Lucid is a collection of infrastructure and tools for research in neural network interpretability.

In particular, it provides state of the art implementations of feature visualization techniques, and flexible abstractions that make it very easy to explore new research directions.

Notebooks corresponding to the Feature Visualization article

Notebooks corresponding to the Building Blocks of Interpretability article

• Feature Visualization
• The Building Blocks of Interpretability
• Using Artificial Intelligence to Augment Human Intelligence
• Visualizing Representations: Deep Learning and Human Beings

You may use this software under the Apache 2.0 License. See LICENSE.

This project is research code. It is not an official Google product.

We use naming conventions to help differentiate tensors, operations, and values:

• Suffix variable names representing tensors with _t
• Suffix variable names representing operations with _op
• Don't suffix variable names representing concrete values

Usage example:

global_step_t = tf.train.get_or_create_global_step()
global_step_init_op = tf.variables_initializer([global_step_t])
global_step = global_step_t.eval()

Use tox to run the test suite on all supported environments.

To run tests only for a specific module, pass a folder to tox: tox tests/misc/io

To run tests only in a specific environment, pass the environment's identifier via the -e flag: tox -e py27.

After adding dependencies to setup.py, run tox with the --recreate flag to update the environments' dependencies.