DemonRangerOptimizer
imagenette
DemonRangerOptimizer | imagenette | |
---|---|---|
1 | 9 | |
23 | 877 | |
- | 0.0% | |
0.0 | 0.0 | |
over 3 years ago | over 1 year ago | |
Python | Jupyter Notebook | |
- | Apache License 2.0 |
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
DemonRangerOptimizer
-
[R] AdasOptimizer Update: Cifar-100+MobileNetV2 Adas generalizes with Adas 15% better and 9x faster than Adam
The results are interesting, but in terms of novelty of the main theory - isn't it almost identical to Baydin et al.? https://arxiv.org/pdf/1703.04782.pdf It seems the difference may be in some implementation details, like using a running average for the past gradient. If it's useful, I implemented a bunch of optimizers with options to synergize different techniques (https://github.com/JRC1995/DemonRangerOptimizer) including hypergradient updates for stuffs (and taking into account decorrelated weight decay and per-parameter lrs for hypergradient lr) when I was bored before practically abandoning it all together. I didn't really run any experiments with it though, but some people tried although they may not have got any particularly striking results.
imagenette
-
[P] Graph path traversal with semantic graphs
This idea isn't exclusive to text, the same can be done for images. See this example from the imagenette dataset.
-
Ask HN: In 2022, what is the proper way to get into machine/deep learning?
FastAI has ready-to-run code that does just this. They seem to have an ImageNet package https://github.com/fastai/imagenette
-
How can I download ImageNet dataset with only 20 or 30 classes?
Imagenette is a smaller subset with only 10 classes. https://github.com/fastai/imagenette
You can try this https://github.com/fastai/imagenette which is subset of the main dataset.
-
[D] How can I download ImageNet dataset with only 20 or 30 classes?
Download entire imagenet and annotations. Filter out all annotations that do not contain the classes you're interested in. Consider imagenette.
-
[R] Dataset for research paper
You can try Imagenette, which is a 10-class subset of ImageNet but with the same number of images per class. There are also two datasets linked in the README for increased difficulty.
-
ResNet from scratch - ImageNet
If you're looking for something more bite-sized, how about Imagenette?
-
[D] Tiny-Imagenet original size images
Fastai has made something like that. Does this fit the bill https://github.com/fastai/imagenette?
-
[R] AdasOptimizer Update: Cifar-100+MobileNetV2 Adas generalizes with Adas 15% better and 9x faster than Adam
You don't need Imagenet to verify it really works or not, Checkout https://github.com/fastai/imagenette the fastai folks have a small subset of Imagenet, which has 3 types of the dataset, test on them. If AdasOptimizer really works it you should be able to beat their results, or at least see where it stands.
What are some alternatives?
pytorch-optimizer - torch-optimizer -- collection of optimizers for Pytorch
tiny-imagenet
ML-Optimizers-JAX - Toy implementations of some popular ML optimizers using Python/JAX
AdasOptimizer - ADAS is short for Adaptive Step Size, it's an optimizer that unlike other optimizers that just normalize the derivative, it fine-tunes the step size, truly making step size scheduling obsolete, achieving state-of-the-art training performance
Gradient-Centralization-TensorFlow - Instantly improve your training performance of TensorFlow models with just 2 lines of code!
txtai - 💡 All-in-one open-source embeddings database for semantic search, LLM orchestration and language model workflows
RAdam - On the Variance of the Adaptive Learning Rate and Beyond