Bumsub Ham

Prompt-based continual learning can now capture diverse image distributions, overcoming the limitations of prompt collapse that hinder performance across tasks.

Shift-and-sum quantization reduces reconstruction errors in visual autoregressive models, achieving state-of-the-art performance in post-training quantization.

By relating input features to output feature changes, RFC substantially boosts the accuracy of feature caching in diffusion transformers, leading to faster and more efficient image generation.