mastodon.world is one of the many independent Mastodon servers you can use to participate in the fediverse.
Generic Mastodon server for anyone to use.

Server stats:

8.1K
active users

#convolutionalneuralnetwork

0 posts0 participants0 posts today

I remembered playing with DeepDream computer vision program that uses CNN (Convolutional Neural Network) almost 10 years ago. How fun it was to be able to run it locally on one my machines despite being very slow. Luckily I was able to find an already trained model, because training it locally was not feasible for me.

While we're (needfully) obsessing over AI in the world of direct human affairs, there's an explosion of productive exploitation of deep learning in the natural (often anthropogenically-influenced) world.

And these applications need to survive more than an elevator pitch.

Side-product, perennially surfaced during weekly climate research trawl:

sciencedirect.com/science/arti

#DeepLearning
#TreeMortality
#WildFire
#ConvolutionalNeuralNetwork

phys.org/news/2023-09-ai-algor

"…typically composed of stacks of #graphene layers with a relative twist…attracted immense attention from the #condensedmatter community…due to their high tunability and…make these systems a perfect playground for testing theories from #stronglycorrelatedphenomena…but directly obtaining these details from experimental data is often an ill-defined inverse problem…we trained a #convolutionalneuralnetwork…to recognize features of #nematicity from the data…"

Phys.orgAI algorithm learns microscopic details of nematicity in moiré systemsBy João A. Sobral

Want to share a milestone on a #ml #project I’ve been working on for a while on the side. I’m working on a #computervision application for #chess to detect the game state from a photo.
I used #transferlearning to fine tune a #convolutionalneuralnetwork with a new head for #regression to predict the four corners of the board.
I found synthetic datasets online (~6k images) and labeled ~1k real photos. I trained the model with a mix of random augmentations and projections with small errors.