Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Quantum Visual Word Sense Disambiguation: Unraveling Ambiguities Through Quantum Inference Model

Wenbo Qiao, Peng Zhang, Qinghua Hu

TL;DR

This work addresses semantic bias in visual word sense disambiguation by proposing a Quantum Inference Model for Unsupervised VWSD (Q-VWSD) that encodes multiple glosses into a superposition and measures a quantum posterior to select the best image. It formalizes Q-VWSD as a quantum generalization of classical probability, introduces a quantum circuit implementation (Q-VWSD$_{QC}$) and a heuristic quantum-inspired variant (Q-VWSD$_{QI}$), and demonstrates that interference among glosses can mitigate bias. Empirical results on the SE23 dataset show state-of-the-art performance, with pronounced gains when leveraging glosses from large language models, while remaining feasible on classical hardware. The study highlights the practical potential of quantum machine learning concepts for real-world tasks and suggests that quantum-inspired approaches can guide improved classical models even before quantum hardware becomes mature.

Abstract

Visual word sense disambiguation focuses on polysemous words, where candidate images can be easily confused. Traditional methods use classical probability to calculate the likelihood of an image matching each gloss of the target word, summing these to form a posterior probability. However, due to the challenge of semantic uncertainty, glosses from different sources inevitably carry semantic biases, which can lead to biased disambiguation results. Inspired by quantum superposition in modeling uncertainty, this paper proposes a Quantum Inference Model for Unsupervised Visual Word Sense Disambiguation (Q-VWSD). It encodes multiple glosses of the target word into a superposition state to mitigate semantic biases. Then, the quantum circuit is executed, and the results are observed. By formalizing our method, we find that Q-VWSD is a quantum generalization of the method based on classical probability. Building on this, we further designed a heuristic version of Q-VWSD that can run more efficiently on classical computing. The experiments demonstrate that our method outperforms state-of-the-art classical methods, particularly by effectively leveraging non-specialized glosses from large language models, which further enhances performance. Our approach showcases the potential of quantum machine learning in practical applications and provides a case for leveraging quantum modeling advantages on classical computers while quantum hardware remains immature.

Quantum Visual Word Sense Disambiguation: Unraveling Ambiguities Through Quantum Inference Model

TL;DR

This work addresses semantic bias in visual word sense disambiguation by proposing a Quantum Inference Model for Unsupervised VWSD (Q-VWSD) that encodes multiple glosses into a superposition and measures a quantum posterior to select the best image. It formalizes Q-VWSD as a quantum generalization of classical probability, introduces a quantum circuit implementation (Q-VWSD) and a heuristic quantum-inspired variant (Q-VWSD), and demonstrates that interference among glosses can mitigate bias. Empirical results on the SE23 dataset show state-of-the-art performance, with pronounced gains when leveraging glosses from large language models, while remaining feasible on classical hardware. The study highlights the practical potential of quantum machine learning concepts for real-world tasks and suggests that quantum-inspired approaches can guide improved classical models even before quantum hardware becomes mature.

Abstract

Visual word sense disambiguation focuses on polysemous words, where candidate images can be easily confused. Traditional methods use classical probability to calculate the likelihood of an image matching each gloss of the target word, summing these to form a posterior probability. However, due to the challenge of semantic uncertainty, glosses from different sources inevitably carry semantic biases, which can lead to biased disambiguation results. Inspired by quantum superposition in modeling uncertainty, this paper proposes a Quantum Inference Model for Unsupervised Visual Word Sense Disambiguation (Q-VWSD). It encodes multiple glosses of the target word into a superposition state to mitigate semantic biases. Then, the quantum circuit is executed, and the results are observed. By formalizing our method, we find that Q-VWSD is a quantum generalization of the method based on classical probability. Building on this, we further designed a heuristic version of Q-VWSD that can run more efficiently on classical computing. The experiments demonstrate that our method outperforms state-of-the-art classical methods, particularly by effectively leveraging non-specialized glosses from large language models, which further enhances performance. Our approach showcases the potential of quantum machine learning in practical applications and provides a case for leveraging quantum modeling advantages on classical computers while quantum hardware remains immature.
Paper Structure (22 sections, 22 equations, 9 figures, 6 tables)

This paper contains 22 sections, 22 equations, 9 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Quantum Machine Learning
  4. Visual Word Sense Disambiguation
  5. Preliminaries
  6. Quantum Computing
  7. Quantum Probability Theory
  8. Quantum Visual Word Sense Disambiguation
  9. Task Definition of Unsupervised VWSD
  10. Quantum Inference Circuit
  11. Formalization of Quantum Inference Circuit
  12. Quantum-Inspired Inference Model
  13. Glosses from Large Language Models
  14. Experiments
  15. Setup
  16. ...and 7 more sections

Figures (9)

  • Figure 1: Different sources of glosses focus on various perspectives, which can lead to erroneous VWSD results.
  • Figure 2: Fidelity between the target words constructed using glosses from WordNet and those constructed using glosses from ChatGPT 3.5. A shows that a large number of samples did not achieve 90% fidelity, indicating a semantic bias between WordNet and ChatGPT 3.5. B shows that after representation with quantum superposition states, the fidelity between WordNet and ChatGPT 3.5 reached 95%, suggesting hope for mitigating semantic bias.
  • Figure 3: Main structure of quantum inference circuit (Q-VWSD$_{QC}$). C denotes Eq. (\ref{['e3']})/(\ref{['e6']}) and indicates the initialization of a quantum state. S denotes Eq. (\ref{['e4']}) and indicates the construct of the superposition states of the glosses. M denotes Eq.(\ref{['e10']}) and indicates the construct of the observation.
  • Figure 4: Q-VWSD vs. classical methods. (a) Traditional image-text matching directly computes the cosine similarity. (b) The classical probabilistic model uses the law of total probability and applies Bayesian inference to accumulate evidence, thereby obtaining the posterior probability. (c) Our Q-VWSD employs quantum probability, and the "interference effects" between different glosses must also be considered.
  • Figure 5: Main structure of quantum-inspired inference model (Q-VWSD$_{QI}$). C denotes Eq. (\ref{['e3']})/(\ref{['e6']}) and indicates the initialization of a quantum state. A denotes Eq. (\ref{['e7']}) and refers to calculating the probability amplitude. S denotes Eq. (\ref{['e4']}) and indicates the construct of the superposition states of the glosses. M denotes Eq.(\ref{['e10']}) and indicates quantum measurement.
  • ...and 4 more figures

Theorems & Definitions (2)

  • Claim 1
  • proof