What it is
Model distillation is a machine learning technique used to create more efficient and lightweight AI models. In this process, a smaller, typically faster model (the "student") is trained to reproduce the output and internal representations of a larger, more powerful "teacher" model. The student model learns from the teacher's soft targets (e.g., probability distributions for classifications) rather than just the hard labels, allowing it to achieve comparable performance with significantly fewer parameters and lower computational requirements.
Model distillation is increasingly important for deploying AI models on edge devices, in resource-constrained environments, or for real-time applications where large models are impractical. It enables companies to reduce inference costs and latency, making advanced AI more accessible and scalable. This technique is crucial for optimizing enterprise AI solutions and mobile applications, as it allows for the deployment of sophisticated AI capabilities without the need for extensive compute resources, impacting product development and market adoption.
Why it matters
Distillation makes AI models smaller and faster, enabling them to run on more devices and reduce costs, leading to wider AI adoption and new applications.
Reviewed under editorial standardsUpdated September 26, 2026Not investment advice