Interleaving a Human/ Machine learning Technique

Interleaving in Human learning

What is interleaving?

Interleaving is a learning technique that involves mixing together different topics or forms of practice, in order to facilitate learning

Imaging you are in a study session preparing for a test, a test that the would be comprised of various topics. How should you study for them? One at a time or switching between them?

Research suggest that you shouldn’t study one idea, topic or type of problem for too long. Instead you should switch it up often. Interleaving like this might seem harder than sticking to one topic or material for a long time, but it is more helpful in the long run. This strategy is particularly useful for topics regarding problem solving like math or physics. Interleaving can help you choose the correct strategy to solve a problem, it can also help you see the links or similarities between ideas.

Other things to consider with is idea with regards to learning are:

How often do you switch topics?

While it’s good to switch between ideas, don’t switch too often between ideas or spend to little time on anyone idea. You need to make sure to cover some ground before tackling a new idea. Switching too often would verge towards “multi-tasking” which is not as effective.

What should you do while Interleaving?

Don’t just switch an forget about each idea once you’ve switched, instead try to make links between different ideas as you switch them. And then go back over the ideas again in a different order to strengthen your understanding.

Furthermore, creating a link between different subjects/fields of study have be the catalyst to great intellectual achievements. On such instance, is the story of Rene Descartes, he unified the geometry and algebra(seemingly unrelated fields at the time) which giving rise to the emergence of analytic or Cartesian geometry, which is responsible for digital technology as you know it today. I talk about this in more detail in a different article I wrote; Descartes’ Paradise ( I know again with the shameless self promotion, I promise I paid me)

Why and how interleaving works

The benefits of interleaving are attributed to a number of cognitive mechanisms, as a result of the varied ways in which interleaving can benefit learners, as well as the varied ways and situations in which interleaving can be implemented. Some of these mechanisms include:

• Contextual interference: Essentially mixing materials increases interference during the performance of the task,which promotes the use of effective learning strategies by learners. Hence, this leads to better learning of the material.
• Discriminative contrast: Mixing materials helps learners notice the similarities and differences between the concepts they are learning which helps them learn the concepts better.
• Another notable mechanism behind interleaving is that it pushes learners to figure out what strategies, techniques, and information they need to use in order to solve problems that they encounter.

Interleaving in Machine learning

Similarly to the concept behind interleaving in human learning, Netflix uses interleaving technique to accelerate the pace of algorithm innovation by increasing the rate of learning which to leads to even more personalized , hence better recommendations.

So basically, Netflix uses multiple ranking algorithms, each optimized for different purposes. For example the Trending now row incorporates recent popularity trends , while the Top picks row is on the homepage and makes recommendations based on personalized ranking on videos. These Algorithms used with other are used construct personalized homepages for millions of members.

Netflix uses the interleaving technique, to increase the rate of learning by testing a broad set of ideas quickly. They have devised a two-stage online experimentation process. The first stage is a fast pruning step in which we identify the most promising ranking algorithms from a large initial set of ideas. The second stage is a traditional A/B test on the pared-down set of algorithms to measure their impact on longer-term member behavior.

The first stage finishes in a matter of days, leaving then with a small group of the most promising ranking algorithms. The second stage uses only these select algorithms, which allows them assign fewer members to the overall experiment and to reduce the total experiment duration compared to a traditional A/B test.

Conclusion

Just like in both human and machine learning, interleaving can be a powerful technique for more productive learning process. In Machine learning, this technique is used to accelerate ranking algorithm innovation at Netflix. It allows them to sensitively measure member preference for ranking algorithms and to identify the most promising candidates within days. This has enabled them to quickly test a broad set of new algorithms, and thus increase their rate of learning.

In humans interleaving can improve your ability to retain new information, acquire new skills and even enhance already existing abilities in a wide range of domains. When deciding which material to interleave, make sure the items aren’t too similar or too different. Effectiveness varies based on people, the materials involved and environment. It’s not an easy process at first, one has to be patient and persistent.

References:

(1) Rohrer, D., & Taylor, K. (2007). The shuffling of mathematics problems improves learning. Instructional Science, 35, 481–498.

(2) Rohrer, D. (2012). Interleaving helps students distinguish among similar concepts. Educational Psychology Review, 24, 355–367.

https://netflixtechblog.com/interleaving-in-online-experiments-at-netflix-a04ee392ec55