FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W2034794057> ?p ?o ?g. }

• W2034794057 endingPage "502" @default.
• W2034794057 startingPage "500" @default.
• W2034794057 abstract "In redundant neural networks, many different combinations of connection weights will produce the same output, thereby providing many possible solutions for a given computation. In this issue of Neuron, Rokni et al. propose that the arm movement representations in the cerebral cortex act like redundant networks that drift randomly between different synaptic configurations with equivalent input-output behavior because of random noise in the adaptive learning mechanism. In redundant neural networks, many different combinations of connection weights will produce the same output, thereby providing many possible solutions for a given computation. In this issue of Neuron, Rokni et al. propose that the arm movement representations in the cerebral cortex act like redundant networks that drift randomly between different synaptic configurations with equivalent input-output behavior because of random noise in the adaptive learning mechanism. The arm and hand contain a few dozen muscles innervated by a few thousand spinal motor neurons, but tens of millions of neurons in several motor cortical areas contribute to their control. Because of this redundancy, it is theoretically possible for the response properties of single cortical neurons to change with time while the global cortical output signal continues to produce the same movements. Rokni et al., 2007Rokni U. Richardson A.G. Bizzi E. Seung H.S. Neuron. 2007; 54 (this issue): 653-666Abstract Full Text Full Text PDF PubMed Scopus (153) Google Scholar present evidence that the motor representations in the primary motor cortex (M1) and the supplementary motor area (SMA) of monkeys vary with time at the single-neuron level. They then show that a simple redundant neural network with a noisy learning mechanism will form a motor representation that is variable but still produces accurate movements and adapts to external forces. During neural recordings, one sometimes observes gradual changes in the activity of a neuron, such as its baseline tonic rate or its response to a stimulus, over the course of minutes or hours. The origin of these changes is usually unknown, but the analytical complications they present can often be managed by such strategies as randomized task designs. In some studies, however, gradual drifts in neural activity over time are a problem. Suppose that one wants to study how neural activity changes while an animal learns a new stimulus-response rule. If neural responses are inherently stable, then one can assume that all observed activity changes are related to the learning process. However, if neural responses are not stable, an indefinable part of the observed changes are not learning related, which confounds their interpretation. A case in point was a recent series of studies of neural activity in M1 and SMA of monkeys while they adapted to external forces during arm movements (Li et al., 2001Li C.S.R. Padoa-Schioppa C. Bizzi E. Neuron. 2001; 30: 593-607Abstract Full Text Full Text PDF PubMed Scopus (322) Google Scholar, Padoa-Schioppa et al., 2004Padoa-Schioppa C. Li C.S.R. Bizzi E. J. Neurophysiol. 2004; 91: 449-473Crossref PubMed Scopus (97) Google Scholar). During each daily learning session, monkeys made reaching movements in different directions in three consecutive blocks of 160 trials: (1) a baseline block with no external forces; (2) an adaptation block during which a velocity-dependent “viscous-curl” field pushed on the arm orthogonal to the direction of movement; (3) a washout block without forces. The monkeys quickly adapted to the curl field to straighten out their hand paths during the adaptation block and then readapted to the absence of forces in the washout block. Neurons in M1 and SMA have broad directional tuning curves centered on a particular preferred movement direction. During the learning sessions, neurons showed a variety of patterns of directional tuning changes across trial blocks, which suggested that they were distinct functional classes of neurons that contributed to different aspects of motor learning. The preferred direction of some neurons rotated from the baseline to the adaptation block and then rotated back toward their original tuning during washout, consistent with a role in compensating for the forces (“dynamic” neurons). The tuning curves of other neurons showed only minor changes in all three blocks (“kinematic” neurons). Finally, some neurons changed their tuning curves either only in the adaptation block or the washout block and never returned to their baseline tuning, as if they stored information about the adaptation and readaptation episodes (“memory” neurons). However, the tuning curves of kinematic and dynamic neurons also usually showed some residual change between the baseline and washout blocks. Furthermore, when the monkeys performed three consecutive “baseline” blocks without external forces, many M1 and SMA neurons still showed response changes between the first and last blocks that were as large as those seen over the same time frame in the learning session (Li et al., 2001Li C.S.R. Padoa-Schioppa C. Bizzi E. Neuron. 