Conditional Idempotent Generative Networks

TMLR Paper2921 Authors

Abstract:

We propose Conditional Idempotent Generative Networks (CIGN), a new approach that expands upon Idempotent Generative Networks (IGN) to enable conditional generation.
While IGNs offer efficient single-pass generation, they lack the ability to control the content of the generated data.
CIGNs address this limitation by incorporating conditioning mechanisms, allowing users to steer the generation process towards specific types of data.

We establish the theoretical foundations for CIGNs, outlining their scope, loss function and evaluation metrics.
We then present two potential architectures for implementing CIGNs, which we call channel conditioning and filter conditioning.
We discuss experimental results obtained on the MNIST dataset, demonstrating the effectiveness of both conditioning approaches.
Our findings pave the way for further exploration of CIGNs on larger datasets and more complex use cases.

Submission Length: Regular submission (no more than 12 pages of main content)

Changes Since Last Submission: Fixing the formatting of the images at page 12 and 13.

Assigned Action Editor: ~Grigorios_Chrysos1

Submission Number: 2921