AI Factory Glossary

u-net denoiser, generative models

Architecture for diffusion models.

ulpa filter (ultra-low particulate air),ulpa filter,ultra-low particulate air,facility

Even more efficient than HEPA removes 99.999% of particles.

ultimate sd upscale, generative models

Tiled upscaling method.

umbrella sampling, chemistry ai

Sample along reaction coordinate.

uncertainty budget, metrology

Breakdown of uncertainty sources.

uncertainty quantification, ai safety

Uncertainty quantification estimates model confidence in predictions.

uncertainty quantification,ai safety

Estimate confidence in predictions.

uncertainty-based rejection,ai safety

Abstain based on uncertainty estimates.

uncertainty,confidence,epistemic

Uncertainty quantification distinguishes epistemic (model) from aleatoric (data) uncertainty. Know when to abstain.

under-sampling majority class, machine learning

Remove common examples.

undertraining, training

Insufficient training.

unidirectional attention,transformer

Each token only attends to previous tokens (GPT-style).

unified vision-language models,multimodal ai

Models handling multiple vision-language tasks.

unipc sampling, generative models

Unified predictor-corrector sampler.

universal adversarial triggers,ai safety

Inputs that reliably cause unwanted behavior.

universal domain adaptation, domain adaptation

Handle target classes not in source.

universal transformers,llm architecture

Recurrent depth with adaptive computation.

universally slimmable networks, neural architecture

Switch between widths seamlessly.

unlearning,ai safety

Remove specific knowledge or capabilities from a trained model.

unobserved components, time series models

Unobserved components models represent time series as sums of latent stochastic components like trends cycles and seasonal patterns.

unplanned maintenance, production

Emergency repairs.

unscented kalman, time series models

Unscented Kalman Filter propagates uncertainty through nonlinear transformations using deterministic sampling.

unscheduled maintenance, manufacturing operations

Unscheduled maintenance responds to unexpected equipment problems.

unstructured pruning, model optimization

Unstructured pruning removes individual weights creating sparse networks.

unstructured pruning,model optimization

Remove individual weights creating sparse tensors.

unsupervised domain adaptation,transfer learning

Adapt without target labels.

up-sampling, training

Use data more than once.

update functions, graph neural networks

Update functions in GNNs combine aggregated neighbor information with node features to compute new representations.

upscaling techniques, generative models

Increase image resolution.

upw, upw, environmental & sustainability

Ultra-Pure Water exceeds deionized water specifications with sub-ppb impurity levels and filtered particles for critical semiconductor processes.

usage-based maintenance, production

Maintain after set amount of use.

uv disinfection, uv, environmental & sustainability

UV disinfection uses ultraviolet light to destroy microorganisms in water without chemical addition.

uv mapping, uv, multimodal ai

UV mapping parameterizes 3D surfaces for texture coordinate assignment.