AI Factory Glossary

gradient accumulation,model training

Accumulate gradients over multiple mini-batches before updating weights.

gradient bucketing, distributed training

Group gradients for efficient communication.

gradient clipping, training techniques

Gradient clipping bounds gradient norms preventing privacy leakage and training instability.

gradient clipping,model training

Cap gradient magnitude to prevent exploding gradients.

gradient compression techniques, distributed training

Reduce gradient communication.

gradient flow preservation,model training

Maintain gradient flow in sparse networks.

gradient masking, ai safety

Make gradients uninformative.

gradient penalty, generative models

Regularize gradient magnitude (GANs).

gradient quantization for communication, distributed training

Quantize gradients for transmission.

gradient reversal layer, domain adaptation

Reverse gradients for adversarial training.

gradient synchronization, distributed training

Aggregate gradients across devices.

gradient-based nas, neural architecture

Optimize architecture with gradients.

gradient-based pruning, model optimization

Gradient-based pruning estimates weight importance using gradient information.

gradient-based pruning,model optimization

Use gradients to determine importance.

grain boundaries, defects

Interfaces between crystallites.

grain boundary characterization, metrology

Analyze grain boundary structure and energy.

grain boundary energy, defects

Energy per area of boundary.

grain boundary segregation, defects

Impurities collect at boundaries.

grain growth in copper,beol

Increase grain size to reduce resistance.

grammar-based decoding, llm optimization

Grammar-based decoding generates text following formal grammar specifications.

grammar-based generation, graph neural networks

Grammar-based generation uses formal grammars to ensure syntactic validity of generated graphs.

gran, gran, graph neural networks

Graph Recurrent Attention Networks generate graphs through sequential block-wise generation with recurrent state tracking for scalability.

granger causality, time series models

Granger causality tests whether past values of one time series provide statistically significant information for predicting another series.

granger non-causality, time series models

Granger non-causality tests null hypothesis that past values of one series don't help predict another.

graph attention networks (gat),graph attention networks,gat,graph neural networks

Use attention in GNNs.

graph completion, graph neural networks

Graph completion predicts missing nodes or edges in incomplete graphs for knowledge graph construction.

graph convolution, graph neural networks

Graph convolution generalizes convolutional operations to irregular graph structures by aggregating features from neighboring nodes with learnable weights.

graph convolutional networks (gcn),graph convolutional networks,gcn,graph neural networks

Convolutional operations on graphs.

graph generation, graph neural networks

Generate new graphs.

graph isomorphism network (gin),graph isomorphism network,gin,graph neural networks

Most expressive message-passing GNN.

graph laplacian, graph neural networks

Operator encoding graph structure.

graph neural network,gnn,node

GNNs operate on graph-structured data. Message passing between nodes. Social networks, molecules.

graph neural odes, graph neural networks

Apply Neural ODEs to graph-structured data.

graph neural operators,graph neural networks

Operators on graph-structured data.

graph optimization, model optimization

Graph optimization transforms computation graphs improving efficiency through fusion reordering and elimination.

graph pooling, graph neural networks

Downsample graphs in GNNs.

graph recurrence, graph neural networks

Graph recurrence applies RNNs to sequences of graph snapshots learning temporal graph dynamics.

graph serialization, model optimization

Graph serialization stores model structure and parameters in portable format for deployment.

graph u-net, graph neural networks

Graph U-Net applies encoder-decoder architecture with skip connections to graphs for node classification.

graph unpooling, graph neural networks

Upsample graphs.

graph vae, graph neural networks

Variational autoencoder for graphs.

graph wavelets, graph neural networks

Wavelet transforms on graphs.

graph-based relational reasoning, graph neural networks

Use graph networks for relational tasks.

graphaf, graph neural networks

Graph Autoregressive Flow generates graphs by sequentially adding nodes and edges with normalizing flows.

graphgen, graph neural networks

Sequential graph generation.

graphnvp, graph neural networks

GraphNVP applies normalizing flows to graph generation enabling exact likelihood computation and efficient sampling of molecular structures.

graphrnn, graph neural networks

GraphRNN generates graphs sequentially by modeling node and edge formation as a sequence generation problem using recurrent neural networks.

graphrnn, graph neural networks

RNN-based graph generation.

graphsage, graph neural networks

GraphSAGE generates node embeddings by sampling and aggregating features from local neighborhoods enabling inductive learning on unseen graphs.

graphsage,graph neural networks

Inductive GNN with sampling.