Results for "data → model"

Probabilistic energy-based neural network with hidden variables.

Maintaining alignment under new conditions.

High-fidelity virtual model of a physical system.

Architecture that retrieves relevant documents (e.g., from a vector DB) and conditions generation on them to reduce hallucinations.

A preference-based training method optimizing policies directly from pairwise comparisons without explicit RL loops.

Exponential of average negative log-likelihood; lower means better predictive fit, not necessarily better utility.

System design where humans validate or guide model outputs, especially for high-stakes decisions.

Allows model to attend to information from different subspaces simultaneously.

Models trained to decide when to call tools.

Logged record of model inputs, outputs, and decisions.

Controls amount of noise added at each diffusion step.

Increasing model capacity via compute.

Prompt augmented with retrieved documents.

A structured collection of examples used to train/evaluate models; quality, bias, and coverage often dominate outcomes.

Measure of consistency across labelers; low agreement indicates ambiguous tasks or poor guidelines.

Learning only from current policy’s data.

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

Neural networks that operate on graph-structured data by propagating information along edges.

Two-network setup where generator fools a discriminator.

CNNs applied to time series.

Belief before observing data.

Storing results to reduce compute.

Software pipeline converting raw sensor data into structured representations.

Models estimating recidivism risk.

A function measuring prediction error (and sometimes calibration), guiding gradient-based optimization.

Maximum number of tokens the model can attend to in one forward pass; constrains long-document reasoning.

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

Constraining outputs to retrieved or provided sources, often with citation, to improve factual reliability.

Ensuring model behavior matches human goals, norms, and constraints, including reducing harmful or deceptive outputs.

The shape of the loss function over parameter space.