Results for "distribution distance"

A mismatch between training and deployment data distributions that can degrade model performance.

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

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

Converts logits to probabilities by exponentiation and normalization; common in classification and LMs.

A measure of randomness or uncertainty in a probability distribution.

Measures how much information an observable random variable carries about unknown parameters.

Models that define an energy landscape rather than explicit probabilities.

Probabilistic model for sequential data with latent states.

Generative model that learns to reverse a gradual noise process.

Controls amount of noise added at each diffusion step.

Maintaining alignment under new conditions.

Performance drop when moving from simulation to reality.

Learning policies from expert demonstrations.

Restricting distribution of powerful models.

Learning where data arrives sequentially and the model updates continuously, often under changing distributions.

Methods that learn training procedures or initializations so models can adapt quickly to new tasks with little data.

The relationship between inputs and outputs changes over time, requiring monitoring and model updates.

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

Techniques that discourage overly complex solutions to improve generalization (reduce overfitting).

Harmonic mean of precision and recall; useful when balancing false positives/negatives matters.

Scalar summary of ROC; measures ranking ability, not calibration.

Methods to set starting weights to preserve signal/gradient scales across layers.

A model that assigns probabilities to sequences of tokens; often trained by next-token prediction.

Feature attribution method grounded in cooperative game theory for explaining predictions in tabular settings.

Model-generated content that is fluent but unsupported by evidence or incorrect; mitigated by grounding and verification.

When some classes are rare, requiring reweighting, resampling, or specialized metrics.

Expanding training data via transformations (flips, noise, paraphrases) to improve robustness.

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

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

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