Neurologica

chevron-icon
RFP Proposals
Top
chevron-icon
project-presentation-img
user-profile-img
Prasad Kumkar
Project Owner

Neurologica

Expert Rating

n/a

Overview

NeuroLogica integrates differentiable logic programming (PyNeuraLogic) and Kolmogorov–Arnold Networks (KANs) into PRIMUS. By embedding experiential rules from AIRIS and higher-order logic into deep networks, it enhances interpretability, data efficiency, and reasoning across graphs and continuous models. PyNeuraLogic provides rule-based inference with learnable weights; KANs deliver compact, interpretable continuous models. Together they enable adaptive learning from small datasets, bridging data-driven and symbolic reasoning. NeuroLogica will demonstrate proof-of-concepts in smart-grid forecasting and logical reasoning, delivering a reproducible toolkit and evaluation metrics by month 9.

RFP Guidelines

Neural-symbolic DNN architectures

Complete & Awarded
  • Type SingularityNET RFP
  • Total RFP Funding $160,000 USD
  • Proposals 17
  • Awarded Projects 1
author-img
SingularityNET
Apr. 14, 2025

This RFP invites proposals to explore and demonstrate the use of neural-symbolic deep neural networks (DNNs), such as PyNeuraLogic and Kolmogorov Arnold Networks (KANs), for experiential learning and/or higher-order reasoning. The goal is to investigate how these architectures can embed logic rules derived from experiential systems like AIRIS or user-supplied higher-order logic, and apply them to improve reasoning in graph neural networks (GNNs), LLMs, or other DNNs. Bids are expected to range from $40,000 - $100,000.

Proposal Description

Our Team

Prasad - 6+ years in Web3 R&D and protocol design, with deep expertise in decentralized ledger consensus, incentive models, and smart-contract architectures. Bachelor’s in Computer Science & Engineering.

Siva - Academic background in cryptography and blockchain. Computer Engineering degree with strong systems background; Advocates for statistically sound evaluation; has designed and executed large-scale benchmark suites for consensus protocols and cryptographic primitives.

Company Name (if applicable)

Chainscore Labs

Project details

NeuroLogica is a hybrid neural-symbolic research initiative that combines differentiable logic programming with advanced continuous-function networks to achieve both experiential learning and higher-order reasoning. Our goal is to embed symbolic rules—derived either from agent-driven exploration systems or from human experts—directly into deep neural network architectures, and to complement this with compact, interpretable function approximators. By integrating PyNeuraLogic and KANs within the PRIMUS cognitive framework on Hyperon, NeuroLogica will demonstrate a novel, reproducible approach to building AI systems that learn from limited data, reason over complex structures, and explain their inferences in human-readable form.

Motivation & Rationale
Pure deep learning models excel at pattern recognition but lack transparency and struggle with small datasets or structured reasoning. Pure symbolic engines offer clear logic but are brittle when faced with noisy or high-dimensional inputs. NeuroLogica bridges this divide by:

  • Leveraging Prior Knowledge: Symbolic rules guide neural learning, reducing data requirements.

  • Ensuring Explainability: Learned models preserve rule structure and yield interpretable functions.

  • Enhancing Reasoning: Logic inference and numeric prediction work together to solve structured and continuous tasks.
    This synergy aligns with the RFP’s emphasis on experiential rule embedding, higher-order reasoning, data efficiency, and integration into PRIMUS.

Core Components

  1. PyNeuraLogic Module

    • Implements differentiable forward-chaining inference: symbolic clauses become neural computation nodes with trainable weights.

    • Supports graph and ontology data: relational structures are handled natively, enabling tasks like knowledge-graph completion and rule-guided classification.

    • Facilitates dynamic rule updates: rules generated by AIRIS or provided by users are loaded into Atomspace, and PyNeuraLogic adjusts its inference graph accordingly.

  2. KAN Module

    • Realizes learnable univariate functions on each network edge, replacing fixed activations and scalar weights.

    • Achieves superior approximation efficiency: compact networks reach target accuracies with fewer parameters and less training data.

    • Exposes edge functions for post-training visualization: researchers can extract and inspect mathematical curves that the model learned, translating them into domain-relevant formulas.

Integration with PRIMUS/Hyperon
NeuroLogica’s components will be orchestrated via MeTTa scripts within Hyperon’s Atomspace knowledge graph. Key integration pathways include:

  • Rule Exchange: PyNeuraLogic reads AIRIS-generated rules and writes back inferred facts, all represented as Atomspace atoms.

  • Numeric Services: KANs are exposed as Python services callable from MeTTa; prediction results and extracted functions are stored in Atomspace for downstream planning and explanation.

  • Cognitive Loop: Experience → Rule Generation → Logic-Neural Inference → Numeric Prediction → Planning → New Experience, enabling continual learning and reasoning.

