Suggest edit — 20-07-13 Deep Ideas in Class Discovery

Title

Name

Note

---
title: "20-07-13 Deep Ideas in Class Discovery"
visibility: public
---

# 20-07-13 Deep Ideas in Class Discovery

Category: [[idea-lists-upon-request|Idea Lists (Upon Request)]]

[Read the original document](https://docs.google.com/document/d/1iHTetpMb2OBE2ge87ggjy3cu38J-aTRGKAxJa--LSS8/edit?usp=drivesdk&sa=D&ust=1596495076753000&usg=AOvVaw13K__yUczx634NOUksUrnD)

---

1. Transfer objective
2. David’s layer type (anisotropic)
3. Training in sim
4. Recursive self-improvement
   1. Training leads to improved class creation
5. Modifying the ontology dynamically
   1. Class merging, splitting, creation
6. Learnability as the prior
7. Start with contrastive learning (to get a good clusterable representation)
8. One-vs-all / Object Detection / Escape the one class per image assumption
9. Internet scale training
   1. Discover everything for incredible transfer
10. Transfer because you’ve already seen and classified similar data
11. Curriculum as you transfer to similar classes

Mathematical possibilities:
* David’s Anisotropic Layer
* Semi-supervised clustering
* Geometry of metric space

Obstacles and Limitations:
* Deep metric clustering outperformed by raw image space clustering
* Computational expense of transfer evaluations
* Noisy label training

---

*Source: [Original Google Doc](https://docs.google.com/document/d/1iHTetpMb2OBE2ge87ggjy3cu38J-aTRGKAxJa--LSS8/edit?usp=drivesdk&sa=D&ust=1596495076753000&usg=AOvVaw13K__yUczx634NOUksUrnD)*