Mixture of Experts
A neural network architecture that routes inputs to specialised sub-networks (experts), improving efficiency by only activating relevant parts of the model for each input.
In plain language
An AI design where different specialised sub-models handle different types of questions. Instead of one giant brain processing everything, it's a team of experts who activate based on the topic.
Why this matters
Mixture-of-experts architectures offer efficiency gains and can be easier to audit than monolithic models. Understanding model architecture is important for your governance framework when assessing performance, explainability and failure modes.
Relevance
ImplementationTechnical understanding of model architecture supports better governance decisions around testing, monitoring and risk assessment.
Related terms
Putting mixture of experts into practice in your organisation?
Ready to transform your AI strategy?
Partner with Australia's AI strategy and governance specialists. From adoption roadmaps to ISO 42001 audit readiness.