Independent component analysis is a computational method that separates a multivariate signal into its additive components. Sounds complicated, doesn’t it? Let’s break the definition down, then.

In computing, a multivariate signal is simply a signal that contains several distinguishable components. So you can think of it as a complete song—with music, lyrics, and even additional sound effects and backup vocals. Suppose you break the piece down into its parts. In that case, you can apply independent component analysis, so to speak, by separating each instrument, singer, and object making the different sounds from one another. In this example, the song is the multivariate signal while the instruments, singers, and objects are its additive components.

Other interesting terms…

Read More about “Independent Component Analysis”

If you’re still having a hard time digesting what independent component analysis is, maybe this simplistic illustrative explanation can help.

Think of the chart below as a multivariate signal or, as in our example, a complete song with five additive components—lead singer, guitar, drums, piano, and backup singer.

multivariate signal example

If independent component analysis is applied to it, you’ll get these results, which show what objects made the song sound the way it does.

The song won’t sound the same if any additive component is removed. The more elements you take out, the less it would sound the same. But independent component analysis can also allow you to adjust individual components, like making the piano louder than the guitar, to create a different version.

Note, however, that sound waves don’t usually look as clean as the charts above depict. Instead, they look a lot more complex than the waves we used.

The Cocktail Party Problem: What Does It Have to Do with Independent Component Analysis?

Ever been to a cocktail party? It gets loud, doesn’t it? That’s why it’s typically used to describe how independent component analysis works. The process solves the problem of isolating what one speaker says from what everyone else is saying.

So, imagine two people talking at a cocktail party. If you equip each one with a microphone, their voices are picked up at different volumes. The recordings would depend on how far the mics are from the speakers’ mouths and how loud their voices are. But how can you separate the two voices to obtain isolated recordings of each speaker?

Enter independent component analysis, which transforms each speaker’s voice into a distinct set of vectors. Think along the wave lines in our song example earlier. Once the waves are identified, you can quickly tell each speaker’s words.

Who Do We Have to Thank for Independent Component Analysis?

The first general framework for independent component analysis was introduced by Jeanny Hérault and Bernard Ans in 1984. Christian Jutten further developed their work in 1985 and 1986. After that, it was refined by Pierre Comon in 1991, who also popularized the concept in a paper published in 1994. By 1995, Tony Bell and Terry Sejnowski introduced the first fast and efficient independent component analysis algorithm.

What Are Common Applications of Independent Component Analysis?

Some independent component analysis use cases are:

  • Optical imaging of neurons: Doctors, specifically neurologists and neurosurgeons, can use independent component analysis on optical images of human brains to observe them closely. That way, they can surmise what’s wrong with the organ and fix it.
  • Neuronal spike sorting: You can also use independent component analysis to group neurons into clusters based on similarities in their spikes.
  • Facial recognition: Independent component analysis algorithms are also used in facial recognition systems to detect identifying characteristics that make a person’s face, such as iris patterns.
  • Predicting stock market prices: Much like detecting neuronal spikes, independent component analysis can also be used to track the progress of particular stocks once they’re offered to buyers.
  • Mobile phone communications: In telecommunications, independent component analysis makes noise reduction possible, allowing hands-free communication that’s as clear as possible.
  • Color-based detection of fruit ripeness: Independent component analysis can also help farmers handle huge plantations come harvest time. Algorithms can be crafted to gauge how ripe fruits are based on their color.
  • Astronomy and cosmology: Independent component analysis can also be applied to monitor the movements of stars, comets, and other heavenly bodies. It can help astrologists and cosmologists predict their movements over time.
  • Finance: Much like you would use independent component analysis to predict stock market changes and heavenly body movements, you can also employ it for other finances.

Based on the documented applications above, it’s easy to see that independent component analysis is helpful for medicine, IT, finance, telecommunications, agriculture, astronomy, and cosmology. Over time, more use cases are bound to crop up.

Key Takeaways

  • Independent component analysis is a computing process that identifies the individual parts that make up a multivariate signal.
  • The Cocktail Party Problem is most often used to explain the underlying concept behind independent component analysis.
  • Independent component analysis is helpful for medicine, IT, finance, telecommunications, agriculture, astronomy, and cosmology.