5 papers
Waypoint-1.5: A Real-Time Video World Model for Consumer Hardware
Rajit Rajpal, Shahbuland Matiana, Liew Wei Pyn +13
We present Waypoint 1.5, a real-time diffusion world model for interactive video generation on consumer-grade hardware. Unlike general video diffusion models, interactive world mod…
Decoupling Appearance and Geometry in Diffusion World Models
Rajit Rajpal
We study the problem of learning diffusion-based world models that jointly capture scene geometry and visual appearance in complex first-person environments. Existing approaches ty…
Low-Rank Friction for Memory-Efficient Transformer Pretraining
Rajit Rajpal, Benedict Leimkuhler
iKFAD is a recently proposed optimiser that replaces adaptive learning rates with adaptive friction in the momentum dynamics, yet performs as well as Adam. Its limitation is that t…
Adaptive Momentum and Nonlinear Damping for Neural Network Training
Aikaterini Karoni, Rajit Rajpal, Benedict Leimkuhler +1
Momentum Stochastic Gradient Descent (mSGD) relies on a fixed momentum coefficient shared across all parameters, failing to account for the heterogeneous structure of modern loss l…
Adaptive Stepsizing for Stochastic Gradient Langevin Dynamics in Bayesian Neural Networks
Rajit Rajpal, Benedict Leimkuhler, Yuanhao Jiang
Bayesian neural networks (BNNs) require scalable sampling algorithms to approximate posterior distributions over parameters. Existing stochastic gradient Markov Chain Monte Carlo (…