What if you could replace a core part of a Transformer with something simpler and stronger? Researchers from Princeton, NYU, and CMU present Derf. They swapped the standard "normalization" layer with a simple, element-by-element function called Derf (based on a Gaussian error
MACHINE LEARNING
-
9 Distance Metrics for Machine Learning and Data Science
By
–
9 Distance Metrics used in #DataScience and #MachineLearning (advantages & pitfalls): https://
maartengrootendorst.com/blog/distances/
———
#AI #ML #Statistics #Mathematics #DataScientist #Algorithms
———
See 730+ page book "Encyclopedia of Distances" (3rd edition): http://
amzn.to/3skDOWm -
New Edition of Transformers and Generative AI Technical Guide
By
–
Very impressive! >> "Transformers for #NLProc and #ComputerVision: Explore #GenerativeAI and Large Language Models #LLMs with Hugging Face, GPT-4V, DALL-E 3,…" [3rd Edition] http://
amzn.to/3TkBknI
————
Huge, 728 pages of #AI #DeepLearning #MachineLearning #DataScience joy! -
State-of-the-art reinforcement learning algorithms replacing traditional methods
By
–
#PaperADay 11
Discovering state-of-the-art reinforcement learning algorithms https://nature.com/articles/s41586-025-09761-x
… The paper is about replacing the traditional RL target generation algorithms (policy gradient, GAE, TD-lambda, etc) that look at a sequential block of agent frame predictions -
Iterative Improvements and Error Correction in Neural Networks
By
–
It's a neural network. Every minor update fixed errors and got smoother. Do you want to admit to what errors they fixed? I wouldn't. "We retrained the system so it doesn't run red lights in a small community in Idaho."
-
Running R1 model locally requires significant memory
By
–
I’m waiting for them to release a 768 GB version so I can run full R1 model locally.
-
Future 5.X versions aim to improve writing and general-purpose models
By
–
"We will make future versions of 5.X much better at writing" "The future is in general-purpose models"
-
Claude Code to get prompt suggestions and new Thinking effort selector
By
–
Besides this – Claude Code is about to get prompt suggestions. – A new Thinking effort selector will be added to the model selector.
-
Fine-Tuning Video Models for Visuomotor Robot Control
By
–
"Fine-Tuning Models for Visuomotor Control and Planning" This paper proposes Cosmos Policy, showing a pretrained latent video diffusion model (Cosmos-Predict2) can be adapted into a SoTA robot policy via a single post-training stage on robot demonstrations, without
-
Early Grok 4.20 checkpoint achieves +10% gain on Prediction Arena
By
–
An early Grok 4.20 checkpoint has been spotted on Prediction Arena, achieving +10% gain after a 2 weeks long round. Soon?