LiNNA

LiNNA (Linear Neural Network Abstraction) is a Python library for abstracting feed-fordward neural networks. The abstraction can then be used to speed up verification or to gain insights about redundancies in the network. The core idea is to replace neurons by a linear combination of neurons and thereby replace the overall size of the network.

Authors

LiNNA is developed and maintained at the Technical University of Munich by:

• Stefanie Mohr

• Jan Křetı́nský

• Calvin Chau

Publications

Ashok, P., Hashemi, V., Křetínský, J., Mohr, S. (2020). DeepAbstract: Neural Network Abstraction for Accelerating Verification. ATVA 2020. (Paper, arXiv).

API Documentation

Contents:

• Modules
  • automata Module
    • linna.network
      • Network
      • NetworkLayer
    • linna.utils
      • load_tf_network()
      • get_accuracy()
      • load_model()
      • load_experiment()
      • save_results()
      • get_counterexamples()
      • forward()
      • writeNNet()

Indices and tables

• Index

• Module Index

• Search Page