AI Factory Glossary

skin lesion classification,healthcare ai

Classify skin conditions from photos.

skipnet, model optimization

SkipNet learns to skip layers adaptively for different inputs reducing computation.

sla, sla, supply chain & logistics

Service Level Agreements define expected supplier performance including delivery times defect rates and response requirements.

sliding window attention patterns,llm architecture

Only attend to recent tokens within fixed window.

sliding window attention,transformer

Attention mechanism that only attends to nearby tokens within a fixed window to handle long sequences.

sliding window, llm optimization

Sliding window attention restricts attention to local neighborhoods reducing complexity.

sliding window, time series models

Sliding window forecasting maintains fixed-length training history discarding oldest observations.

slimmable networks, neural architecture

Train single network supporting multiple widths.

slot-based architectures, neural architecture

Represent information in discrete slots.

small language model, llm architecture

Small language models provide efficiency with fewer parameters.

smiles generation, smiles, chemistry ai

Generate molecular SMILES strings.

smoothgrad, explainable ai

Average gradients over noisy inputs.

smote, smote, advanced training

Synthetic Minority Over-sampling Technique generates synthetic examples by interpolating between minority class instances for imbalanced data.

smote, smote, machine learning

Generate synthetic minority examples.

smt-based verification, ai safety

Use SMT solvers to verify properties.

snail, snail, meta-learning

Attention-based meta-learning.

snapshot ensembles,machine learning

Ensemble from single training run.

snapshot graphs, graph neural networks

Snapshot graphs represent temporal networks as sequences of static graphs at discrete time points.

so equivariant, so(3), graph neural networks

SO(3) equivariant networks maintain rotational symmetries in 3D molecular and geometric data.

soft defect, failure analysis advanced

Soft defects are latent failures that manifest intermittently or under specific conditions requiring specialized dynamic fault localization techniques.

soft routing, llm architecture

Soft routing weights expert outputs rather than selecting discrete subset.

softplus, neural architecture

Smooth approximation of ReLU.

software pipelining, model optimization

Software pipelining overlaps iterations from different loop executions improving throughput.

solder joint inspection, failure analysis advanced

Solder joint inspection uses X-ray or cross-sectioning to detect voids cracks and insufficient wetting.

solubility prediction, chemistry ai

Predict compound solubility.

solvent distillation, environmental & sustainability

Solvent distillation separates and purifies organic solvents based on boiling point differences.

solvent recovery, environmental & sustainability

Solvent recovery systems capture and purify organic solvents from process exhaust enabling reuse and reducing disposal costs.

sort pooling, graph neural networks

Sort pooling orders nodes by learned structural roles enabling use of 1D convolutions for graph classification.

sortpool variant, graph neural networks

SortPool variants improve graph classification by learning node importance for sorted feature concatenation.

sound source localization, multimodal ai

Locate sound sources using vision.

source-free domain adaptation, domain adaptation

Adapt without access to source data.

sparse attention,transformer

Attention pattern where each token only attends to a subset of tokens rather than all.

sparse autoencoders for interpretability, explainable ai

Extract interpretable features.

sparse mixture, llm architecture

Sparse mixture of experts activates subset of experts per input.

sparse model,model architecture

Model where only subset of parameters activate (MoE).

sparse training, model optimization

Sparse training maintains sparsity throughout training never materializing dense networks.

sparse transformer patterns, transformer

Different sparsity patterns for attention.

sparse upcycling,model architecture

Convert dense model to MoE by splitting weights into experts.

sparse weight averaging, model optimization

Sparse weight averaging combines multiple sparse models improving generalization.

sparse-to-sparse training,model training

Maintain sparsity throughout training.

spatial attention, model optimization

Spatial attention focuses on informative regions by weighting spatial locations.

specialist agent, ai agents

Specialist agents focus on specific capabilities contributing expertise to team efforts.

specification gaming, ai safety

Specification gaming exploits reward function loopholes achieving objectives harmfully.

specification waiver, production

Authorization to operate out-of-spec.

spectral graph convolutions, graph neural networks

Graph convolutions using spectral methods.

spectral graph theory, graph neural networks

Analyze graphs via eigenvalues.

spectral normalization in gans, generative models

Stabilize GAN training.

spectral normalization, ai safety

Normalize by largest singular value.

spectral normalization, generative models

Normalize weights by spectral norm.

spectral residual, time series models

Spectral residual detects anomalies by computing deviations between original and smoothed frequency spectra in time series saliency maps.