IAGA: Identity-Aware Gaussian Approximation for Efficient 3D Molecular Generation

Gaussian Probability Path based Generative Models (GPPGMs) generate data by reversing a stochastic process that progressively corrupts samples with Gaussian noise. While these models have achieved state-of-the-art performance in 3D molecular generation, their practical deployment remains constrained by the high computational cost of long generative trajectories, involving hundreds to thousands of steps during model training and sampling. In this work, we introduce a novel method that improves the efficiency of 3D molecular generation without sacrificing training granularity or inference fidelity. Our key insight is that different data modalities will exhibit markedly different rates of convergence to Gaussianity in the forward process of GPPGMs. We analytically identify a characteristic step at which the data has acquired sufficient Gaussianity, and then replace the remaining generation trajectory with a closed-form Gaussian approximation. Unlike existing techniques that accelerate the generation process via reformulating or coarsening the trajectories, our method preserves the full resolution of learning dynamics while avoiding redundant distributional transport with little data identity remained. Empirical results across different 3D molecular generation datasets demonstrate substantial improvements in both sample quality and computational efficiency.