AngularGrad Optimizer

This repository contains the oficial implementation for AngularGrad: A New Optimization Technique for Angular Convergence of Convolutional Neural Networks in PyTorch.

AngularGrad reduces the zig-zag effect in the optimization trajectory. Fluctuations are significantly smoothed, tracing a more direct path towards the minimum of the cost function.

You can import the optimizer as follows:

from myoptims.tanangulargrad import tanangulargrad
from myoptims.cosangulargrad import cosangulargrad
...
model = YourModel()
optimizer = tanangulargrad(model.parameters())
...
for input, output in data:
  optimizer.zero_grad()
  loss = loss_function(output, model(input))
  loss.backward()
  optimizer.step()
...

If you have questions or suggestions, please feel free to open an issue. Please cite as:

@article{roy2021angulargrad,
  title={AngularGrad: A New Optimization Technique for Angular Convergence of Convolutional Neural Networks},
  author={S.K. Roy, M.E. Paoletti, J.M. Haut, S.R. Dubey, P. Kar, A. Plaza and B.B. Chaudhuri},
  journal={arXiv preprint arXiv:2105.10190},
  year={2021}
}

Experiments

Experiments in the paper:

Analitycal

cd analitycal/
python main.py

CIFAR-10/100

cd cifar/
python main.py --dataset <cifar10/cifar100> --model <r18/r34/.../vgg16/d121> --alg <adam/sgd/.../cosangulargrad/tanangulargrad> --lr <float>
Example:
python main.py --dataset cifar10 --model r50 --alg cosangulargrad --lr 1e-3

Mini-ImageNet:

cd mini-imagenet/
wget URL dataset
python main.py DATADIR --alg <adam/sgd/.../cosangulargrad/tanangulargrad> --lr <float>
Example:
python main.py ./split_mini/ --alg cosangulargrad --model r50 --lr 1e-3

Fine-Grained:

cd fine-grained/
wget URL datasets
python main.py DATADIR --dataset <dsetname> --alg <adam/sgd/.../cosangulargrad/tanangulargrad> --lr <float>
Example:
python main.py ./data/Car196/ --dataset cars --alg adam --lr 1e-3