# What is Model Pruning? Definition & Technical Mechanics

> **Source:** [SPIDITS AI AI Glossary](https://spidits.com/ai-glossary)  
> **Category:** Model Optimization  
> **Canonical Citation:** [https://spidits.com/ai-glossary/model-pruning](https://spidits.com/ai-glossary/model-pruning)

## Definition
Model Pruning is a model compression technique that removes non-essential weights or neurons from a trained network. By zeroing out parameters that have minimal impact on output predictions, it reduces model file sizes and execution latency.

## Architectural Mechanics & Deep Dive
Model Pruning is a parameter compression method that deletes non-essential connections or layers in a trained neural network. By identifying weights that have minimal impact on output activations and setting them to zero, pruning reduces model size and accelerates inference speeds, making models more suitable for edge-device serving.

## Industry Applications & Real-World Use Cases
Edge device AI deployment, model footprint reduction, and faster inference execution

## Frequently Asked Questions

### What is structured vs. unstructured pruning?
Unstructured pruning removes individual weights anywhere in the network. Structured pruning removes entire channels or layers, which is easier for standard hardware to accelerate.

### Does pruning degrade accuracy?
A slight drop can occur, but it is typically recovered by running short recovery fine-tuning epochs after pruning.

## Related Terminology
- [distillation](https://spidits.com/ai-glossary/distillation)
- [quantization](https://spidits.com/ai-glossary/quantization)
- [models](https://spidits.com/ai-glossary/models)

---
*From the SPIDITS AI Knowledge Base (300+ terms indexed). Explore full technical definitions: [https://spidits.com/ai-glossary](https://spidits.com/ai-glossary)*