Multi-class Temporal Logic Neural Networks

Danyang Li,Roberto Tron

2024-06-25

Abstract:Time-series data can represent the behaviors of autonomous systems, such as drones and self-driving cars. The task of binary and multi-class classification for time-series data has become a prominent area of research. Neural networks represent a popular approach to classifying data; However, they lack interpretability, which poses a significant challenge in extracting meaningful information from them. Signal Temporal Logic (STL) is a formalism that describes the properties of timed behaviors. We propose a method that combines all of the above: neural networks that represent STL specifications for multi-class classification of time-series data. We offer two key contributions: 1) We introduce a notion of margin for multi-class classification, and 2) we introduce STL-based attributes for enhancing the interpretability of the results. We evaluate our method on two datasets and compare it with state-of-the-art baselines.

Machine Learning

What problem does this paper attempt to address?

The paper primarily aims to address the multi-class classification problem of time series data and improve the interpretability of classification results by combining neural networks with Signal Temporal Logic (STL).

Specifically, the objectives of the paper include:

1. **Proposing a new method**: Combining neural networks and Signal Temporal Logic (STL) to achieve multi-class classification of time series data, and describing the behavioral properties of signals through STL formulas to enhance the interpretability of classification results.

2. **Introducing the concept of boundaries in multi-class classification**: To improve the separation between different classes, the authors propose a new boundary concept for multi-class classification tasks and design a loss function based on this concept, aiming to maximize these boundaries.

3. **Enhancing interpretability**: By introducing methods based on STL properties, the classification results become easier to understand. This includes defining two key contributions:
- Introducing the boundary concept in multi-class classification.
- Enhancing the interpretability of results based on STL properties.

The proposed Multi-class Temporal Logic Neural Networks in the paper can generate STL specifications for multi-class classification of time series data. Additionally, a new boundary concept and a new boundary-based loss function are introduced to improve the separation between different classes of data. Finally, the authors validate the effectiveness of the proposed method on 2 datasets and compare it with existing baselines.

In summary, the main goal of this paper is to develop a new method that can effectively and interpretably perform multi-class classification of time series data.