Thread by @TheSequenceAI

What are the most intriguing areas of deep learning?

Let's use the examples of the biggest AI labs in the world. What do they believe is key to the near future of AI?

7 companies and the techniques they're focused on.
⬇️

1. @DeepMind — Reinforcement Learning

Evolution:
1) AlphaGo mastered Go
2) AlphaGo-Zero learned Go without knowing the rules
3) AlphaZero mastered perfect-information environments: Chess & Shogi
3) MuZero masters complex environments: Atari

2. @facebookai — Self-Supervised Learning

Examples:
1) DINO discovers and segments objects in an image or a video with absolutely no supervision
2) HuBERT learns both acoustic and language models from continuous inputs

3. @OpenAI — Transformers

Following GPT-3, the AI powerhouse continues applying transformer models into new areas such as computer vision with exciting results.

1) DALL·E creates images from text
2) CLIP connects text and images

4. @GoogleAI — AutoML

Google is advancing AutoML in highly diverse areas such as time-series analysis or computer vision.

- Model search finds optimal ML models
- Applying AutoML to transformer architectures

5. @IBMResearch — Quantum Machine Learning

At the moment, IBM has the most sophisticated quantum computing stack in the market and is actively using it to evaluate new deep learning techniques.

6. @MSFTResearch — Machine Teaching
ML research in:
- speech analysis
- computer vision

7. @amazon — Transfer Learning
- @alexa99 digital assistant