@jake - you should find this helpful.
@Paul_Lamb - you should like this in light of your work with the Cortical IO retina.
Part of how I keep the interest in the material is to switch focus on regular bases; hence totally unrelated projects are a good idea in education.
There is a reason why education has settled on the model of cycling through different topics in classes. One of the basic principles of neural networks is “orthogonal axis of representation” in the source material to be learned.
If two things are being learned and you have no way to tell them apart they blur into one. What is desired is that as you learn more details they become separable and learn-able. If there is nothing to relate the material to it is just meaningless noise.
During sleep, you consolidate this new information and it becomes available as part of your parsing for new information. This learned material aids in this process of taking in more material - the landscape now has more features to use as landmarks in placing additional material.
It seems that an hour or two of new information seems to be about as much as most students can absorb in a given area of the brain in a day. Then we switch to an orthogonal branch of information and train that up for the day.
It is helpful if this new information adds details to the “edges” of some of the other information being absorbed.
In deep learning, we present the entire body of material with small learning rates to allow this landscape to form, with multiple cycles to allow the orthogonal axis’s and islands to be formed incrementally. Error back propagation is usually the method to determine “novelty.”
If you think about it - the brain uses much the same process solved in slightly different ways.
With humans, we add the novelty and personal relevance coloring to enhance the learning rate as this hour of material is absorbed. I think that personal relevance is self-explanatory and leave the details to your imagination.
Novelty is a different kettle of fish. I am pretty sure that everything you learn is always parsed in terms of what you have learned before. All new learning is “delta coded” in relation to what is already in the cortex. If what you are being presented with is fully recognized then there is nothing novel - no difference is generated to be learned in the hippocampus. If it is not recognized then it is something new to be learned.
The novel “learning” is clustered with the related material with some sort of learned relationship. The proximity between existing material and the presentation establishes this cluster relationship. There must be a chain (no matter how far removed) to the original grounding in our physical body senses and emotional shading to those sensations.
Humor or pleasant surprise (or fear 
) is likely to help consolidate the material. The continuum of the various emotions helps place a value on the material when recalled.
An example of this grouping can be seen in the Cortical IO retina data structure.[1] Yes - the consolidation process adds this delta to the existing cortex to form a more detailed landscape. The capacity of the hippocampus to hold this delta coding is the capacity to learn new material in a given daily cycle.
So recapping - for effective learning:
- it has to be novel. This is likely to vary by student.
- It has to be personally relevant. This is likely to vary by student.
- It has to be orthogonal (linearly separable); you can learn in more than one area each day.
- New information should be paired is some way with known information. This relationship should be selected to highlight the feature to be learned.
- It has to be in bite-sized chunks.
- You have to sleep on it - with good REM cycles to consolidate it.