Announcing Parfait — private aggregation & retrieval, federated, analytics, inference, & training — a GitHub organization that we’ve developed to demonstrate our state-of-the-art methods across 4 privacy pillars. Learn how it was created & is being used →
https://
goo.gle/4axVNfN
MACHINE LEARNING
-
Google Announces Parfait: Private Aggregation & Federated Learning
By
–
-
Machine Learning Analyzes Directorial Styles in Theater
By
–
Join today's @StanfordHAI seminar with Michael Rau to hear how machine learning and computer vision can be used to analyze directorial styles and other aspects of theater performances. https://
stanford.io/3EfZXga -
Upcoming Chapter Content and Bonus Multi-Head Attention Materials
By
–
Yes, stay tuned! (And there's also the "Comparing Efficient Multi-Head Attention Implementations" bonus content if you are still bored after completing the chapter :))
-
Tiny 1.5B Model Outperforms GPT-4o, Runs Offline
By
–
“outperforms GPT-4o and Claude-3.5-Sonnet on math benchmarks”
— Aaron Ng (@localghost) 22 janvier 2025
Here it is running on a phone, completely offline.
Tiny 1.5B AI model that will run on almost any hardware, even without internet.
Are you getting it yet? https://t.co/RjXTKvsRPd pic.twitter.com/6GimUSBKOo“outperforms GPT-4o and Claude-3.5-Sonnet on math benchmarks” Here it is running on a phone, completely offline. Tiny 1.5B AI model that will run on almost any hardware, even without internet. Are you getting it yet?
-
AI Solving Major Hard Science Challenges Beyond Capex
By
–
It literally is solving the biggest challenges in the hard sciences im not even talking about capex. But you do you.
-
Approximately 1 billion people use BPE tokenizers and LLMs
By
–
according to my napkin math, about 1 billion people use BPE tokenizers (/LLMs)
-
Data Augmentation vs Generative Models: Key Differences Explained
By
–
yes, a kind of flavor of it. but also not quite. in traditional data augmentation (say image/cnn contexts) you add noise to the data, rescale it etc. here, the difference is that you actually generate that data. in image/cnn land that would be using a diffusion model to generate
-
Linear Approximations in Modern Machine Learning: Why Simplicity Works
By
–
that's nowadays literally a multi-million dollar question . Btw there are lots of (linear) approximations that came and went in the past few years, but it's still more or less the same olde because it just works so well!
-
Kimi K1.5: Scaling Reinforcement Learning with LLMs
By
–
Oh!!! kimi k1.5 is coming! Kimi k1.5: Scaling Reinforcement Learning with LLMs https://
github.com/MoonshotAI/kim
i-k1.5?tab=readme-ov-file
… -
OmniManip: General Robotic Manipulation via Spatial Constraints
By
–
OmniManip: Towards General Robotic Manipulation via Object-Centric Interaction Primitives as Spatial Constraints https://
omnimanip.github.io https://
arxiv.org/abs/2501.03841