Slow Feature Analysis Example

Encouraging representations of temporally close frames to exhibit only small differences.

- Slow bc high-level visual signals change slowly over time.

- Ex. slow-feature based approaches would only require that images of people in nearby poses be mapped close to one another.

Slow and Steady Feature Analysis

Steady Feature Analysis Example

- Ex. steady-feature based approaches would not require images of people in nearby poses be mapped to another, but rather they will also be influenced by object motions and other types of visual transformations. Ex. like how colors of objects in the sunlight change over the course of a day, or how the views of a static scene change as a camera moves around it.

Intuition: Temporal Coherence

Description

Task 1: Object Recognition

Evaluation

Overview

- First, train feature function z on unlabeled data (unsupervised learning).

- Then, use these pretrained embeddings and evaluate on supervised learning approaches.