Optimization of Learning: Summary

Piotr Wozniak, 1990

From P.A.Wozniak, Optimization of Learning, updated and corrected May 20, 1997

Here is the list of the most important points of the thesis (terminology given in boldface was included in the Glossary):

  1. The SuperMemo method used in repetition spacing was presented (Chapter 3). The following elements of the method marked the most significant steps in its development:
    • application of the recall principle (Chapter 3.1)
    • application of the minimum information principle (Chapters 2 and 3.1)
    • application of the optimum repetition spacing principle (Chapters 2 and 3.1)
    • introducing E-factors (Chapter 3.2)
    • introducing the function of optimum intervals (Chapter 3.3 and 3.4)
    • application of interval dispersing (Chapter 3.5)
    • application of the propagation of changes in the matrix of optimal intervals (Chapter 3.7)
  2. Software implementation of the SuperMemo method was described (Chapter 4)
  3. SuperMemo on paper was described (Chapter 7)
  4. The function of optimum intervals was found by means of three methods:
    • specially designed experiment (p.16)
    • univalent matrices of optimal factors in the Algorithm SM-5 (Chapter 3.6)
    • model of intermittent learning (Chapter 11.4)
  5. A comprehensive analysis of the SuperMemo learning process was presented:
    • an accurate simulation model of the SuperMemo process was constructed (Chapter 5)
    • function of the acquisition rate was found (p.68)
    • all-life, maximum acquisition rate was predicted to be about 230 item/year/min (this value may be substantially lower in case of ill-structured SuperMemo databases, or, possibly, higher in case of further development of the knowledge structuring techniques) (p.68)
    • the all-life capacity of the human brain was estimated to be about several million SuperMemo items (p. 72)
    • long-term acquisition of knowledge with the use of the SuperMemo method was shown to be close to linear (p.69)
    • workload function was found (p. 65)
    • reducing the forgetting index was found as of little value for the speed and quality of learning
    • eliminating items characterized by low E-factors was demonstrated to be crucial for the speed of learning (p. 68)
    • it was found that only 5% of the learning process can be spent on acquisition of knew knowledge, the rest is consumed by repetitions of the old material
    • forgetting rate in case of the cessation of repetitions was found to be much higher than the acquisition rate (e.g. after 5 years of the process, 60% of knowledge is lost in the first year after the cessation) (p. 70)
    • burden parameter was proposed as a very accurate measure of the learning progress in SuperMemo (p.53)
    • model of intermittent learning was constructed (Chapter 11)
    • relationship between the forgetting index and knowledge retention was found to be close to linear. For the index equal to 10%, as in the Algorithm SM-5, the long-term retention was predicted to be 94% (the presently reported retention reaches 96%) (p. 154)
    • the increase of the stability of memories was found to be the greatest if the intervals are twice as long as the optimal intervals (it corresponds to the forgetting index equal to 20%) (p. 156)
    • the function of the workload-retention trade-off was found (p. 155)
    • the function of the workload-retention trade-off was used to determine that the desirable value of the forgetting index falls in the rage 5% to 10% (p. 155)
  6. Method-independent prerequisites of the successful application of SuperMemo were formulated (Chapter 6)
  7. Results of a questionnaire collecting opinions of SuperMemo students were presented (Chapter 8)
  8. Biological aspects of learning in the light of the SuperMemo method were analyzed:
    • distinction between stochastic and deterministic learning was proposed (Chapter 10.2)
    • optimum repetition spacing in stochastic learning was found to be possibly less dense than that of deterministic learning (p. 95)
    • illustrative, hypothetical models of neural circuitry involved in stochastic and deterministic learning were described (p. 106)
    • arguments for the presynaptic character of the facilitation in case of stochastic learning were listed as well as arguments for heterosynaptic facilitation in deterministic learning (Chapter 10.2)
    • discussion of the nature of short-term and long-term memory was included (Chapter 10.1 and 10.3)
    • new arguments for the postsynaptic membrane as the location of long-term memory were put forward (Chapter 10.3)
    • E-factors were proposed as a reflection of the number of synapses involved in remembering particular items (Chapter 10.4.1)
    • existence of at least two components of memory was postulated and demonstrated: retrievability and stability (Chapter 10.4.2)
    • phosphorylation of proteins was considered as possibly responsible for retrievability (Chapter 10.4.3)
    • the number of postsynaptic receptors was considered as possibly responsible for stability (Chapter 10.4.3)
  9. Possible future applications of SuperMemo were outlined (Chapter 12). As an illustration, software supervising a touch typing training was described (p. 87). A simple method for using SuperMemo in learning to play musical instruments was presented (p. 92). Universal nature of learning based on repetition spacing was suggested (p. 165).
  10. The mere existence of the SuperMemo METHOD refutes or calls in question a pretty large number of common sense conceptions and dogmas of the psychology of learning. The most prominent examples are listed below:
    • opposition of memorization to logical thinking is pointless. Memorization (or according to my terminology deterministic learning) lays ground for the refinement of the circuitry of the brain which is later on used in the process of thinking (p. 167)
    • ever-lasting memory acquired by a single learning act is unlikely. Cases of supernatural memory refer to either short-term memory, mutant individuals or must be otherwise seriously reconsidered [Luria, 73] (p. 168)
    • time necessary to learn a given material is proportional to the first, not second power of the size of the material. This refers only to long-term memory and properly spaced repetition process (p. 169)
    • forgetting has a biochemical nature (trace-decay theory) and is only partially caused by interference (interference theory). Proper application of the minimum information principle, principle of univocality, mnemonic techniques etc. allows to one to reduce interference to a negligible level (Chapter 10.4)