Explore the second part of a comprehensive discussion on STL Net, focusing on its application in STL specification checking for RNN and transformer outputs. Delve into the Student-Teacher Network Paradigm, STL Loss Function, and the architecture overview of two models. Learn about adjusting neural network results to meet specifications, with emphasis on converting specifications to DNF form. Examine example trace generation, metrics, and results from both generated data and air quality prediction data. Access accompanying slides and the original STL Net paper for deeper understanding. This video, part of the Neuro Symbolic Channel, offers insights into the intersection of symbolic methods and deep learning, contributing to advancements in artificial intelligence and machine learning.
Neural Networks, Temporal Logic, and Verification with STL Net - Part 2
Overview
Syllabus
Setup
Student-Teacher Network Paradigm
STL Loss Function
Two Models
Architecture Overview
Adjusting Neural Network Results to Meet a Specification
Key Idea: Converting Specification to DNF Form
Example Trace Generation
Metrics
Results on Generated Data
Results on Air Quality Prediction Data
Taught by
Neuro Symbolic
