5arXiv cs.AI (Artificial Intelligence)·
SoftReason: Fully differentiable neuro-soft-symbolic architecture for deductive reasoning over perceptual data
A new arXiv preprint introduces SoftReason, a neuro-soft-symbolic architecture that eliminates the discrete gradient gap in classical neuro-symbolic pipelines by representing deductive state as a soft interpretation tensor over candidate constants and predicates. The system integrates probabilistic perception, Knowledge Graph evidence injection, and differentiable deductive closure into a single end-to-end trainable architecture. The core innovation is a learned differentiable lift of the immediate-consequence operator using predicate-definition embeddings. The framework is instantiated on Knowledge-aware Visual Question Answering (KVQA) as a proof-of-concept.
Related guides (1)
Related events (8)
DeepSWIP: Counterfactual reasoning for neural probabilistic logic programs via quotient-WMC
DeepSWIP introduces a single-world counterfactual semantics for DeepProbLog, enabling causal inference over neurosymbolic programs that combine neural perception with probabilistic logic. The approach uses neural materialization to reduce neural predicates to standard ProbLog choices, then applies Single World Intervention Programs (SWIPs) and weighted model counting to compute exact counterfactuals from a single transformed program. Experiments on MPI3D validate the method against a DeepTwin construction across 12,000 queries and show a 2.14× inference speedup, while a SUMO HOV experiment demonstrates that neural calibration degradation biases plug-in causal estimates and that a correctly scoped AIPW estimator removes most first-order bias.
G-RRM: Neuro-symbolic approach guides SAT solvers with recurrent neural reasoning models
Researchers introduce G-RRM (Guiding with Recurrent Reasoning Models), a neuro-symbolic framework that uses symbol-equivariant recurrent neural networks (SE-RRMs) to guide classical symbolic solvers—including backtracking and SAT solvers Glucose 4.1 and CaDiCaL 3.0.0—for constraint satisfaction problems. On 9×9 Sudoku, the approach achieves 33.3× speedup for backtracking and 1.70× for Glucose 4.1, but shows no significant gain for CaDiCaL due to its overhead-dominated runtime and inability to overwrite injected hints. The paper identifies two conditions for neural guidance to be effective: a large combinatorial search space and a solver architecture capable of dynamically overriding imperfect neural hints.
RSF-GLLM: Differentiable graph reasoning framework for multi-hop knowledge graph QA
RSF-GLLM is a new framework for multi-hop question answering over knowledge graphs that decouples differentiable graph reasoning from LLM-based answer generation. The core Recurrent Soft-Flow (RSF) module uses a GRU-guided query updater with dynamic gating to traverse semantically dissimilar bridge nodes, with flow sparsity regularization guaranteeing convergence to discrete reasoning paths. Extracted paths are textualized to fine-tune an LLM, grounding generation in factual graph topology. Experiments on WebQSP and CWQ benchmarks show competitive accuracy with improved inference efficiency over LLM-heavy baselines.
SciReasoner: Multimodal foundation model for native structural reasoning across biology, chemistry, and materials science
Researchers introduce SciReasoner, a multimodal scientific foundation model that discretizes molecular coordinates, topologies, and periodic connectivities into a unified structure-aware vocabulary for reasoning over proteins, small molecules, and inorganic crystals. The model achieves state-of-the-art performance on 67 of 86 benchmarks, with notable gains in Gene Ontology prediction (F_max 0.42→0.55), retrosynthesis accuracy (0.63→0.72), and materials phase separation. Expert evaluators rated its reasoning traces as preferred or comparable to a frontier LLM in 98% of cases, suggesting meaningful interpretability alongside accuracy gains.
KGRL: Neuro-symbolic reinforcement learning for parametrized action spaces using Datalog knowledge bases
A new arXiv preprint introduces KGRL (Knowledge- and Gradient-Guided Reinforcement Learning), a neuro-symbolic algorithm for Parametrized Action Markov Decision Processes (PAMDPs) where each decision involves both a symbolic action and continuous numerical parameters. KGRL uses a Datalog knowledge base to prune infeasible actions and constrain parameter spaces, then applies gradient-based refinement to estimate optimal parameters during training and deployment. The approach improves sample efficiency and episodic return over state-of-the-art PAMDP baselines, and provides local procedural explanations for its decisions.
SD-GPS: Solver-Driven Autoformalization and Theorem Proposing for Geometry Problem Solving
Researchers propose SD-GPS, a neuro-symbolic framework for geometry problem solving that treats a symbolic solver as an execution oracle during both formalization and deduction stages. The system combines solvability-guided reinforcement learning for autoformalization (built on QwenVL3-2B) with an impasse-aware agent that proposes and symbolically verifies auxiliary lemmas. Evaluations on Geometry3K and PGPS9K show SD-GPS outperforms existing multimodal, neural, and neuro-symbolic baselines across multiple task regimes. The work advances the line of research on grounding neural agents in formal systems for verifiable mathematical reasoning.
NeSyCat Torch: Differentiable tensor framework unifying neurosymbolic semantics via monadic abstraction
NeSyCat Torch extends the NeSyCat/ULLER neurosymbolic framework with neural network support for predicates and functions, implemented via probabilistic programming and tensor backends (HaskTorch, JAX, PyTorch). The key technical contribution is a lazy log-tensor monad over the log-semiring enabling numerically stable, differentiable training, alongside a batch monad for efficient batched inference. On MNIST addition benchmarks, the implementations outperform LTN and DeepProbLog in speed and accuracy while remaining within a uniform categorical framework that generalizes across first-order neurosymbolic approaches. The work positions itself as a unifying foundation for classical, fuzzy, probabilistic, and neural truth semantics.
MemDreamer: Hierarchical graph memory and agentic retrieval for long video understanding
MemDreamer is a plug-and-play framework that decouples perception and reasoning for long-video understanding by incrementally building a three-tier Hierarchical Graph Memory capturing spatiotemporal and causal relations. During inference, a reasoning model uses an Observation-Reason-Action loop with agentic tool-augmented retrieval to navigate the memory graph, constraining the context window to 2% of full-context ingestion while achieving a 12.5-point absolute accuracy gain. The system reaches SOTA on four benchmarks, narrowing the gap with human experts to 3.7 points. The authors also report a strong linear correlation between logical reasoning performance and long-video understanding, proposing agentic capability scaling as a new paradigm for multimodal comprehension.