Invertibility of Convolutional Generative Networks from Partial Measurements
Authors: Fangchang Ma, Ulas Ayaz, Sertac Karaman
NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We further demonstrate, empirically, that the same conclusion extends to networks with multiple layers, other activation functions (leaky Re LU, sigmoid and tanh), and weights trained on real datasets. |
| Researcher Affiliation | Collaboration | Fangchang Ma* MIT fcma@mit.edu Ulas Ayaz MIT uayaz@mit.edu uayaz@lyft.com Sertac Karaman MIT sertac@mit.edu Both authors contributed equally to this work. Ulas Ayaz is presently affiliated with Lyft, Inc. |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | The sample code is available at https://github.com/fangchangma/invert-generative-networks. |
| Open Datasets | Yes | We rescale the raw grayscale images from the MNIST dataset [11] to size of 32 ˆ 32. A similar study is conducted on a generative network trained on the Celeb Faces [12] dataset. |
| Dataset Splits | No | The paper does not explicitly specify training, validation, or test dataset splits or percentages. It only mentions the datasets used and the training framework. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used to run the experiments (e.g., GPU models, CPU types). |
| Software Dependencies | No | The paper mentions using "conditional deep convolutional generative adversarial networks (DCGAN) framework" and "Adam with learn rate 0.1" but does not specify version numbers for any software or libraries. |
| Experiment Setup | Yes | The first layer has 16 channels and the second layer has 1 single channel. Both layers have a kernel size of 5 and a stride of 3. We used Adam with learn rate 0.1 to optimize the latent code z . The optimization process usually converges within 500 iterations. The input noise to the generator is set to have a relatively small dimension 10 to ensure a sufficiently expanding network. |