AI Factory Glossary

killer defect,production

Defect large or critical enough to cause chip failure.

kinetic monte carlo, simulation

Simulate time evolution of processes.

kink effect,device physics

Sudden change in I-V curve due to floating body.

kirkendall voids, failure analysis advanced

Kirkendall voids form at bond interfaces from differential diffusion rates weakening interconnections.

knn-lm (k-nearest neighbor language model),knn-lm,k-nearest neighbor language model,llm architecture

Augment LM with nearest neighbor retrieval.

knn,nearest neighbor,instance

K-nearest neighbors classifies by proximity. No training, just lookup.

knowledge distillation for edge, edge ai

Distill large model to small for edge.

knowledge distillation variants, model compression

Different distillation approaches.

knowledge distillation, model optimization

Knowledge distillation trains compact student models to mimic larger teacher model predictions.

knowledge distillation,model optimization

Train small student model to mimic large teacher model.

knowledge editing, model editing

Modify specific facts in models.

knowledge editing,model training

Update model's factual knowledge without full retraining.

knowledge extraction attacks, privacy

Extract knowledge without full replication.

knowledge freshness, rag

How up-to-date information is.

knowledge graph embedding, graph neural networks

Knowledge graph embeddings map entities and relations to vectors for link prediction and reasoning.

knowledge graph embedding,rag

Represent entities and relations as vectors.

knowledge graph embeddings (advanced),knowledge graph embeddings,advanced,graph neural networks

Learn representations of KG entities and relations.

knowledge graph rec, recommendation systems

Knowledge graph-based recommendations leverage entity relationships for enhanced item understanding and cold-start handling.

knowledge graph to text,nlp

Verbalize knowledge graphs.

knowledge graph,kg,entity,relation

Knowledge graphs store entities and relations. Query for structured facts to augment LLM with precise knowledge.

knowledge graph,ontology,semantic

Knowledge graphs represent facts as nodes and edges. Enable reasoning. Combine with LLMs for grounding.

knowledge localization, explainable ai

Find where knowledge is stored.

knowledge masking, nlp

Mask factual knowledge during training.

knowledge neuron, interpretability

Knowledge neurons are individual units encoding specific factual knowledge identifiable through causal intervention.

knowledge neurons, explainable ai

Neurons storing factual knowledge.

knowledge retention techniques, continual learning

Preserve learned information.

known good die (kgd),known good die,kgd,testing

Dies tested and verified before packaging.

known good die for chiplets, kgd, advanced packaging

Pre-tested dies for multi-die packages.

koala,berkeley,dialog

Koala is Berkeley dialog model. Web-scraped conversations.

koboldcpp,roleplay,local

KoboldCpp runs LLMs for roleplay and writing. Built on llama.cpp.

koh etch,etch

Potassium hydroxide for anisotropic silicon etching along crystal planes.

kolmogorov-arnold networks (kan),kolmogorov-arnold networks,kan,neural architecture

Networks using learnable activation functions on edges.

koopman operator theory, control theory

Linear representation of nonlinear dynamics.

kosmos,multimodal ai

Multimodal large language model from Microsoft.

krf (krypton fluoride),krf,krypton fluoride,lithography

248nm DUV light source.

krippendorff's alpha, evaluation

Krippendorff's alpha measures agreement accounting for chance and data types.

krippendorff's alpha,evaluation

Reliability coefficient for multiple annotators.

krum aggregation, federated learning

Select closest updates.

kruskal-wallis, quality & reliability

Kruskal-Wallis test compares distributions across multiple groups.

kserve,kubernetes,inference

KServe is Kubernetes inference platform. Serverless, autoscaling.

kubeflow, infrastructure

Kubernetes-based ML workflow platform.

kubeflow,kubernetes,ml

Kubeflow runs ML workflows on Kubernetes. Pipelines, training, serving. Google origin.

kubernetes for ml,infrastructure

Use K8s to deploy and manage ML workloads.

kv cache eviction, kv, optimization

Remove old entries from cache.

kv cache management,optimization

Efficiently store and reuse key-value pairs.

kv cache optimization, kv, optimization

Efficiently manage key-value cache.

kv cache quantization, kv, optimization

Reduce KV cache memory.

kv cache, kv, llm optimization

KV cache stores key-value attention states avoiding recomputation during generation.

kv cache,key value,memory

KV cache stores computed keys and values for past tokens. Grows with sequence length. Major memory consumer.

kv cache,llm architecture

Cached key-value pairs from previous tokens to speed up autoregressive generation.