Results for "restricted sampling"

Stochastic generation strategies that trade determinism for diversity; key knobs include temperature and nucleus sampling.

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

Samples from the smallest set of tokens whose probabilities sum to p, adapting set size by context.

Simplified Boltzmann Machine with bipartite structure.

Sampling from easier distribution with reweighting.

Approximating expectations via random sampling.

Probabilistic energy-based neural network with hidden variables.

Isolating AI systems.

Sampling-based motion planner.

Selecting the most informative samples to label (e.g., uncertainty sampling) to reduce labeling cost.

Scales logits before sampling; higher increases randomness/diversity, lower increases determinism.

Sampling multiple outputs and selecting consensus.

Learning from data generated by a different policy.

The internal space where learned representations live; operations here often correlate with semantics or generative factors.

Systematic differences in model outcomes across groups; arises from data, labels, and deployment context.

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

Artificially created data used to train/test models; helpful for privacy and coverage, risky if unrealistic.

Raw model outputs before converting to probabilities; manipulated during decoding and calibration.

Structured dataset documentation covering collection, composition, recommended uses, biases, and maintenance.

Variability introduced by minibatch sampling during SGD.

Balancing learning new behaviors vs exploiting known rewards.

Models that learn to generate samples resembling training data.

Generative model that learns to reverse a gradual noise process.

Learns the score (∇ log p(x)) for generative sampling.

Autoencoder using probabilistic latent variables and KL regularization.

Exact likelihood generative models using invertible transforms.

Variable whose values depend on chance.

Randomizing simulation parameters to improve real-world transfer.

Space of all possible robot configurations.

Unequal performance across demographic groups.