2001; 30: 593-607Abstract Full Text Full Text PDF PubMed Scopus (322) Google Scholar, Padoa-Schioppa et al., 2004Padoa-Schioppa C. Li C.S.R. Bizzi E. J. Neurophysiol. 2004; 91: 449-473Crossref PubMed Scopus (97) Google Scholar, Rokni et al., 2007Rokni U. Richardson A.G. Bizzi E. Seung H.S. Neuron. 2007; 54 (this issue): 653-666Abstract Full Text Full Text PDF PubMed Scopus (153) Google Scholar). Rokni et al., 2007Rokni U. Richardson A.G. Bizzi E. Seung H.S. Neuron. 2007; 54 (this issue): 653-666Abstract Full Text Full Text PDF PubMed Scopus (153) Google Scholar argue that these response changes across time even in constant task environments show that cortical motor representations are inherently variable and that learning processes are superimposed on that unstable background. To support this hypothesis, Rokni et al., 2007Rokni U. Richardson A.G. Bizzi E. Seung H.S. Neuron. 2007; 54 (this issue): 653-666Abstract Full Text Full Text PDF PubMed Scopus (153) Google Scholar present a neural network model with two input units, a highly redundant middle layer of 10,000 “neurons,” and two output units. The input and output units coded the (x,y) components of the direction of movement and forces, respectively. The connection weights from the input units to the neurons were adjusted after each trial by a learning signal that attempted to minimize the sensed error in that trial and also by random noise that was uncorrelated across all connections. Rokni et al., 2007Rokni U. Richardson A.G. Bizzi E. Seung H.S. Neuron. 2007; 54 (this issue): 653-666Abstract Full Text Full Text PDF PubMed Scopus (153) Google Scholar show that this network can adapt to a force field and then readapt when the field is removed. During the learning session, many network neurons undergo correlated rotations of their directional tuning across blocks, similar to M1 and SMA dynamic neurons, despite the random noise in the learning mechanism. When performing three consecutive baseline blocks, the noise causes uncorrelated random changes in the responses of single neurons across blocks, but the network still generates the desired outputs. This latter finding indicates that many different specific configurations of the motor representation can produce the same motor output and so comprise an “optimal manifold” of functionally equivalent representations. The motor representation meanders within the confines of this range of functionally equivalent states, due to the continual interplay between random noise and error-based learning signals. One intriguing implication of this model is that the motor cortex is in a dynamic equilibrium of continual flux even in a constant environment. Small motor errors are converted to learning signals to stay within the range of adequate motor representations for that environment by limiting the drift caused by the intrinsic noise in the adaptive process. When the learning signal is blocked, Rokni et al., 2007Rokni U. Richardson A.G. Bizzi E. Seung H.S. Neuron. 2007; 54 (this issue): 653-666Abstract Full Text Full Text PDF PubMed Scopus (153) Google Scholar show that the response properties of neurons drift unchecked and the performance of the network rapidly degrades. The same feedback-based learning signal that keeps the motor representation within the manifold for one environment causes it to adapt when the environment changes. The consistent trial-to-trial bias in the learning signal caused by the change in conditions drives the motor representation toward a new range of acceptable configurations despite the random noise element. Whether the motor representation drifts within the manifold of equivalent states or adapts by evolving toward a new range of states is determined by the trial-to-trial statistics of the error-based learning signal. Another important implication is that some neural response changes are behaviorally irrelevant and may even be functionally deceiving. Rokni et al., 2007Rokni U. Richardson A.G. Bizzi E. Seung H.S. Neuron. 