NeurIPS 2026
Dynamic Expert Sharing: Decoupling Memory from Parallelism in Mixture-of-Experts Diffusion LLMs
Reducing expert activations and inference latency in mixture-of-experts diffusion language models through sequence-level expert sharing.
Generative models
Diffusion language models
Image and visual diffusion
-
AAAI 2026
HierarchicalPrune: Position-Aware Compression for Large-Scale Diffusion Models
Compressing large text-to-image diffusion models for efficient inference while preserving perceptual quality.
-
ICLR 2025
ConceptPrune: Concept Editing in Diffusion Models via Skilled Neuron Pruning
A training-free method that removes undesirable concepts from text-to-image diffusion models by pruning a small set of specialised weights.
-
ICML GenLaw Workshop · 2024
Memorized Images in Diffusion Models Share a Subspace That Can Be Located and Deleted
Locating and pruning a shared subspace associated with memorised training images in text-to-image diffusion models.
-
VCIP 2024 · Oral
SketchAnimator: Animate Sketch via Motion Customization of Text-to-Video Diffusion Models
Animating sketches by combining their visual appearance with customised motion from text-to-video diffusion models.
-
ICLR 2023 · Spotlight
SketchKnitter: Vectorized Sketch Generation with Diffusion Models
Generating recognisable vector sketches through a diffusion process over stroke-point locations and pen states.
Selected papers co-authored by Da Li. For the full publication list, visit Google Scholar.