What's new in SuperMemo 17?

Contents


SuperMemo 17 represents a radical break with the past. A great deal of functionality was removed on the basis of:

  1. little popularity and
  2. high cost (in confusion, bugs, programming, maintenance, etc.).

For that reasons, review the list of warnings carefully before you upgrade.

New features

Top ten

  1. Neural creativity - boost your creativity and problem solving powers using the same tricks as your brain uses in information processing
  2. Algorithm SM-17 - new algorithm allows of an effective repetition at any time: in 20 sec or with a 20-year delay. SuperMemo can handle it all. If you play with intervals or do subset review, the algorithm improves SuperMemo's predictive powers by 5-10%, which is a lot in terms of learning efficiency!
  3. Concepts and concept maps - link information in between branches of the knowledge tree. For example, group together all you know about the flu virus and beat the flu by knowing its tricks (e.g. using neural creativity)
  4. Full user lists in registries - no more limit on member usership. Now you can find all elements using texts, pictures, etc. even if these go into hundred thousands
  5. Multithreaded image downloads - no more wait for pictures to download. They will get to SuperMemo before you have time to review them
  6. Image references - keep all information about the image (incl. the source URL) and prevent importing the same image twice even if you change its name, overwrite the actual file, etc.
  7. Deprecating URLs - if your partly-processed Wikipedia article is 10 years old, you might prefer to deprecate, deprioritize, and import an up-to-date version without wasting prior work. URL deprecation will make sure that you will be directed to the new version at duplicate detection or when you plan a review
  8. Fit all pictures - with a keystroke, economize space in your element containing multiple pictures of various sizes
  9. Web import : Browse duplicates - check all prior duplicated web imports to see if they are worth replacing with newer/up-to-date articles, pictures, etc.
  10. Commander: Import file - you can download files from the web directly to a component or import files from a disk

Algorithm

Concepts

Incremental learning

Learning

Components

Templates

Registries

Browser

Contents

Statistics

Search

Imports

Exports

Interface

SleepChart

Tasklists

Plan

Performance

Other

Bugs and annoyances

Learning

Components

Search

Registries

Browser

Contents

References

Interface

Imports

Exports

Video

SleepChart

Plan

Algorithm

Files

Repair

HTML

XML

Mail

Warnings

Examples

Concept map

SuperMemo: An exemplary concept map displaying outgoing links for the current element

Fit all pictures example

Fit all pictures for images in the Commander makes it possible to save element space and organize multiple images.

For example:

  1. Image download: After a download of seven images from Wikipedia, the text is displaced by a large number of big image components:
    Fitall1.jpg
  2. Image tiling: Those image components can be tiled into a single column to reduce the image area width:
    Fitall2.jpg
  3. Image fitting: In the text step, Fit all pictures can be used. Fit all is the same as Fit picture, but it is executed on all images. The images are pushed to the right and all space left over is reused maximally to provide more room for the text:
    Fitall3.jpg

Priority dialog

SuperMemo: Priority dialog

Figure: Element priority dialog box which you can use to set or change the priority of an element in SuperMemo. Position is the position in the priority queue. Percent is the priority (0% means maximum priority). Colors reflect the priority: red - highest, blue - lowest. Before and After show the elements placed before and after the current element in the priority queue.


SuperMemo backgrounds in Windows 10

SuperMemo: In Windows 10, SuperMemo background can be used as a nice visual collection chooser between active collections opened in separate instances of SuperMemo running simultaneously

Figure: In Windows 10, SuperMemo background can be used as a nice visual collection chooser between active collections opened in separate instances of SuperMemo running simultaneously. Use Alt+Tab to switch between collections, or click the collection of your choice. This new feature comes with Windows 10, not with SuperMemo 17. You can also use SuperMemo background pictures in this way in SuperMemo 16 in Windows 10.

Contents window with descendant counts

SuperMemo: The contents window displaying the knowledge tree (i.e. the hierarchical structure of elements) of a collection
Figure: The Contents window displaying the knowledge tree (i.e. the hierarchical structure of elements) of a collection. In SuperMemo 17, each branch shows the number of its descendants in the parentheses. Concepts use their dedicated icon, which can be seen in the root (collection root is always a concept).

Recall matrix visualization

SuperMemo: The Recall[] matrix graph showing actual recall differs from predicted retrievability

Figure: The Recall[] matrix graph shows that the actual recall differs from predicted retrievability. For higher stabilities and difficulties, it is harder to reach the desired recall level.

First forgetting curve spanning a decade

SuperMemo: The first forgetting curve (first repetition for items with no lapses)

Figure: The first forgetting curve (first repetition for items with no lapses). Unlike it was the case in earlier SuperMemos, where all forgetting curves were exponential, the first forgetting curve in SuperMemo 17 is approximated using power regression. This provides for a more accurate mapping due to the heterogeneity of the learning material introduced in the learning process that results in superposition of exponential forgetting with different decay constants. The use of power regression explains why the first interval might be slightly shorter in Algorithm SM-17. On a semi-log graph, the power regression curve is logarithmic (in yellow), and appearing almost straight. The curve shows that in the presented collection recall drops merely to 58% in four years, which can be explained by a high reuse of the memorized knowledge in real life. In earlier SuperMemos, recall data would only be collected in the span of 20 days, and negatively exponential forgetting curve would make for far lower retrievability predictions. The first optimum interval for the forgetting index of 10% is 3.76 days. The forgetting curve can be described with the formula R=0.987*power(interval,-0.07), where 0.987 is the recall on Day 1, while -0.07 is the decay constant. This is case, the formula yields 89.5% recall after 4 days, which is then used as the first rounded optimum interval. Almost 77,000 repetition cases were used to plot the presented graph. Steeper drop in recall will occur in collections with a higher mix of difficult items, in poorly formulated collections, or in new users with lesser mnemonic skills.

Notes

Registries provide access to all member users

In older SuperMemos, registry member users lists are limited to 500 elements/objects for performance reasons.

Access to full user lists is extremely useful but longer lists would significantly slow down SuperMemo. Lists might be kept in contiguous file in memory to resolve this problem, however, one contiguous file would slow down SuperMemo even further esp. if it had to be paged. Keeping lists for individual members would solve the speed issue, but this would require as many files as there are members in registries, which would go into millions for larger collections.

SuperMemo 17 adopts an golden mean solution: short lists are kept in the LST file as before, longer lists are kept in contiguous files loaded to memory on demand. The cut off point for memory files was chosen on the basis of performance statistics (to minimize the number of files without noticeably degrading performance). The optimum was set for collections with a million elements. Small collections will behave as before. Large collections will store long lists in dedicated usership files, which should rarely number more than a dozen or for most used registries (esp. texts).

For large collections, the performance might degrade slightly, but this should be well compensated by the increase in the speed of computers in the two years that passed since the release of SuperMemo 16.

Element registration with concept groups is not obligatory

Elements do not register with the default concept group when being added to a collection.

The reasons:

Better memory management

SuperMemo 17 provides superior memory management based on Delphi's native solutions rather than the old memory manager written in assembly language back in 1995. Large block memory operations should now be faster making SuperMemo faster in many contexts

Repetition history import

File : Import : Repetition history provides several options:

Delay parameter has been removed

Delay parameter has been inaccurate due to changes to U-Factors with auto-postpone. Retrievability is a better marker of delays. You can sort the browser by retrievability to see which items are most delayed. In short, low R means high delay.