Auto-encoder

Click on a tile to change the color scheme:

default slate

1. Overall

2. Feature Disentangle

Application: Voice Conversion

3. Discrete Latent Representation

• The discrete, finite options in the middle represent basic features.
• If we use other forms rather than vectors in the middle, we can get some human readable summaries!
• Use a discriminator to force the decoder to generate human readable summaries. (equipped with RL)

4. Anomaly Detection

To address the problem that it is hard to collect abnormal data if we want to use classifiers.

---

Last update: June 16, 2023

Authors: Colin