Deep Parameter Interpolation for Scalar Conditioning
This addresses a bottleneck in generative modeling for researchers and practitioners by enabling more flexible and accurate scalar conditioning without restricting architecture choices.
The paper tackles the challenge of designing neural networks that effectively integrate scalar inputs, such as time or noise levels, with high-dimensional vector inputs in generative models like diffusion and flow matching, by proposing deep parameter interpolation (DPI), which improves denoising performance and sample quality while maintaining computational efficiency.
We propose deep parameter interpolation (DPI), a general-purpose method for transforming an existing deep neural network architecture into one that accepts an additional scalar input. Recent deep generative models, including diffusion models and flow matching, employ a single neural network to learn a time- or noise level-dependent vector field. Designing a network architecture to accurately represent this vector field is challenging because the network must integrate information from two different sources: a high-dimensional vector (usually an image) and a scalar. Common approaches either encode the scalar as an additional image input or combine scalar and vector information in specific network components, which restricts architecture choices. Instead, we propose to maintain two learnable parameter sets within a single network and to introduce the scalar dependency by dynamically interpolating between the parameter sets based on the scalar value during training and sampling. DPI is a simple, architecture-agnostic method for adding scalar dependence to a neural network. We demonstrate that our method improves denoising performance and enhances sample quality for both diffusion and flow matching models, while achieving computational efficiency comparable to standard scalar conditioning techniques. Code is available at https://github.com/wustl-cig/parameter_interpolation.