Rewriting a Deep Generative Model
This work addresses the challenge of interpretability and control in generative AI for researchers and practitioners, though it appears incremental as it builds on existing model architectures.
The paper tackles the problem of understanding and modifying specific rules encoded in deep generative models, such as GANs, by introducing a method to manipulate layers as linear associative memories, resulting in demonstrated advantages over fine-tuning and edit transfer algorithms on multiple datasets.
A deep generative model such as a GAN learns to model a rich set of semantic and physical rules about the target distribution, but up to now, it has been obscure how such rules are encoded in the network, or how a rule could be changed. In this paper, we introduce a new problem setting: manipulation of specific rules encoded by a deep generative model. To address the problem, we propose a formulation in which the desired rule is changed by manipulating a layer of a deep network as a linear associative memory. We derive an algorithm for modifying one entry of the associative memory, and we demonstrate that several interesting structural rules can be located and modified within the layers of state-of-the-art generative models. We present a user interface to enable users to interactively change the rules of a generative model to achieve desired effects, and we show several proof-of-concept applications. Finally, results on multiple datasets demonstrate the advantage of our method against standard fine-tuning methods and edit transfer algorithms.