Deep Tensor Graphical Model for Sparse Graph Recovery
This repository contains the preliminary implementation of DETECT,
a deep tensor graphical model designed for sparse graph estimation
and high-dimensional precision matrix learning.
The current release focuses on the algorithmic core and code structure.
A full version including theoretical background, benchmark experiments,
and tensor-based extensions will be released in a future update.
DETECT introduces a deep unrolled optimization framework that learns
the sparse structure of conditional dependencies among variables.
It generalizes classical graphical model estimation to tensor-valued settings
by embedding the optimization process into a differentiable neural architecture.
Key features include:
- Tensor-aware modeling: preserves high-order structural dependencies.
- Adaptive optimization: learns hyperparameters dynamically through backpropagation.
- Unrolled Alternating Minimization: interpretable iteration-by-iteration updates.
- Differentiable regularization: integrates soft-thresholding within neural layers.
- GPU-accelerated training: implemented in PyTorch for scalability.