MSTG (Multi-Stage Tree Graph)

Overview

Multi-Stage Tree Graph (MSTG) is a hierarchical vector index developed by MyScale that overcomes limitations of traditional IVF (Inverted File Index) through a hierarchical design.

Key Innovation

MSTG creates multiple layers unlike IVF which has a single layer of cluster vectors. This hierarchical approach enables:

• More efficient search traversal
• Better handling of complex vector distributions
• Improved recall at comparable latency

Hierarchical Design

The multi-stage architecture allows for:

• Progressive refinement of search results
• Better balance between accuracy and speed
• More flexible trade-offs between recall and performance

Advantages Over IVF

Single Layer vs Multi-Layer

• IVF: Single layer of cluster vectors
• MSTG: Multiple hierarchical layers for progressive search

Search Efficiency

The tree-based structure enables faster traversal compared to flat IVF indexes, especially for large-scale datasets.

Use Cases

• Large-scale vector search
• Applications requiring high recall
• Systems needing balanced performance/accuracy trade-offs
• Enterprise vector databases

Performance Characteristics

Provides improved performance over traditional IVF while maintaining computational efficiency through intelligent hierarchical organization.

Pricing

Available as part of MyScale vector database.