Ben Goertzel wrote:
Hi,
BTW, I just read this paper
For example, in Loosemore & Harley (in press) you can find an analysis of a
paper by Quiroga, Reddy, Kreiman, Koch, and Fried (2005) in which the latter
try to claim they have evidence in favor of grandmother neurons (or sparse
collections of grandmother neurons) and against the idea of distributed
representations.
which I found at
http://www.vis.caltech.edu/~rodri/
and I strongly disagree that
We showed their conclusion to be incoherent. It was deeply implausible,
given the empirical data they reported.
The claim that Harley and I made - which you quote above - was the
*conclusion* sentence that summarized a detailed explanation of our
reasoning.
That reasoning was in our original paper, and I also went to the trouble
of providing a longer version of it in one of my last posts on this
thread. I showed, in that argument, that their claims about sparse vs
distributed representations were incoherent, because they had not
thought through the implications contained in their own words - part of
which you quote below.
Merely quoting their words again, without resolving the inconsistencies
that we pointed out, proves nothing.
We analyzed that paper because it was one of several that engendered a
huge amount of publicity. All of that publicity - which, as far as we
can see, the authors did not have any problem with - had to do with the
claims about grandmother cells, sparseness and distributed
representations. Nobody - not I, not Harley, and nobody else as far as
I know - disputes that the empirical data were interesting, but that is
not the point: we attacked their paper because of their conclusion
about the theoretical issue of sparse vs distributed representations,
and the wider issue about grandmother cells. In that context, it is not
true that, as you put it below, the authors "only [claimed] to have
gathered some information on empirical constraints on how neural
knowledge representation may operate". They went beyond just claiming
that they had gathered some relevant data: they tried to say what that
data implied.
Richard Loosemore
Their conclusion, to quote them, is that
"
How neurons encode different percepts is one of the most intriguing
questions in neuroscience. Two extreme hypotheses are
schemes based on the explicit representations by highly selective
(cardinal, gnostic or grandmother) neurons and schemes that rely on
an implicit representation over a very broad and distributed population
of neurons1–4,6. In the latter case, recognition would require the
simultaneous activation of a large number of cells and therefore we
would expect each cell to respond to many pictures with similar basic
features. This is in contrast to the sparse firing we observe, because
most MTL cells do not respond to the great majority of images seen
by the patient. Furthermore, cells signal a particular individual or
object in an explicit manner27, in the sense that the presence of the
individual can, in principle, be reliably decoded from a very small
number of neurons.We do not mean to imply the existence of single
neurons coding uniquely for discrete percepts for several reasons:
first, some of these units responded to pictures of more than one
individual or object; second, given the limited duration of our
recording sessions, we can only explore a tiny portion of stimulus
space; and third, the fact that we can discover in this short time some
images—such as photographs of Jennifer Aniston—that drive the
cells suggests that each cell might represent more than one class of
images. Yet, this subset of MTL cells is selectively activated by
different views of individuals, landmarks, animals or objects. This
is quite distinct from a completely distributed population code and
suggests a sparse, explicit and invariant encoding of visual percepts in
MTL.
"
The only thing that bothers me about the paper is that the title
"
Invariant visual representation by single neurons in
the human brain
"
does not actually reflect the conclusions drawn. A title like
"
Invariant visual representation by sparse neuronal population encodings
the human brain
"
would have reflected their actual conclusions a lot better. But the paper's
conclusion clearly says
"
We do not mean to imply the existence of single
neurons coding uniquely for discrete percepts for several reasons:
"
I see some incoherence between the title and the paper's contents,
which is a bit frustrating, but no incoherence in the paper's conclusion,
nor between the data and the conclusion.
According to what the paper says, the authors do not claim to have
solve the neural knowledge representation problem, but only to have
gathered some information on empirical constraints on how neural
knowledge representation may operate.
-- Ben G
-------------------------------------------
agi
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/
Modify Your Subscription: https://www.listbox.com/member/?&;
Powered by Listbox: http://www.listbox.com
-------------------------------------------
agi
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/
Modify Your Subscription:
https://www.listbox.com/member/?member_id=8660244&id_secret=120640061-aded06
Powered by Listbox: http://www.listbox.com
