Welcome to IDL Package

IDL (Implicit Deep Learning) is a Python package that implements implicit deep learning models (with a specialized recurrent version) and the state-driven training approach.

Key Features

  • Implicit Modeling with optional low-rank training

  • Special case of Implicit Modeling with recurrent structure

  • State-driven Implicit Modeling (SIM) training approach with multiple ready-to-use solvers

  • Easy to use, seamless integration with Pytorch autograd

Getting Started

Installation

Install using pip:

pip install torchidl

Install from source:

git clone https://github.com/HoangP8/Implicit-Deep-Learning
cd Implicit-Deep-Learning
pip install -e .

Basic Usage

Here’s a simple example using ImplicitModel:

from torchidl import ImplicitModel

# Normal data processing
train_loader, test_loader = ...  # Any dataset users use (e.g., CIFAR10, time-series, ...)

# Define the Implicit Model
model = ImplicitModel(
   hidden_dim=100,  # Size of the hidden dimension
   input_dim=3072,  # Input dimension (e.g., 3*32*32 for CIFAR-10)
   output_dim=10,   # Output dimension (e.g., 10 classes for CIFAR-10)
)

# Normal training loop
optimizer = ...  # Choose optimizer (e.g., Adam, SGD)
loss_fn = ...    # Choose loss function (e.g., Cross-Entropy, MSE)

for _ in range(epoch):
   ...
   optimizer.zero_grad()
   loss = loss_fn(model(inputs), targets)
   loss.backward()
   optimizer.step()
   ...

  • Load your dataset and train as usual, the forward and backward passes are fully packaged.

  • ImplicitModel is simple to use with just a few lines of code.

Sections: