liveBook · Manning

Low-rank subspace

As explained in Applied Reinforcement Learning

A low-rank subspace is a restricted set of directions whose dimensionality is smaller than that of the full space being represented.

A low-rank subspace captures changes using fewer independent directions than the full space permits. In fine-tuning, weight updates may have this structure, so LoRA represents them with two smaller matrices rather than changing the original weights. section 11.3

A compact adapter

For a 4096 × 4096 weight matrix, a rank-16 LoRA adapter adds 131,072 parameters, less than 1% of the original matrix’s 16.8 million. section 11.3

Figure 11.3 LoRA architecture for a single linear layer.

Use the single-layer diagram to see how LoRA adds a low-rank adapter path alongside frozen weights. section 11.3

How Applied Reinforcement Learning covers it

How deep it goes
Defines rank as independent directions and explains representing updates with two smaller matrices.
Examples
Uses repeated row patterns as intuition and compares rank-16 adapters with a 4096 × 4096 weight matrix.
What you can do after
Understand how low-rank updates can motivate smaller trainable adapters during model fine-tuning.
How it connects
Connects matrix rank and weight updates to LoRA adapters and fine-tuning.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Applied Reinforcement Learning. The section links point to the parts of the book it draws on.