What is the Indian Buffet Process 🌱

Motivation

  • Dirichlet Process can give us clusters for data points, but it cannot cluster data into overlapping clusters.
  • In many applications, data points exhibit properties of multiple latent features:
    • Images contain multiple objects, movies contain aspects of multiple genres, etc.
  • Can we model a system where:
    • we allow each data point to exhibit multiple latent features, to varying degrees
    • we make the number of features unbounded a posteriori (like with the Dirichlet Process)

The Process

IBP

  • \(m_{k}\) is the number of previous customers who have sampled the dish.
\[\text{ If } Z \text{ follows this process then } Z \sim IBP(\alpha)\]

Beta Bernoulli Model

  • The Beta-Bernoulli model where we take the limit as the number of latent features goes to infinity describes the Indian Buffet Process

IBP

IBP

Infinite Gamma-Poisson Process

IBP

IBP

IBP

Notes mentioning this note

There are no notes linking to this note.


Here are all the notes in this garden, along with their links, visualized as a graph.