2007; 54 (this issue): 653-666Abstract Full Text Full Text PDF PubMed Scopus (153) Google Scholar make the provocative suggestion that different response patterns (e.g., kinematic, dynamic) do not arise from separate functional classes of neurons. Instead, they argue that the pattern of changes in activity of a given neuron with time is largely an epiphenomenon shaped by the unique history of noise- and error-driven changes in synaptic input during each learning session. Their model makes some testable predictions. First, the behavior of a neuron could change randomly between kinematic, dynamic, and memory patterns in different learning sessions. Less evident but equally important, the model requires that the change in synaptic efficacy caused by noise and by the learning signal on any given trial should be of similar size. If the learning signals have a substantially larger effect than the noise term, most neurons would show dynamic properties. In contrast, if noise and learning signals have a similar impact on synaptic strengths on M1 neurons but the neurons still show consistent patterns of response changes across different learning sessions, then there must be some underlying constraint on their behavior, which would support the existence of distinct functional classes. A third prediction is that as a task becomes more demanding, there will be a decrease in the range of acceptable variability of the motor representation and of noise-related neural activity changes. The conclusions drawn by this study are only as valid as the neural data that motivated it. The sizeable degree of variability of single-neuron activity in the motor cortex over short time frames observed these studies, even in constant task conditions (Li et al., 2001Li C.S.R. Padoa-Schioppa C. Bizzi E. Neuron. 2001; 30: 593-607Abstract Full Text Full Text PDF PubMed Scopus (322) Google Scholar, Padoa-Schioppa et al., 2004Padoa-Schioppa C. Li C.S.R. Bizzi E. J. Neurophysiol. 2004; 91: 449-473Crossref PubMed Scopus (97) Google Scholar, Rokni et al., 2007Rokni U. Richardson A.G. Bizzi E. Seung H.S. Neuron. 2007; 54 (this issue): 653-666Abstract Full Text Full Text PDF PubMed Scopus (153) Google Scholar) has not been documented before. It is essential to confirm this important finding in further studies in a range of tasks. Some of the neural results are not easily explained by random noise. For instance, the neurons showed a progressive increase in average discharge rate with each successive trial block (Li et al., 2001Li C.S.R. Padoa-Schioppa C. Bizzi E. Neuron. 2001; 30: 593-607Abstract Full Text Full Text PDF PubMed Scopus (322) Google Scholar, Padoa-Schioppa et al., 2004Padoa-Schioppa C. Li C.S.R. Bizzi E. J. Neurophysiol. 2004; 91: 449-473Crossref PubMed Scopus (97) Google Scholar). This trend suggests the presence of a process that has a cumulative effect rather than a random effect throughout the recording period. This could be related to fatigue or the trauma inflicted by the electrode on local synaptic circuitry, or other factors. This cumulative process may have also contributed to the directional tuning changes across time. Furthermore, some of the examples of time-dependent response changes in M1 are from neurons with very low peak discharge rates (Rokni et al., 2007Rokni U. Richardson A.G. Bizzi E. Seung H.S. Neuron. 2007; 54 (this issue): 653-666Abstract Full Text Full Text PDF PubMed Scopus (153) Google Scholar). This suggests that these neurons were on the margins of the task-related population and made only a minor contribution to task performance. While the activity changes in those neurons might be statistically significant, it is not clear to what degree they are functionally significant or provide strong evidence for an unstable motor representation. Despite these reservations, the study by Rokni et al., 2007Rokni U. Richardson A.G. Bizzi E. Seung H.S. Neuron. 2007; 54 (this issue): 653-666Abstract Full Text Full Text PDF PubMed Scopus (153) Google Scholar is important. It draws attention to the possibility that the movement representation in the motor cortex is not as stable as is generally assumed. It presents some interesting speculations about the implications of an unstable redundant motor representation on motor cortex function. It suggests that sensory signals from the periphery are used not only for feedback correction for errors and to guide motor learning, but also to maintain the motor representation within a range of equivalent functional states against debilitating drift caused by stochastic noise in adaptive components of the system. Finally, it makes strong predictions that should be readily testable by experiments. Motor Learning with Unstable Neural RepresentationsRokni et al.NeuronMay 24, 2007In BriefIt is often assumed that learning takes place by changing an otherwise stable neural representation. To test this assumption, we studied changes in the directional tuning of primate motor cortical neurons during reaching movements performed in familiar and novel environments. During the familiar task, tuning curves exhibited slow random drift. During learning of the novel task, random drift was accompanied by systematic shifts of tuning curves. Our analysis suggests that motor learning is based on a surprisingly unstable neural representation. Full-Text PDF Open Archive" @default.
• W2034794057 created "2016-06-24" @default.
• W2034794057 creator A5009601137 @default.
• W2034794057 creator A5084480226 @default.
• W2034794057 date "2007-05-01" @default.
• W2034794057 modified "2023-10-15" @default.
