Other Flashcards
Name some models created by Deepmind? What are some cool applications of ML at Deepmind?
AlphaTensor:
Finding more efficient ways to do matrix multiplication
Turns problem into single-player game and solves w/ RL
Found solutions better than Strassen’s Algorithm
Using AI to more efficiently cool data centers
Thousands of sensor readings to a NN Model
Chooses action that minimizes energy consumption
Chooses action that passes all safety tests
Forecasting energy generation from wind turbine farms
Make power grid commitments 36hrs in advance
Push for more carbon-free energy on grids worldwide
AlphaCode:
Solving competitive coding problems using transformers
Top 54%. Can solve, but not super efficient
AlphaFold:
Predict 3D structure of proteins
What MLOps tool do you like?
Personally I have used weights and biases and MLflow. I prefer the W&B, but MLflow is easy to integrate with GCP/Azure/AWS.
What are your favorite papers of 2022?
AlphaTensor: Finding more efficient matrix multiplications using RL
AlphaCode: Competition-Level Code Generation on Codeforces
Lambda: Dialog Models for Everything
- 1.56T words (40x previous model)
- Generates safety scores and candidates, then filters
Imagen - Diffusion based image generation (forcing generated images by adding text repr into images at multiple points)
Parti - GAN based image generation
What are some of your favorite papers?
Attention is all you need, Bert, RoBERTa, Imagen, Parti, AlphaTensor, AlphaCode, AlphaFold, Glove, Word2Vec, Elmo, Can a Fruit Fly Learn Word Embeddings?, SentencePiece, WordPiece, BPE.
