ZICQ
中 Log in / Sign up
Newsroom LLMs & Foundation Models #Neurosymbolic #Edge Computing #Inference Optimization #Resource-Constrained Devices #AI Reasoning

Neurosymbolic Routing: A New Approach for Efficient Reasoning on Edge Devices

Avatar of Mr.Xu

By Mr.Xu Compiled & Reviewed by Editorial

Published: · 6 views

中文阅读 (Chinese) English Version
Co-reading Select any sentence in the article to highlight, annotate, or copy a quote card. Join discussion ↓

Summary:This paper introduces Neurosymbolic Routing, a novel approach to address the unreliability of reasoning on resource-constrained edge devices when running language models. By classifying queries and dispatching them to the most appropriate solver, the method achieves efficient and accurate reasoning. Experimental results show that the approach achieves 100% routing accuracy and 98.3% overall accuracy on a Raspberry Pi 4B, significantly outperforming existing baselines.


Background and Challenges

Running language models on resource-constrained edge devices often faces challenges in terms of reasoning accuracy and efficiency. Small models struggle with complex reasoning tasks such as arithmetic, algebra, and formal logic, while traditional probabilistic models sacrifice accuracy and energy when approximating these tasks.

Methodology and Innovation

This paper introduces Neurosymbolic Routing, a novel approach that addresses the above issues through the following:

  1. Query Classification and Dispatching: A classifier categorizes each incoming query and dispatches it to the most appropriate solver. Structured tasks are sent to deterministic engines, while open-ended word problems are handled by a small language model (SLM).

  2. Learning Routing Logic: Instead of hand-coding the routing logic, a deterministic finite automaton (DFA) is learned using the L* grammatical inference algorithm. The SLM serves as a membership oracle, and labeled data acts as an equivalence oracle.

  3. Efficient Solving: Since formatted queries never reach the model, the router can answer these queries in 1-11 milliseconds. In its 30-token configuration, the router runs 8.8x faster and is 2.8x more energy-efficient than Program-of-Thought.

Experimental Results

Evaluated on a Raspberry Pi 4B (8 GB RAM, no GPU) with 100 untested prompts from DeepMind Mathematics, GSM8K, and RuleTaker, the learned routing achieves 100% routing accuracy and 98.3% overall accuracy (93.3% on word problems), compared to 72.0% for the strongest agent baseline, Program-of-Thought, and 58.7% for a tool-calling agent given the same solvers.

Industry Impact and Developer Recommendations

Neurosymbolic Routing offers a new solution for achieving efficient and accurate reasoning on edge devices, particularly in applications where privacy and low latency are critical. Developers can leverage this approach to enhance the performance of AI models in resource-constrained environments by combining the strengths of symbolic reasoning and neural networks.


Source: ArXiv AI (cs.AI) (2026-09-30)

— END —

Tags: #Neurosymbolic #Edge Computing #Inference Optimization #Resource-Constrained Devices #AI Reasoning

Editorial & Fact-Checking Note: This article is compiled from primary research, official release documentation, and source papers by the ZICQ Newsroom pipeline with automated entity verification and human editorial review. If you notice any technical inaccuracy, please submit a correction via our corrections policy or email our editorial desk directly.

What's your take on this story?
🚀 Explore & Submit Works

Community Comments

Loading live comments and annotations…