Vicarious develops Recursive Cortical Network™, a visual perception system that interprets the contents of photographs and videos.
Their website says:
- a probabilistic generative model
- a structured probabilistic model
- which can discover causal semantics
- has parts-based representations and compositionality
- similar to a grammar based model
- has a network architecture used for generalization
- can provide tight control for invariance-selectivity tradeoffs
They don’t use:
- convolutional neural networks
- Hubel and Wiesel model
Prediction: Visual object recognition and video commentaries is the furthest they will go in AI research. This is because they steal the features discovered by evolution over millions of years. Even if they succeed in developing various stances towards different types of visual phenomena e.g. physical stance, designed stance and intentional stance, they will unable to do so other sensory modalities in the near future. But they will go further than everyone before regarding computer vision.