ABSTRACT:
	A physiological explanation of the long-term memory effect, 
Retroactive Interference, is proposed.  We consider associative memory to 
be a dynamical system of cortical spatial patterns, manifested in the electrical 
fields of dendritic currents measured via electroencephalographic electrodes. 
Stimulus input causes state transitions with convergence toward a template 
pattern. In associative memories, new pattern storage affects all previously 
stored patterns.  We believe RI is caused by accumulation of these changes, 
leading to failure of the system to converge toward a prior template pattern.  
Recordings from rabbit visual, somatic, and auditory cortices demonstrate 
pattern evolution over time with concomitant imposed learning.

SUMMARY:

	Retroactive Interference, (RI) also called retrograde inhibition, is a 
well studied psychological phenomenon in both humans and animals.  
Typically studies have shown that recall performance for memories 
previously stored are negatively affected by later learning.  Many hypothesis 
for this phenomena have been suggested.  Among those commonly cited are 
memory address retrieval failure, memory trace decay, and competition.  
Scientists since Aristotle have been attempting to model forgetting in an 
effort to understand memory and the brain.  Aristotle believed that memories 
undergo a passive decay over time.  This hypothesis was embraced by E. L. 
Thorndike in 1914, who proposed a 'law of disuse' as the definitive factor 
of trace decay theory.  It was the German scholar Hermann Ebbinghaus 
who first described Retroactive Interference in 1878.  He created nonsense 
syllables and then methodically applied them to the study of memory recall 
performance.   Using sequentially memorized lists of the syllables, he 
showed that forgetting could be attributed to the interfering effects of 
subsequently learned matter.  His basic finding has been widely repeated 
and substantiated with a wide variety of human and animal subjects, 
learning and testing conditions.  One predominant explanation for this 
finding is the two-factor theory of forgetting, which proposes that inhibition 
is due to both competition and unlearning.  Specifically, forgetting is 
attributed to competition among memory traces during recall, and to 
unlearning of the original memory traces due to the subsequent learning.  
The unlearning hypothesis is equivalent to the experimental extinction found 
in animal studies.  These and many modern cognitive psychology theories 
employ analogous or metaphoric structures, such as digital computers, to 
model the psychophysical data.  It is unlikely the brain uses such 
computational algorithms.

	We are proposing another mechanism by which retroactive 
interference may occur.  Findings derived from many years of research into 
the nature of perception in mammals has lead us to a new view of memory.  
We base our hypothesis upon empirical observations of overt natural 
behavior and correlated real-time electrophysiological events.  We have 
found that cortical spatial amplitude patterns of a common carrier wave are 
the conveyor of perceptual memories and stimulus meaning.  Associative 
memories in the mammalian brain are carried by globally distributed, 
chaotic, self organized dynamic activity patterns.  Information retrieval is 
accomplished by external stimulus input causing the neural assembly to 
converge by state transition toward a specific, identifiable template pattern.  
This pattern is internally generated, and not directly dependent upon the 
external stimulus input pattern.  It is profoundly influenced by the collection 
of past experiences of a particular animal.  Memories stored as global 
attractor patterns manifest primarily internally generated stimuli meaning, 
rather than strictly transduced sensory data.  Learning arises via Hebbian 
style synaptic strength changes influenced by correlated pattern output.  
Each learned class is a constraining of the dynamics of the cortical pattern 
into a stable, reproducible wing.  Novel wings are formed by alterations of 
intracortical synapses during learning.  Simultaneous excitation between 
neurons in the nerve cell assemblies are where new associations are 
integrated with all past associations.

	Three Albino New Zealand female adult rabbits were chronically 
implanted with sixty four channel (8x8 with inter-electrode distance of 
.8mm) EEG surface electrode arrays.  The implant sites included the 
Somatosensory,  Visual, and Auditory cortex.  Computers recorded 
6000ms of broadband (1-100 Hertz) electroencephalographic waves from 
familiarized animals in an awake, immobile state.  No stimulus was 
delivered during recordings, however classical conditioning was performed 
before and midway through the duration of the experiment.  Each recording 
session was held twice a week, with 22 total trials.  Spectral decomposition, 
temporal filtering, and multivariate analysis of EEG segments were used to 
extract common wave form spatial amplitude patterns.  Euclidean distance 
measures are calculated for the root mean square (RMS) carrier wave 
patterns generated from each weekly trial and used for tracking the evolution 
of carrier wave spatial amplitude patterns over time and between trials.

	The purpose of the study was to examine how electrophysiological 
(EEG) amplitude patterns evolve over time and with learning.  A variety of 
factors can and do influence both the pattern and its rate of change.  Several 
studies by Freeman have shown that in addition to the particular stimuli 
causing a subject to generate a specific pattern, the meaning of the stimuli, 
the environments it is given in, and the arousal state of the animal, can all 
have a profound influence upon the pattern.  Our findings showed patterns 
typically mutate 5-15% from their starting points over the time span of a 
week.  Prior to learning being imposed, the average intertrial change was 
10%.   The conditioning stimulus modality was changed for each animal 
after the 9th trial.  The average percentile change between trial patterns 
where novel training was imposed jumped to 32% These findings are in 
concert with the theory that brains function with interconnected and 
distributed neural networks.  In this associative view all new memory 
networks should have a small but noticeable effect on all previous memory 
stores.  Our experimental findings of the evolution of EEG amplitude 
patterns are complementary to this theoretical framework.

	Forgetting and RI phenomena can be attributed to a difficulty in 
achieving or replicating previously shaped global cortical spatiotemporal 
patterns.  The difficulty arises due to several factors.  As new associative 
memories are formed in associative Hebbian networks, all previously stored 
memories are subtlety altered by the new connections.  The original 
information is not written over, lost, or in competition with each other, but 
rather it is slowly modified over time, as the landscape of a mountain range 
is altered by weather and erosion over time.  The exact point it becomes 
something new is not clear, and it will always retain some of its original 
character.  

	In an associative memory, changes in any previously established 
memory, stored as spatial pattern, will have a small but significant impact 
upon all of the previously stored patterns.  This characteristic is what allows 
an organism to bring all previous experiences to bear in analyzing a given 
stimulus input.  Forgetting or retrieval failure or error  can be caused by 
accumulation of these pattern changes due to extensive learning over time.  
These changes may lead to failure of the system to converge toward the 
correct template pattern.  Experiments performed on three cortical implant 
sites in rabbits (visual, olfactory, and auditory) have provided data showing 
that these memory carrying patterns gradually evolve over time, and the rate 
of pattern change is influenced by imposed learning.

References

[1]	Baird, B., Nonlinear dynamics of pattern formation and pattern 
recognition in the rabbit olfactory bulb, Physica D. 22, 150-77 (1986)

[2]	Freeman, W. J., Mass action in the nervous system, (1975)
 
[3]	Freeman, W. J.; Schneider, W., Changes in Spatial Patterns of 
Rabbit Olfactory EEG with Conditioning to Odors, 19 (1), 44 (1982)

[4]	Freeman, W. J.; Characterization of State Transitions in Spatially 
Distributed, Chaotic, Nonlinear, Dynamical Systems in Cerebral Cortex, 29 
(3), 294-306 (1994)