Research note
Tensor Network Classifiers for Image Classification
Exploring compact tensor-network representations for supervised learning and interpretable model structure.
Research question
Can tensor network architectures provide useful inductive bias for small-scale image classification while remaining interpretable and computationally controlled?
Motivation
Tensor networks offer a useful bridge between physical structure, compressed representations, and machine learning models.
Methods
Matrix product states Tensor train decompositions Feature maps Supervised classification experiments
Approach
- Compare tensor-network classifiers against small neural baselines.
- Track how bond dimension affects capacity, stability, and training cost.
- Document feature encoding choices and their effect on classification behavior.
Next steps
- Publish reproducible notebooks with controlled toy experiments.
- Write a short note connecting tensor-network compression to practical ML model design.
Current results
Current results are being organized into reproducible notes and experiment summaries.
Tools
Python PyTorch NumPy Tensor network methods
Status: Active research notes