Efficient Training with Denoised Neural Weights

TL;DR

This paper introduces a diffusion-based weight generator that synthesizes neural weights for initialization, significantly reducing training time and improving quality in image-to-image translation tasks.

Contribution

It proposes a novel diffusion model approach to generate neural weights for initialization, enabling faster training and better results compared to traditional methods.

Findings

Achieves 15x faster training time for new concepts.

Produces better image quality than training from scratch.

Requires only 43.3 seconds to initialize training.

Abstract

Good weight initialization serves as an effective measure to reduce the training cost of a deep neural network (DNN) model. The choice of how to initialize parameters is challenging and may require manual tuning, which can be time-consuming and prone to human error. To overcome such limitations, this work takes a novel step towards building a weight generator to synthesize the neural weights for initialization. We use the image-to-image translation task with generative adversarial networks (GANs) as an example due to the ease of collecting model weights spanning a wide range. Specifically, we first collect a dataset with various image editing concepts and their corresponding trained weights, which are later used for the training of the weight generator. To address the different characteristics among layers and the substantial number of weights to be predicted, we divide the weights into equal-sized blocks and assign each block an index. Subsequently, a diffusion model is trained with such a dataset using both text conditions of the concept and the block indexes. By initializing the image translation model with the denoised weights predicted by our diffusion model, the training requires only 43.3 seconds. Compared to training from scratch (i.e., Pix2pix), we achieve a 15x training time acceleration for a new concept while obtaining even better image generation quality.