Results for "reward inference"

Maximizing reward without fulfilling real goal.

Modifying reward to accelerate learning.

Reward only given upon task completion.

Methods to protect model/data during inference (e.g., trusted execution environments) from operators/attackers.

A learning paradigm where an agent interacts with an environment and learns to choose actions to maximize cumulative reward.

Reinforcement learning from human feedback: uses preference data to train a reward model and optimize the policy.

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

Optimizing policies directly via gradient ascent on expected reward.

Inferring reward function from observed behavior.

Time from request to response; critical for real-time inference and UX.

Hardware resources used for training/inference; constrained by memory bandwidth, FLOPs, and parallelism.

Reducing numeric precision of weights/activations to speed inference and reduce memory with acceptable accuracy loss.

Prevents attention to future tokens during training/inference.

Variable enabling causal inference despite confounding.

Differences between training and inference conditions.

Limiting inference usage.

Model trained to predict human preferences (or utility) for candidate outputs; used in RLHF-style pipelines.

Framework for reasoning about cause-effect relationships beyond correlation, often using structural assumptions and experiments.

Updating beliefs about parameters using observed evidence and prior distributions.

Model execution path in production.

Running predictions on large datasets periodically.

Low-latency prediction per request.

Cost to run models in production.

Running models locally.

Acting to minimize surprise or free energy.