4 equations, 8 figures, 4 tables)

This paper contains 24 sections, 4 equations, 8 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Contributions
  3. Related Work
  4. Hallucination detection
  5. In-Context learning
  6. Method overview
  7. Manifold Construction
  8. Anchor Initialization and Local Clustering
  9. Manifold Expansion and Validation
  10. Submanifold Basis Extraction
  11. Training Objectives
  12. Proxy Anchor Loss:
  13. Manifold Point-to-Point Loss:
  14. Distance Calculation.
  15. Experiments
  16. ...and 9 more sections

Figures (8)

  • Figure 1: Accuracy comparisons of MB-ICL vs other methods
  • Figure 2: Perplexity comparison across models on HaluEval dialogue task. Lower perplexity indicates better performance.
  • Figure 3: Perplexity comparison across models on HaluEval QA task. Lower perplexity indicates better performance.
  • Figure 4: Performance of MB-ICL under varying numbers of in-context demonstrations
  • Figure 5: Accuracy vs Prototype Size $Z'$
  • ...and 3 more figures