Learning Finite Linear Temporal Logic Specifications with a Specialized Neural Operator

TL;DR

This paper introduces NeuralLTLf, a neural network architecture that learns compact finite linear temporal logic formulas from system traces, effectively scaling to larger formulas and handling noisy data.

Contribution

It presents a novel neural operator with a specialized recurrent filter that learns and extracts symbolic LTLf formulas from labeled traces.

Findings

NeuralLTLf scales to larger formulas than existing methods.

Maintains high accuracy even with noisy data.

Successfully extracts symbolic formulas from learned truth tables.

Abstract

Finite linear temporal logic ($\mathsf{LTL}_f$) is a powerful formal representation for modeling temporal sequences. We address the problem of learning a compact $\mathsf{LTL}_f$ formula from labeled traces of system behavior. We propose a novel neural network operator and evaluate the resulting architecture, Neural$\mathsf{LTL}_f$. Our approach includes a specialized recurrent filter, designed to subsume $\mathsf{LTL}_f$ temporal operators, to learn a highly accurate classifier for traces. Then, it discretizes the activations and extracts the truth table represented by the learned weights. This truth table is converted to symbolic form and returned as the learned formula. Experiments on randomly generated $\mathsf{LTL}_f$ formulas show Neural$\mathsf{LTL}_f$ scales to larger formula sizes than existing approaches and maintains high accuracy even in the presence of noise.