Bowen Chen

Sparse neuron dictionaries are not just passive interpretability tools: actively steering frozen VLM tokens through query-guided concept activation delivers up to a +9.5 point accuracy boost on complex spatial reasoning without retraining.

Visual token pruning creates a silent vulnerability vector: lightweight perturbations can manipulate token selection to blind compressed vision-language models while leaving uncompressed inference completely intact.

Aligning luminance histograms can transform diffusion models into powerful HDR generators without any retraining, achieving remarkable fidelity and detail.

Scale-Frustum Representations offer a surprisingly effective way to model objects at drastically different scales in ultra-wide area remote sensing images, boosting segmentation accuracy and convergence speed.