AI Factory Glossary

multi-line code completion, code ai

Complete multiple lines at once.

multi-node training, distributed training

Distributed across machines.

multi-objective nas, neural architecture

Optimize accuracy latency and size together.

multi-query attention (mqa),multi-query attention,mqa,llm architecture

Share key/value across heads to reduce memory and speed up inference.

multi-resolution hash, multimodal ai

Multi-resolution hash encoding stores features at multiple scales in hash tables.

multi-resolution training, computer vision

Train on multiple resolutions.

multi-scale discriminator, generative models

Discriminators at different resolutions.

multi-scale generation, multimodal ai

Multi-scale generation produces images at multiple resolutions simultaneously or progressively.

multi-source domain adaptation,transfer learning

Adapt from multiple source domains.

multi-stage moderation, ai safety

Multiple checks at different points.

multi-step jailbreak,ai safety

Use multiple steps to gradually bypass restrictions.

multi-step jailbreaks, ai safety

Gradually elicit harmful behavior.

multi-style training, audio & speech

Multi-style training uses diverse acoustic conditions during ASR training for robustness.

multi-target domain adaptation, domain adaptation

Adapt to multiple target domains.

multi-task pre-training, foundation model

Pre-train on multiple objectives simultaneously.

multi-task training, multi-task learning

Train on multiple tasks together.

multi-teacher distillation, model compression

Learn from multiple teacher models.

multi-tenancy in training, infrastructure

Share resources across teams.

multi-token prediction, llm optimization

Multi-token prediction forecasts several future tokens enabling faster generation.

multi-view learning, advanced training

Multi-view learning leverages multiple representations or modalities of data to improve model robustness and performance.

multi-view learning, machine learning

Learn from different views of data.

multilingual model, llm architecture

Multilingual models handle multiple languages through diverse training data.

multilingual neural mt, nlp

Single model for many language pairs.

multilingual pre-training, nlp

Pre-train on many languages together.

multimodal bottleneck, multimodal ai

Compress multimodal information.

multimodal chain-of-thought,multimodal ai

Reasoning across modalities.

multimodal fusion strategies, multimodal ai

Combine information from modalities.

multimodal sentiment,multimodal ai

Combine text audio and visual cues.

multimodal transformer av, audio & speech

Multimodal transformers process audio and visual sequences with cross-modal attention mechanisms.

multimodal translation, multimodal ai

Translate between modalities.

multinomial diffusion, generative models

Diffusion over categorical distributions.

multitask instruction, training techniques

Multitask instruction learning trains on diverse tasks simultaneously improving generalization.

multivariate tpp, time series models

Multivariate temporal point processes model interdependent event sequences across multiple event types with cross-excitation.

murphy yield model, yield enhancement

Murphy yield model incorporates defect clustering through alpha parameter representing degree of spatial correlation.

muse, multimodal ai

Muse uses masked generative transformers for parallel image generation from text.

music transformer, audio & speech

Music Transformer applies relative positional encoding to transformers enabling generation of long expressive musical sequences.

mutual learning, model compression

Multiple models learn from each other.

mutually exciting, time series models

Mutually exciting point processes model how events of one type trigger events of other types.