A lot of fun in the Appendix, e.g. how GeLU can be used for multiplication / bypassing it as identity, use of LayerNorm for division, or bypassing that as identity, etc.
MACHINE LEARNING
-
Transformer Equations as Low-Level Microcode and Assembly Instructions
By
–
Nice! Like the track of work. Equations of Transformer are a bit like low-level microcode, this track tries to "go up" to uncover an implied assembly instruction set (e.g. RAW "read-arithmetic-write" operator?), and implemented algorithms on top of that for e.g. ridge regression.
-
MLCommons Algorithms Benchmark Standardizes Neural Network Training Comparison
By
–
Today at 3:30pm, come by the #NeurIPS2022 Google booth to hear @zacharynado talk about the @MLCommons Algorithms benchmark, an effort to standardize how researchers compare training algorithms for neural networks. https://
github.com/mlcommons/algo
rithmic-efficiency
… -
AI Predicts Heart Attack and Stroke Risk From Chest X-rays
By
–
#AI program can predict heart attack, stroke risk within decade using single chest X-ray https://
studyfinds.org/ai-program-hea
rt-disease-x-ray/
… -
Allen AI’s CLOSE Completes Visual Tasks From Text Inputs Alone
By
–
No Images Are Needed! Allen AI’s CLOSE Learns to Complete Visual Tasks From Text Inputs Alone
-
Efficient Pre-training and Fine-tuning Approaches for Large Models
By
–
As the scale of models increases, the computational effort needed for training grows rapidly. Check out a pair of new approaches that can significantly improve model performance through thoughtful pre-training and fine-tuning with only minimal compute. https://
goo.gle/3UicYbD -
AWS DataZone and ML Forecasting Close Enterprise Data Gaps
By
–
Close #data gaps! with ' #software that works the way you think' #reInvent From integrative #data #governance with #DataZone to enhanced #discovery with 'Q' #ML based #forecasting #ReInvent #Security #CyberSecurity @Paula_Piccard @RetailInstant @awslambdas #coding #Analytics
-
Conformal Prediction for Uncertainty in AI-Generated Images
By
–
Hyper-realistic AI-generated images are becoming increasingly commonplace.
— MIT CSAIL (@MIT_CSAIL) 29 novembre 2022
Using conformal prediction, researchers show how to construct uncertainty intervals w/meaning, i.e. a person’s hair color or the amount they’re smiling: https://t.co/GqTCP7fWnu@swamiviv1 @phillip_isola pic.twitter.com/bBKKxgFQ2zHyper-realistic AI-generated images are becoming increasingly commonplace. Using conformal prediction, researchers show how to construct uncertainty intervals w/meaning, i.e. a person’s hair color or the amount they’re smiling: http://
bit.ly/3Fbuzxq @swamiviv1 @phillip_isola -
CICERO by Meta AI: Fluent Conversations with Multiple Diplomacy Players
By
–
What impresses @ml_perception the most about #CICERObyMetaAI is its ability to hold fluent, human-like conversations with up to Diplomacy players simultaneously. Learn how we created an interface between language and strategic reasoning models:
-
Predibase CEO presents low-code ML systems at TMLS2022
By
–
Hope to see everyone at #TMLS2022 today! Piero Molino, Predibase CEO & creator of open-source http://
Ludwig.ai is leading a talk on how #lowcode declarative ML systems enable teams to rapidly build, iterate and deploy state-of-the-art models. https://
pbase.ai/TMLS22