Results for "high-risk"

Optimization using curvature information; often expensive at scale.

AI subfield dealing with understanding and generating human language, including syntax, semantics, and pragmatics.

Matrix of second derivatives describing local curvature of loss.

Generating speech audio from text, with control over prosody, speaker identity, and style.

Balancing learning new behaviors vs exploiting known rewards.

A theoretical framework analyzing what classes of functions can be learned, how efficiently, and with what guarantees.

Legal or policy requirement to explain AI decisions.

Systematic error introduced by simplifying assumptions in a learning algorithm.

Generates audio waveforms from spectrograms.

Optimizing policies directly via gradient ascent on expected reward.

Maintaining two environments for instant rollback.

Models whose weights are publicly available.

Mathematical foundation for ML involving vector spaces, matrices, and linear transformations.

Vector whose direction remains unchanged under linear transformation.

Sensitivity of a function to input perturbations.

Eliminating variables by integrating over them.

Approximating expectations via random sampling.

Ensuring AI systems pursue intended human goals.

Methods like Adam adjusting learning rates dynamically.

Model exploits poorly specified objectives.

Maximizing reward without fulfilling real goal.

Loss of old knowledge when learning new tasks.

Model relies on irrelevant signals.

Small prompt changes cause large output changes.

Mechanism to disable AI system.

Maximum system processing rate.

Robots made of flexible materials.

Differences between simulated and real physics.

Directly optimizing control policies.

Sampling-based motion planner.