Thanks @mlopscommunity for this really great animation based on our Eng Blog article by Shaji Chennan Kunnummel and Iaroslav Tymchenko. #machinelearning#engineering
https://t.co/cPw2T4O3hV
Excited to announce the launch of GPT-42.
- Half the size of GPT-3 (100 billion parameters)
- runs on 775 watts a day (2000 calories)
- can do one-shot learning
- multi-modal
- does NOT require 9,000 GPUs
- 300,000 years worth of evolution research!
Inference API coming soon!
Many like to say intelligence is not a single thing or a reification of assumptions, IQ is terribly flawed, and the concept has been/is used for bad causes. But so is poverty. So is wealth. So is power. So is inequality. Many important and predictive "things" are like that.
Working on a repo where you can build pipelines and productionize them using AWS and GCP. Use it to start your own project! #MachineLearning#DataScience#softwareengineering https://t.co/0Q9JwfXgTR
@abhi1thakur Just watched your first video and I really liked it! Quick question about how you setup the project: is there an advantage to running the https://t.co/XhTnwF28hv file using ‘python -m src.train’ over ‘python src/train.py’? Thanks!
So, "deep learning" is the idea of doing representation learning via a chain of learned feature extractors. It's all about describing some input data via *deep hierarchies of features*, where features are *learned*.
A further question is then: is the brain "deep learning"?
Looking back, my last decade was like a neural network. Some parts were linear. Some were nonlinear. I never seemed to get enough data, and always got stuck in local minima. There was a lot of learning. I can't explain how any of it worked, but the results came out alright.
The scikit-learn TransformedTargetRegressor can make a fantastic addition to your workflow. Instead of transforming your target before fitting and then after prediction, it automates the process for you.
Here's an example from the Kaggle Housing competition