NeuroLogica will validate its approach through two primary POCs [TBD - but this is just a few basic ones we could thiink of right now]:

  • Smart-Grid Forecasting & Control: Embed grid-topology and safety rules via PyNeuraLogic; forecast load fluctuations with KANs; demonstrate how combined reasoning and prediction yield actionable insights and clear explanations for operators.

  • Clinical Decision Support: Encode medical ontology rules (symptoms, diagnoses, drug interactions) in PyNeuraLogic; learn patient-specific response curves with KANs; deliver a diagnostic assistant that provides both risk scores and logical justifications for recommendations.


NeuroLogica rbings a true convergence of neural learning and symbolic reasoning within a scalable AGI framework. By seamlessly embedding experiential and abstract rules into neural architectures, and by harnessing function-rich approximators, it fulfills the RFP’s vision of adaptive + explainable AI. At project close, NeuroLogica will deliver a versatile toolkit, ready for extension into additional domains, and a blueprint for integrating hybrid architectures in next-generation cognitive systems.

 

Open Source Licensing

MIT - Massachusetts Institute of Technology License

Links and references

Pitch Deck - https://drive.google.com/file/d/1MwU1DdiRAv4TY-TLPrdqdJqvg4v7MZ4b/view?usp=sharing

Proposal Video

Not Avaliable Yet

Check back later during the Feedback & Selection period for the RFP that is proposal is applied to.

  • Total Milestones

    3

  • Total Budget

    $80,000 USD

  • Last Updated

    27 May 2025

Milestone 1 - Research Plan & Architecture Design

Description

Over the first two months we will perform an in-depth survey of recent neural-symbolic research (post-2022) with focus on PyNeuraLogic and KANs and map their capabilities against PRIMUS/Hyperon requirements. We’ll define the overall NeuroLogica architecture specify module interfaces design rule templates for experiential and higher-order logic and establish development environments data pipelines and Atomspace integration patterns.

Deliverables

Comprehensive research brief summarizing SOTA architectures tools and integration strategies. Detailed architecture specification document with component diagrams and data flows. Development environment configured: Hyperon sandbox PyNeuraLogic connector repo PyKAN service scaffold MeTTa wrappers.

Budget

$16,000 USD

Success Criterion

Research brief and architecture spec reviewed and approved by the core team. Development environments provisioned and smoke-tested (e.g., sample rule read/write in Atomspace, KAN service startup). Clear project roadmap and risk register established.

Milestone 2 - Module Prototyping & Initial Integration

Description

During months 3–6 we will build and validate prototype implementations of the two core NeuroLogica modules. The PyNeuraLogic connector will load rules from Atomspace perform differentiable inference on sample graphs and write back new facts. The KAN service will train on continuous datasets expose prediction APIs and support post-training function extraction. We’ll also create MeTTa scripts to orchestrate end-to-end flows and develop data converters for smart-grid and clinical samples.

Deliverables

PyNeuraLogic module integrated with Atomspace and MeTTa capable of rule ingestion inference and fact export. KAN service with CLI/API for training prediction and function-curve export. MeTTa orchestration scripts demonstrating a simple workflow: ingest rules → run inference → call KAN predictor. Sample datasets and converters for power-grid and healthcare scenarios.

Budget

$32,000 USD

Success Criterion

Both modules pass unit and integration tests against defined use cases. End-to-end demo (via MeTTa) executes without errors and produces expected outputs. Team coin-reviews confirm code quality, documentation, and reproducibility.

Milestone 3 - POC Demonstrations Evaluation & Handover

Description

In the final three months we will execute two full proof-of-concepts: a smart-grid forecasting/control demo and a clinical decision-support assistant. We’ll run experiments to measure forecasting accuracy rule-based inference fidelity and explainability then refine models and integrations. Comprehensive documentation and user guides will be prepared to ensure reproducibility and ease of adoption.

Deliverables

Smart-grid POC: live dashboard showing rule triggers and KAN forecasts with extracted function visualizations. Clinical POC: diagnostic assistant that outputs risk assessments alongside logical justification traces. Evaluation report detailing metrics on accuracy data efficiency and interpretability with comparisons to baseline systems. Final code release deployment scripts developer guides and tutorial notebooks.

Budget

$32,000 USD

Success Criterion

POCs achieve or exceed target metrics (e.g., ≥90% of baseline accuracy with ≤20% data, clear rule/fn explanations). Stakeholder demonstration sessions completed with sign-off. All deliverables published in the project repository with full documentation and reproducible pipelines.

Join the Discussion (0)

Expert Ratings

Reviews & Ratings

    No Reviews Avaliable

    Check back later by refreshing the page.

Welcome to our website!

Nice to meet you! If you have any question about our services, feel free to contact us.