Challenging the widespread assumption that connectionism and distributed representations go hand-in-hand. Bowers, J., S.
Challenging the widespread assumption that connectionism and distributed representations go hand-in-hand [pdf]Paper  abstract   bibtex   
One of the central claims associated with the parallel distributed processing approach popularized by D. E. Rumelhart, J. L. McClelland and the PDP Research Group is that knowledge is coded in a distributed fashion. Localist representations within this perspective are widely rejected. It is important to note, however, that connectionist networks can learn localist representations and many connectionist models depend on localist coding for their functioning. Accordingly, a commitment to distributed representations within the connectionist camp should be considered a specific theoretical claim regarding the structure of knowledge rather than a core principle, as often assumed. In this article, it is argued that there are fundamental computational and empirical challenges that have not yet been addressed by distributed connectionist theories that are readily accommodated within localist approaches. It is concluded that the common rejection of localist coding schemes within connectionist architectures is unwarranted. distributed vs. localist representations 3 Challenging the widespread assumption that connectionism and distributed representations go hand-in-hand. Since the publication of the two volume set Parallel Distributed Processing (McClelland, Rumelhart, & the PDP Research Group 1986; Rumelhart, McClelland, & the PDP Research Group, 1986), connectionist models have played a central role in theorizing about perception, memory, language, and cognition more generally. This approach has reintroduced learning as a core constraint in theory development, it shows promise of identifying a set of general principles that apply across a wide range of cognitive domains, and it provides a possible bridge between theories of cognition and the neural structures that mediate these functions. And by linking theories of cognition with theories of learning and neurobiology, connectionism holds the promise of identifying principled constraints as to why cognitive systems are organized the way they are as opposed to some other plausible alternatives; that is, this approach shows promise of supporting explanatory rather than descriptive theories (Seidenberg, 1993a). One claim often associated with this framework is that knowledge is coded in a distributed fashion. That is, information is coded as a pattern of activation across many processing units, with each unit contributing to many different representations. Indeed, at this point, the concepts connectionism and distributed representations are so strongly associated that distributed representations are sometimes described as one of the core " connectionist principles " (e.g., Seidenberg, 1993b, p. 300). This challenges one of the fundamental assumptions of more traditional " symbolic " approaches to cognitive theorizing according to which knowledge is coded in a localist format, with separate representations coding for distinct pieces of information.
@article{
 title = {Challenging the widespread assumption that connectionism and distributed representations go hand-in-hand},
 type = {article},
 id = {890d5f26-bc32-3c3b-8228-8fdff1beec9a},
 created = {2017-09-01T15:53:36.889Z},
 file_attached = {true},
 profile_id = {80da7853-f7b7-36a9-8e4c-d7ddb2d9e538},
 group_id = {a2333ea3-15a4-3d40-8d36-f0d9590ca926},
 last_modified = {2017-09-01T15:53:36.996Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 abstract = {One of the central claims associated with the parallel distributed processing approach popularized by D. E. Rumelhart, J. L. McClelland and the PDP Research Group is that knowledge is coded in a distributed fashion. Localist representations within this perspective are widely rejected. It is important to note, however, that connectionist networks can learn localist representations and many connectionist models depend on localist coding for their functioning. Accordingly, a commitment to distributed representations within the connectionist camp should be considered a specific theoretical claim regarding the structure of knowledge rather than a core principle, as often assumed. In this article, it is argued that there are fundamental computational and empirical challenges that have not yet been addressed by distributed connectionist theories that are readily accommodated within localist approaches. It is concluded that the common rejection of localist coding schemes within connectionist architectures is unwarranted. distributed vs. localist representations 3 Challenging the widespread assumption that connectionism and distributed representations go hand-in-hand. Since the publication of the two volume set Parallel Distributed Processing (McClelland, Rumelhart, & the PDP Research Group 1986; Rumelhart, McClelland, & the PDP Research Group, 1986), connectionist models have played a central role in theorizing about perception, memory, language, and cognition more generally. This approach has reintroduced learning as a core constraint in theory development, it shows promise of identifying a set of general principles that apply across a wide range of cognitive domains, and it provides a possible bridge between theories of cognition and the neural structures that mediate these functions. And by linking theories of cognition with theories of learning and neurobiology, connectionism holds the promise of identifying principled constraints as to why cognitive systems are organized the way they are as opposed to some other plausible alternatives; that is, this approach shows promise of supporting explanatory rather than descriptive theories (Seidenberg, 1993a). One claim often associated with this framework is that knowledge is coded in a distributed fashion. That is, information is coded as a pattern of activation across many processing units, with each unit contributing to many different representations. Indeed, at this point, the concepts connectionism and distributed representations are so strongly associated that distributed representations are sometimes described as one of the core " connectionist principles " (e.g., Seidenberg, 1993b, p. 300). This challenges one of the fundamental assumptions of more traditional " symbolic " approaches to cognitive theorizing according to which knowledge is coded in a localist format, with separate representations coding for distinct pieces of information.},
 bibtype = {article},
 author = {Bowers, Jeffrey S}
}
Downloads: 0