← Back to AI Factory Chat

AI Factory Glossary

751 technical terms and definitions

A B C D E F G H I J K L M N O P Q R S T U V W X Y Z All
Showing page 10 of 16 (751 entries)

model discrimination design, doe

Distinguish between competing models.

model distillation for interpretability, explainable ai

Distill into interpretable student model.

model editing,model training

Directly update model weights to fix specific factual errors or behaviors.

model ensemble rl, reinforcement learning advanced

Model ensemble reinforcement learning trains multiple dynamics models to quantify uncertainty and improve decision-making robustness.

model evaluation, evaluation

Model evaluation measures performance across metrics and test sets.

model extraction attack,ai safety

Steal model by querying it repeatedly.

model extraction, interpretability

Model extraction attacks replicate model functionality through black-box querying.

model extraction,stealing,query

Model extraction steals model via queries. Build substitute model. Protect with rate limiting.

model fingerprint,unique,identify

Model fingerprints identify models from behavior. Unique responses to probe inputs.

model flops utilization, mfu, optimization

Effective compute utilization.

model hub,huggingface,weights

Hugging Face Hub hosts open models and datasets. Download weights, run locally, fine-tune, share.

model inversion attack,ai safety

Reconstruct training data from model parameters or outputs.

model inversion attacks, privacy

Reconstruct training data from model.

model inversion defense,privacy

Prevent reconstruction of training data.

model inversion, interpretability

Model inversion attacks reconstruct training data from model parameters or outputs.

model merging,model training

Combine weights from multiple fine-tuned models to get benefits of both.

model monitoring,mlops

Track model performance metrics and detect degradation.

model parallelism strategies,distributed training

Techniques to split models across GPUs (tensor pipeline expert).

model parallelism,model training

Split model layers across devices each device has subset of parameters.

model predictive control in semiconductor, process control

Use predictive models for process control.

model predictive control, manufacturing operations

Model predictive control optimizes future actions using process models and constraints.

model predictive control, mpc, control theory

Optimize actions using predictive model.

model registry,mlops

Central repository for storing and versioning trained models.

model registry,version,deploy

Model registry versions and stages models. MLflow, W&B, SageMaker.

model retraining,mlops

Periodically retrain model on fresh data to maintain performance.

model routing, llm optimization

Model routing directs requests to appropriate models based on query characteristics.

model server,serving,runtime

Model servers (vLLM, TGI, Triton) host models for inference. Handle batching, scaling, API.

model serving platform,infrastructure

Infrastructure for deploying models (Seldon KServe BentoML).

model serving,deployment

Infrastructure to deploy models and handle inference requests.

model size,model training

Disk space required to store model weights.

model soup, model merging

Average fine-tuned models.

model stealing, privacy

Replicate model by querying.

model stitching for understanding, explainable ai

Connect different model parts.

model stitching, model merging

Connect different model parts.

model theft,extraction,protect

Model extraction attacks steal model via API queries. Protect with rate limits, output perturbation, watermarks.

model verification, security

Verify model hasn't been tampered with.

model versioning,mlops

Track different versions of trained models.

model watermarking,ai safety

Embed secret signals in model to prove ownership or detect unauthorized use.

model-agnostic meta-learning for rl, meta-learning

MAML applied to RL.

model-based ocd, metrology

Fit geometric model to optical data.

model-based reinforcement learning, reinforcement learning

Learn environment model to improve sample efficiency.

moderation api, ai safety

OpenAI's content moderation.

modern hopfield networks,neural architecture

Continuous-valued Hopfield networks equivalent to attention.

modified control charts, spc

Adapt traditional charts.

modular networks,neural architecture

Networks built from reusable modules.

modular neural networks, neural architecture

Networks composed of specialized modules.

modularity maximization, graph algorithms

Optimize community structure quality.

moe communication costs, moe

All-to-all communication overhead.

moe,mixture of experts,experts

Mixture-of-Experts (MoE) models route each token through a few experts instead of all layers. This yields very large capacity at lower compute per token.

moisture absorption in emc, packaging

EMC absorbing water.