Decide on the models/tasks/datasets to benchmark this technique

[simi2525]

We should decide on what to evaluate the effectiveness of std loss regularization.

After a short online search I found a number of benchmarks that seem to be designed for this purpose:

• https://dawn.cs.stanford.edu/benchmark/#cifar10-inference-cost
• https://github.com/fsschneider/DeepOBS
• http://tbd-suite.ai/
• https://mlperf.org/

(this can also help us get some more references in the article, since sadly people care about that)
We can use any of these and we could also use more standard ways of testing.

Since we don't have access to much compute power or funding, we should select as few benchmarks as possible that would still produce results as convincing as possible.

Potential models:

• VGG11
• VGG16
• VGG19
• ResNet101
• ResNeXt
• LeNet
• AlexNet
• Simple MLP

Potential datasets:

• CIFAR10
• CIFAR100
• ImageNet 1k
• COCO
• Sentiment140
• Open AI Gym games

	CIFAR10	CIFAR100	ImageNet 1k	COCO	Sentiment140	Open AI Gym games
VGG11	result	result	result	result	result	result
VGG16	result	result	result	result	result	result
VGG19	result	result	result	result	result	result
ResNet101	result	result	result	result	result	result
ResNeXt	result	result	result	result	result	result
LeNet	result	result	result	result	result	result
AlexNet	result	result	result	result	result	result
Simple MLP	result	result	result	result	result	result

*I suggest we use a table like this and put in the link to the folder/file in the repository where we stored the results of the experiment

A list of popular datasets I found:
https://medium.com/towards-artificial-intelligence/the-50-best-public-datasets-for-machine-learning-d80e9f030279