150 likes | 163 Views
This article discusses established learning models and introduces a novel learning model that emphasizes information recognition, processing, and storage in a feedback-control loop.
E N D
Learning as a feedback-control process Michael H.W. Hoffmann University of Ulm, Department of Microwave Techniques, Albert-Einstein-Allee 41, D-89069 Ulm, Germany eMail: Michael.Hoffmann@ieee.org • Outline • On established learning models • A novel learning model • Conclusions • Summary
1. On established learning models There are numerous learning models known from literature. These are classified to belong to different schools: school protagonist(s) humanists Maslow, Rogers, Dewey behaviorists Skinner, Estes, Guthrie, Maltzman cognitivists Atkinson, Shiffrin, Schneider, Craik, Lockhart constructivists Piaget, Vygotskii, Bruner, Bandura organizational-learning theorists Kolb, Race, Argyris, de Bono, Revans Some of the developed models are based on information processing.
Learning as a feedback-control loop Gagné’s model might be seen as an information-processing loop. • If it is assumed additionally that the “response generator” produces responses and causes actions aiming at: • avoiding frustrations • achieving well-being • then the loop might be seen as a • Such models have been analyzed intensively by engineers ! • Based on these ideas, a new learning model has been developed with an emphasis on information recognition, information processing, and information storage. feedback-control loop.
2. A novel learning model Paradigm: Learning is a process in which the assessment of a given set of information and a list of potential consequences by that assessment are changed enduringly as the result of comparativeobservation or comparison of actual experience with former experience. Thus, as a main part of learning, there is the For that purpose, it is necessary to understand how recognition of novel pieces of information works, and how works their classification as being important ! necessity to recognize something to be new.
Comparison and assessment of information Example: Auditory information outer ear acoustic matching and “antenna diversity” middle ear waveguide and bandpass filter inner ear short-term Fourier transform thalamus (part of the brain) classification into • dangerous • potentially meaningful • music • ignore first step of recognition
Speech recognition Classification of potentially meaningful auditory information recognize smallest units of speech phonemes recognizesensible combination of phonemes,of intonation, and of stress intonation stress syllables morphophonological structure recognizegrouping of the speech stream into words syntax recognize (in a multistep process)grammatical structure meaning recognize (in a multistep process)what was meant
Low-Level Learning • Recognition in general is thus a multistep process of comparing pieces of input information with known patterns. • If there is no known pattern for recognition, then such a pattern must be recognized and stored! This is learning on a low level. • This needs: • rational and/or emotional attention to an unknown aspect (otherwise no classification as being worth storing it) • repetition in order to recognize significance
The Control-Loop of Low-level learning repetition and attention necessary! short-termstore short-termstore control-loop sensoryregister long-termstore
Higher-level learning: Example (Learn a language) thinking in the other language Level n+1 Strategies (What, if?) Basic facts new words Level n (What?) Abstract rules (How?) Context context in whichwords are used recognition of grammar rules (What is linked to what?) Procedures (What happens drill of automatic usage in certain combinations when?)
Higher-level learning as nested feedback-loop n5 learningloops Level n+1 feedback Strategies (What, if?) n1 learningloops Basic facts Level n (What?) n4 learningloops Abstract rules (How?) Context n2 learningloops n3 learningloops (What is linked to what?) Procedures (What happens when?)
3. Conclusions • Learning is a nested feedback process with many feedback-loops • Since learning changes the state of memories, feedback loop are time-variant • On higher levels of learning, learning results depend on how active the learner is in order to find rules or to develop strategies. • Learning success is not only depending on offered novel input. It rather depends on the (very complex) internal state of the learner.Thus, behavior of the learner is notjust depending on external stimuli. • Therefore, the process of learning is different for every individual person. It depends on the learner’s inner state and on his or her own activity and capability of recognizing rules and of finding strategies, how fast new patterns are recognized and made accessible for new strategic thoughts.
Conclusions • Every feedback process needs time. Successive feedback-controlled decisions need more time than simple decisions. • This means particularly that acquisition of canonical and strategic knowledge needs more time than acquisition of factual, conceptual, and procedural knowledge. Moreover, learning on higher levels makes necessary to learn on lower levels first. • Time needed for learning cannot be shortened by skipping acquisition of canonical or strategic knowledge, since this would hamper comprehension.
Conclusions • Since learned patterns are not transferred into long-term memory, if they are not repeated sufficiently often, superficially acquired canonical knowledge will be forgotten. • Thus, prematurely aborted learning processes waste time, since later re-learning must compensate for forgotten material (example: engineering maths in a bachelor-course! In a master-course, on advanced maths, time is wasted to re-install canonical knowledge). • Neural networks are non-linear. Two nonlinear nested feedback loops (with more than two stores) might react completely differently on the same input depending on the inner state of the system.This explains why two learners with same IQ and virtually the same education might react completely differently in the same course.
4. Summary Based on methods found in the theory of (non-linear) feedback-control loops, a new model of learning has been used to explain consistently many learning effects. In particular, it could be explained, why achieving comprehension needs more time than learning by rote.