Truong Nguyen

Pixel diffusion’s chronic high-frequency blur is not an architectural bottleneck: a lightweight adversarial post-training phase restores natural spectral power laws without causing mode collapse or requiring changes to the sampling pipeline.

Wavelet transforms in deep learning no longer require trading off learnability for mathematical stability: lattice-structured lifting steps guarantee perfect invertibility and stability under arbitrary network parameter updates.

HumanoidVLN reveals that navigation success in humanoid robots is significantly influenced by physical embodiment, achieving a 43.55% success rate with state-of-the-art models.

Unified Agent outperforms existing multi-agent systems by effectively maintaining a compact state across devices and time, revolutionizing user-agent interactions.

Achieving 90% of oracle performance in bimanual tool use, DexFuture operates 250 times faster than conventional planning methods.

Achieve state-of-the-art image restoration by routing inputs through a dynamically-selected mixture of experts, guided by a spherical embedding of degradation type.

By encoding physical priors into a graph transformer, PhysGraph achieves superior bimanual manipulation with half the parameters and zero-shot generalization to new tools, suggesting learned physics can be replaced with structured inductive biases.

Achieve high diversity in text-to-video generation without sacrificing temporal consistency by operating in latent space, sidestepping the usual backpropagation bottlenecks.

Deeper frequency modulation via multi-level wavelet transforms slashes Gaussian primitive counts in 3D Gaussian Splatting without sacrificing rendering quality.