Breaking the Scale Barrier: One-Shot Knowledge Transfer via Frequency Transform

Transferring knowledge by fine-tuning large-scale pre-trained networks has become a standard paradigm for downstream tasks, yet the knowledge of a pre-trained model is tightly coupled with monolithic architecture, which restricts flexible reuse across models of varying scales. In response to this challenge, recent approaches typically resort to either parameter selection, which fails to capture the interdependent structure of this knowledge, or parameter prediction using generative models that depend on impractical access to large network collections. In this paper, we identify the low-frequency components of model weights as the concrete carrier of foundational, task-agnostic knowledge—its "learngene"—and validate this by demonstrating its efficient inheritance by downstream models and tasks. Based on this insight, we propose FRONT (FRequency dOmain kNowledge Transfer), a novel framework that uses the Discrete Cosine Transform (DCT) to isolate the low-frequency "learngene". This learngene can be seamlessly adapted to initialize models of arbitrary size via simple truncation or padding, a process that is entirely training-free. For enhanced performance, we propose an optional low-cost refinement process that introduces a spectral regularizer to further improve the learngene's transferability. Extensive experiments demonstrate that FRONT achieves the state-of-the-art performance, accelerates convergence by up to $15\times$ in vision tasks, and reduces training FLOPs by an average of 40.5\% in language tasks. Code is available at https://github.com/LUcy0505/FRONT.

Today, powerful AI systems are built by training massive models at a high cost. We wondered if we could easily apply the knowledge learned from these giant models and reuse it to jump-start the training of smaller, more practical models. Previously, this was incredibly difficult because an AI's knowledge is tightly locked into its specific size and shape. We discovered that the core, reusable essence of a model's understanding—which we call its "learngene"—is consistently stored in a specific mathematical pattern within its internal connections, known as "low-frequency" components. We built a framework called FRONT to isolate these specific patterns. By extracting just this low-frequency knowledge, we can easily reshape and plug it into new models of almost any size without needing to train them from scratch. Our experiments show that this approach works exceptionally well. It speeds up the training of visual recognition models by up to 15 times and cuts the computational cost of training language models by 40%. To help the community build faster and more efficient super AI, we have released our code openly.