arxiv_ml 95% Match Research Paper AI Researchers,Machine Learning Engineers,Robotics Engineers 1 week ago

Neurosymbolic Diffusion Models

generative-ai › diffusion

📄 Abstract

Abstract: Neurosymbolic (NeSy) predictors combine neural perception with symbolic reasoning to solve tasks like visual reasoning. However, standard NeSy predictors assume conditional independence between the symbols they extract, thus limiting their ability to model interactions and uncertainty - often leading to overconfident predictions and poor out-of-distribution generalisation. To overcome the limitations of the independence assumption, we introduce neurosymbolic diffusion models (NeSyDMs), a new class of NeSy predictors that use discrete diffusion to model dependencies between symbols. Our approach reuses the independence assumption from NeSy predictors at each step of the diffusion process, enabling scalable learning while capturing symbol dependencies and uncertainty quantification. Across both synthetic and real-world benchmarks - including high-dimensional visual path planning and rule-based autonomous driving - NeSyDMs achieve state-of-the-art accuracy among NeSy predictors and demonstrate strong calibration.

Authors (4)

Emile van Krieken

Pasquale Minervini

Edoardo Ponti

Antonio Vergari

Submitted

May 19, 2025

arXiv Category

cs.LG

arXiv PDF

Key Contributions

Introduces neurosymbolic diffusion models (NeSyDMs) that use discrete diffusion to model dependencies between symbols, overcoming the limitations of conditional independence assumptions in standard NeSy predictors. This enables scalable learning while capturing symbol dependencies and uncertainty quantification, leading to state-of-the-art accuracy.

Business Value

Enables more robust and reliable AI systems for tasks requiring both perception and reasoning, such as autonomous systems and complex decision-making, by better handling uncertainty and generalization.

Paper Metadata

Innovation Type

Algorithmic

Deployment Feasibility

Potentially feasible, as it reuses independence assumptions for scalability, but requires careful integration of symbolic and neural components.

Limitations Addressed

Conditional independence assumption in standard NeSy predictors, leading to overconfident predictions and poor out-of-distribution generalization.

Performance Gains

State-of-the-art accuracy among NeSy predictors.

Technical Tags

neurosymbolic AIdiffusion modelssymbolic reasoninguncertainty quantificationdiscrete diffusionvisual reasoningout-of-distribution generalizationconditional independencestate-of-the-arthigh-dimensional data

Research Topics

Neurosymbolic AIGenerative ModelsReasoning under UncertaintyDeep Learning ArchitecturesModel Generalization

Methods & Architectures

Discrete DiffusionNeurosymbolic PredictionContrastive Learning Neurosymbolic Diffusion Models (NeSyDMs)

Applications & Tasks

Visual Reasoning Autonomous Driving Path Planning Modeling symbol dependenciesHandling uncertaintyImproving out-of-distribution generalizationOvercoming conditional independence assumption Visual reasoningRule-based autonomous drivingHigh-dimensional path planning

Related Fields

Artificial IntelligenceMachine LearningComputer VisionSymbolic AIProbabilistic Graphical Models

Keywords

neurosymbolic AIdiffusion modelssymbolic reasoninguncertaintygeneralizationvisual reasoningautonomous drivingpath planningdiscrete diffusionconditional independencestate-of-the-arthigh-dimensionalNeSyDMs

Academic Context

#Neurosymbolic AI#Generative Models#Reasoning under Uncertainty#Deep Learning Architectures#Model Generalization

Commercial Potential

Potential Products

More robust autonomous driving systemsAdvanced visual reasoning agentsAI systems for complex planning tasks

Target Industries

AutomotiveRoboticsAerospaceLogistics

Use Case Examples

Autonomous vehicles navigating complex scenariosRobots performing tasks requiring symbolic understandingAI assistants for visual question answering

Competitive Edge

Offers a novel approach to neurosymbolic prediction by incorporating diffusion models, aiming to outperform existing methods that struggle with symbol dependencies and uncertainty.

Resource Requirements

Compute Needs

Likely significant due to diffusion models and complex reasoning, but scalability is a design goal.

Data Requirements

Requires datasets suitable for visual reasoning and planning tasks, potentially with symbolic annotations.

Deployment Constraints

Integration complexity of symbolic and neural components, computational resources.

Scalability

Designed for scalable learning by reusing independence assumptions at each diffusion step.

Production Readiness

Maturity Level

Research

Time to Market

Long, requires further development and validation.

Patent Potential

Moderate, for the novel NeSyDM architecture and training methodology.

View Full Paper Back to Papers