SoftReason: A Fully Differentiable Neuro-Soft-Symbolic Deductive Reasoning Architecture over High-Dimensional Perceptual Data

2026-07-22

AI Summary

This work addresses the challenge of performing symbolic reasoning over high-dimensional perceptual data, where a discrete interface between perception and symbols leads to gradient discontinuities and necessitates integration of structured predicates and reliable evidence from knowledge graphs. To bridge this gap, the paper proposes an end-to-end differentiable neuro-symbolic reasoning architecture that unifies probabilistic grounded facts derived from perception with soft evidences from knowledge graphs within a local soft interpretation tensor. This is achieved through a learnable, differentiable immediate consequence operator, leveraging predicate embeddings, latent composition channels, and probabilistic OR aggregation to enable monotonic and differentiable updates of the deductive closure. Evaluated on knowledge-aware visual question answering (KVQA), the approach effectively integrates perceptual grounding, knowledge injection, and symbolic reasoning, demonstrating both expressive power and the advantages of end-to-end trainability.

Abstract

In many reasoning problems, the premises are not observed as discrete symbols, but must be inferred from high-dimensional inputs. Further, the predicate vocabulary, argument structure, and trusted evidence are supplied by a Knowledge Graph (KG), or rule definitions. Classical neuro-symbolic pipelines have a discrete interface between perception and deduction. We present a neuro-soft-symbolic architecture for differentiable deductive reasoning over latent perceptual facts and knowledge-provided predicates. SoftReason removes the gradient gap by representing the deductive state as a local soft interpretation tensor over candidate constants and predicates. Perception proposes probabilistic base facts, KG triples enter as high-confidence soft evidence, and every query anchor, predicate choice, and closure update remains differentiable. Our core innovation is a learned differentiable lift of the immediate-consequence operator. It uses predicate-definition embeddings and latent composition channels to form soft body-predicate mixtures, aggregate over all possible witnesses, propose query-conditioned head facts, and update the interpretation through a monotone probabilistic OR. We instantiate the framework on Knowledge-aware Visual Question Answering (KVQA), and demonstrates how SoftReason supports end-to-end perceptual grounding, KG evidence injection, and differentiable deductive closure in one trainable architecture.

Problem

• neuro-symbolic reasoning
• differentiable deduction
• perceptual grounding
• knowledge graph
• soft logic

Innovation

• differentiable reasoning
• neuro-symbolic architecture
• soft logic
• deductive closure
• perceptual grounding