• W2034794057 title "Is the Movement Representation in the Motor Cortex a Moving Target?" @default.
• W2034794057 cites W1964856529 @default.
• W2034794057 cites W2072583006 @default.
• W2034794057 cites W2130672873 @default.
• W2034794057 doi "https://doi.org/10.1016/j.neuron.2007.05.007" @default.
• W2034794057 hasPubMedId "https://pubmed.ncbi.nlm.nih.gov/17521562" @default.
• W2034794057 hasPublicationYear "2007" @default.
• W2034794057 type Work @default.
• W2034794057 sameAs 2034794057 @default.
• W2034794057 citedByCount "3" @default.
• W2034794057 countsByYear W20347940572014 @default.
• W2034794057 countsByYear W20347940572017 @default.
• W2034794057 crossrefType "journal-article" @default.
• W2034794057 hasAuthorship W2034794057A5009601137 @default.
• W2034794057 hasAuthorship W2034794057A5084480226 @default.
• W2034794057 hasBestOaLocation W20347940571 @default.
• W2034794057 hasConcept C121332964 @default.
• W2034794057 hasConcept C15744967 @default.
• W2034794057 hasConcept C169760540 @default.
• W2034794057 hasConcept C17744445 @default.
• W2034794057 hasConcept C180747234 @default.
• W2034794057 hasConcept C188147891 @default.
• W2034794057 hasConcept C199539241 @default.
• W2034794057 hasConcept C24890656 @default.
• W2034794057 hasConcept C24998067 @default.
• W2034794057 hasConcept C2776359362 @default.
• W2034794057 hasConcept C2777348757 @default.
• W2034794057 hasConcept C2778373776 @default.
• W2034794057 hasConcept C2780226923 @default.
• W2034794057 hasConcept C41008148 @default.
• W2034794057 hasConcept C46312422 @default.
• W2034794057 hasConcept C6757342 @default.
• W2034794057 hasConcept C94625758 @default.
• W2034794057 hasConceptScore W2034794057C121332964 @default.
• W2034794057 hasConceptScore W2034794057C15744967 @default.
• W2034794057 hasConceptScore W2034794057C169760540 @default.
• W2034794057 hasConceptScore W2034794057C17744445 @default.
• W2034794057 hasConceptScore W2034794057C180747234 @default.
• W2034794057 hasConceptScore W2034794057C188147891 @default.
• W2034794057 hasConceptScore W2034794057C199539241 @default.
• W2034794057 hasConceptScore W2034794057C24890656 @default.
• W2034794057 hasConceptScore W2034794057C24998067 @default.
• W2034794057 hasConceptScore W2034794057C2776359362 @default.
• W2034794057 hasConceptScore W2034794057C2777348757 @default.
• W2034794057 hasConceptScore W2034794057C2778373776 @default.
• W2034794057 hasConceptScore W2034794057C2780226923 @default.
• W2034794057 hasConceptScore W2034794057C41008148 @default.
• W2034794057 hasConceptScore W2034794057C46312422 @default.
• W2034794057 hasConceptScore W2034794057C6757342 @default.
• W2034794057 hasConceptScore W2034794057C94625758 @default.
• W2034794057 hasIssue "4" @default.
• W2034794057 hasLocation W20347940571 @default.
• W2034794057 hasLocation W20347940572 @default.
• W2034794057 hasOpenAccess W2034794057 @default.
• W2034794057 hasPrimaryLocation W20347940571 @default.
• W2034794057 hasRelatedWork W134480934 @default.
• W2034794057 hasRelatedWork W1983975291 @default.
• W2034794057 hasRelatedWork W1989883124 @default.
• W2034794057 hasRelatedWork W1997568039 @default.
• W2034794057 hasRelatedWork W2016252867 @default.
• W2034794057 hasRelatedWork W2038514344 @default.
• W2034794057 hasRelatedWork W2082639675 @default.
• W2034794057 hasRelatedWork W2084925968 @default.
• W2034794057 hasRelatedWork W2473912130 @default.
• W2034794057 hasRelatedWork W4205378765 @default.
• W2034794057 hasVolume "54" @default.
• W2034794057 isParatext "false" @default.
• W2034794057 isRetracted "false" @default.
• W2034794057 magId "2034794057" @default.
• W2034794057 workType "article" @default.