Results for "transfer problem"

Learning physical parameters from data.

Deep learning system for protein structure prediction.

Differences between training and deployed patient populations.

Achieving task performance by providing a small number of examples inside the prompt without weight updates.

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

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

Automatically learning useful internal features (latent variables) that capture salient structure for downstream tasks.

Reconstructing a model or its capabilities via API queries or leaked artifacts.

Nonlinear functions enabling networks to approximate complex mappings; ReLU variants dominate modern DL.

Limiting gradient magnitude to prevent exploding gradients.

An RNN variant using gates to mitigate vanishing gradients and capture longer context.

Allows gradients to bypass layers, enabling very deep networks.

Removing weights or neurons to shrink models and improve efficiency; can be structured or unstructured.

Tradeoffs between many layers vs many neurons per layer.

Assigning labels per pixel (semantic) or per instance (instance segmentation) to map object boundaries.

Capabilities that appear only beyond certain model sizes.

Systematic error introduced by simplifying assumptions in a learning algorithm.

All possible configurations an agent may encounter.

Estimating parameters by maximizing likelihood of observed data.

Set of all actions available to the agent.

Simultaneous Localization and Mapping for robotics.

Recovering 3D structure from images.

Running predictions on large datasets periodically.

Interleaving reasoning and tool use.

Agent reasoning about future outcomes.

Agents communicate via shared state.

Eliminating variables by integrating over them.

Restricting updates to safe regions.

Model exploits poorly specified objectives.

Correctly specifying goals.