DeepRob: Deep Learning for Robot Perception – University of Michigan, 2023 An ongoing class on neural network based approaches for robot perception. The class covers advanced topics in computer vision & emerging topics in deep robotic perception. Videos:
https://
youtube.com/playlist?list=
PLf_SmXJixhnXoMs0Qvxe500BrjfbIOwSg
…
RESEARCH
-
DeepRob: Deep Learning for Robot Perception University of Michigan
By
–
-
Hugging Face AI Detection Competition: Identify AI Generated Images
By
–
Join the first ever hugging face competition: AI or Not! Can you predict which images are AI generated?
-
Machine Learning for Remote Sensing Workshop at ICLR 2023 Kigali
By
–
Machine Learning for Remote Sensing Workshop – ICLR 2023, Kigali Machine learning is increasingly finding many useful applications in remote sensing. The first workshop of ML for remote sensing will be held at ICLR 2023 in. Submissions are still open! https://
nasaharvest.github.io/ml-for-remote-
sensing/iclr2023/
… -
ICLR 2023 ML4Materials Workshop Submission Deadline in Two Weeks
By
–
The deadline for the ICLR 2023 ML4Materials workshop is in 2 weeks — we welcome submissions on both ongoing work and work published recently. Our workshop is fully virtual this year so there are no travel requirements. https://
ml4materials.com. Please submit! -
ChatGPT Success Overshadows Misconceptions About AI Models
By
–
#ChatGPT has taken the world by storm, no doubt! But the lessons people talk about wrt what it means is mostly misinformed. Open AI's GPT-3 based models aren't the best performant or most accurate in the world today, and it doesn't do anything usef…
https://
lnkd.in/g57giq8y -
CIFAR-10 Classification Progress: From Manual 94% to Automated Scripts
By
–
Also reminded of this blog post from ~12 years ago. I classified CIFAR10 manually and got… 94%! SOTA then was ~80%, certainly not in 10 seconds. Then I predicted we'd top out around 85-90% (lol). 12 years later: 94% is 10 seconds with one 600-line script https://
karpathy.github.io/2011/04/27/man
ually-classifying-cifar10/
… -
CIFAR10 Training to 94% in Under 10 Seconds on A100
By
–
More on cramming: CIFAR10 hyperlightspeedbench.
Train CIFAR10 to 94% in under 10 seconds on a single A100. With a single readable 600-line http://
main.py, bunch of nice tricks implemented within. -
Mafia concept gains relevance in AI future
By
–
Doesn't seem like it now, give it 3-4 more years. Then "Mafia" would make complete sense.
-
Simula Presents Graph500 BFS Algorithm on Cerebras CSL
By
–
Reminder that tomorrow at 9am PT, Simula's Luk Burchard will be sharing research on the Graph500 BFS algorithm he ported over to the Cerebras Software Language (CSL). Join this event to learn about tooling he's developed to map problems onto CSL Register: https://
hubs.li/Q01zvG4Y0 -
LaMDA Open Domain Model Handles Any Subject
By
–
LaMDA is an “open domain” which means it’s a single model that doesn’t have to be re-trained for different conversations or subjects. It can answer questions about history or the weather or literally anything else.