Need help with icnn?
Click the “chat” button below for chat support from the developer who created it, or find similar developers for support.

About the developer

232 Stars 51 Forks Apache License 2.0 99 Commits 5 Opened issues


Input Convex Neural Networks

Services available


Need anything else?

Contributors list

# 119,932
14 commits

Input Convex Neural Networks (ICNNs)

This repository is by Brandon Amos, Leonard Xu, and J. Zico Kolter and contains the TensorFlow source code to reproduce the experiments in our ICML 2017 paper Input Convex Neural Networks.

If you find this repository helpful in your publications, please consider citing our paper.

  title = {Input Convex Neural Networks},
  author = {Brandon Amos and Lei Xu and J. Zico Kolter},
  booktitle = {Proceedings of the 34th International Conference on Machine Learning},
  pages = {146--155},
  year = {2017},
  volume = {70},
  series = {Proceedings of Machine Learning Research},
  publisher = {PMLR},

Setup and Dependencies

  • Python/numpy
  • TensorFlow (we used r10)
  • OpenAI Gym + Mujoco (for the RL experiments)


└── - Optimize a function over the [0,1] box with the bundle entropy method.
                        (Development is still in-progress and we are still
                        fixing some numerical issues here.)

Synthetic Classification

This image shows FICNN (top) and PICNN (bottom) classification of synthetic non-convex decision boundaries.

├── - Main script.
├── - Create a figure of just the legend.
├── - Make the tile of images.
└── - Run all experiments on 4 GPUs.

Multi-Label Classification

(These are currently slightly inconsistent with our paper and we plan on synchronizing our paper and code.)

├── - Loads the Bibsonomy datasets.
├── - Compare ebundle and gradient descent.
├── - Train a feed-forward net baseline.
├── - Train an ICNN with the bundle entropy method.
├── - Train an ICNN with gradient descent and back differentiation.
└── - Plot the results from any multi-label cls experiment.

Image Completion

This image shows the test set completions on the Olivetti faces dataset over the first few iterations of training a PICNN with the bundle entropy method for 5 iterations.

├── - Train an ICNN with gradient descent and back differentiation.
├── - Train an ICNN with the bundle entropy method.
├── - Plot the results from any image completion experiment.
└── - Loads the Olivetti faces dataset.

Reinforcement Learning


From the RL directory, run a single experiment with:

python src/ --model ICNN --env InvertedPendulum-v1 --outdir output \
  --total 100000 --train 100 --test 1 --tfseed 0 --npseed 0 --gymseed 0
  • Use
    to select a model from
  • Use
    to select a task. TaskList
  • View all of the parameters with
    python -h


The TensorBoard summary is on by default. Use

--summary False
to turn it off. The TensorBoard summary includes (1) average Q value, (2) loss function, and (3) average reward for each training minibatch.

The testing total rewards are logged to log.txt. Each line is

[training_timesteps]  [testing_episode_total_reward]


To reproduce our experiments, run the scripts in the RL directory.


The DDPG portions of our RL code are from Simon Ramstedt's SimonRamstedt/ddpg repository.


Unless otherwise stated, the source code is copyright Carnegie Mellon University and licensed under the Apache 2.0 License. Portions from the following third party sources have been modified and are included in this repository. These portions are noted in the source files and are copyright their respective authors with the licenses listed.



We use cookies. If you continue to browse the site, you agree to the use of cookies. For more information on our use of cookies please see our Privacy Policy.