Results for "real-time"

PEFT method injecting trainable low-rank matrices into layers, enabling efficient fine-tuning.

Search algorithm for generation that keeps top-k partial sequences; can improve likelihood but reduce diversity.

Samples from the k highest-probability tokens to limit unlikely outputs.

Adjusting learning rate over training to improve convergence.

Using same parameters across different parts of a model.

Prevents attention to future tokens during training/inference.

Routes inputs to subsets of parameters for scalable capacity.

Formal framework for sequential decision-making under uncertainty.

Probabilistic model for sequential data with latent states.

Expected cumulative reward from a state or state-action pair.

Diffusion model trained to remove noise step by step.

Optimizing policies directly via gradient ascent on expected reward.

Controls amount of noise added at each diffusion step.

Generative model that learns to reverse a gradual noise process.

Pixel motion estimation between frames.

Generates audio waveforms from spectrograms.

Predicting future values from past observations.

Shift in model outputs.

Organizational uptake of AI technologies.

Differences between training and inference conditions.

Startup latency for services.

Storing results to reduce compute.

Hardware components that execute physical actions.

Algorithm computing control actions.

Classical controller balancing responsiveness and stability.

System returns to equilibrium after disturbance.

Optimal control for linear systems with quadratic cost.

Study of motion without considering forces.

Optimizing continuous action sequences.

Computing collision-free trajectories.