I gave my talk "Logic and computation as biophysics" yesterday. Here is a picture of me making the presentation. I have an imperfect recording of the presentation that will be made available through IASE. If you are interested in access to the recording request that access through me.
Since my May update I have made numerous small revisions to my book's draft "Introductory Remarks." These revisions are mainly derived from feedback on a related submission document to SuperComputing 2011 that comes from a chapter of the book on the Foundations of Computation.
In that submission to the Disruptive Technologies Program I propose a new computational paradigm for certain large scale parallel computing problems so the reviewers were all leading Computer Scientists. This work is not generally available but I did want to propagate some of the important elements of the review back into the Introductory Remarks.
The primary additions from the review relate to some relative comments with respect to Connectionism and Neural Networks.
Honestly, despite its popularity among Neuroscientists, I had dismissed Connectionism long ago and view Neural Networks as a failed technology founded upon a model that is naive and poorly informed.
However, several senior reviewers criticized my failure to mention it directly and, worse, suggested that I had not done adequate research because it was absent.
The truth is that during the late 1980s and early 1990s, before I embarked upon this work, I spent a lot of time with the Churchland work and with McClelland's "Distributed Parallel Processing" text and thought that we had all moved on from what had become a footnote in history :-)
This was wrong of me and I should have included Connectionism in my review, at least to compare my model of distributed representation with the earlier work.
In short, the problem with Connectionism and Neural Networks is that they were founded before a lot of the more detailed evidence became available: they are naive, not founded upon a general theory, they deal with "brains" not organisms, and they ignore physical structure.
My decomposition arguments also hold. Neural Networks do not form a computational parallelism that makes a difference since the parallelism in Neural Networks can be removed (the node computations executed in sequence) without a discernible effect upon the results.
I have started to correct my deficit in the "Introductory Remarks" and, for historical completeness, will add a discussion of Connectionism in a new section of the book. This sort of evolution is why books should always be well reviewed before publication. It's why I publish drafts of my evolving work online.
Although I get few real review comments unless I pointedly ask someone to do it. I tend instead to hear only from people that either love the work and it has changed their entire view of the world (this is nice but not especially useful) or from crazy people that claim that I have stolen their ideas. So far it turns out that such people understand neither their own work nor mine, and precedence is easily established via the Archive.org record.
I also took the opportunity over the past day or two to refine the Introductory description of the model of memory and recognition. This is mostly word craft and clean up of late night bumbling, it is now a clearer technical introduction I hope.
Some controversy online over the claims of a proof that P != NP by Deolalikar of HP Labs. Here is the latest revision of the paper:
Alasdair Urquhart's insightful response on FOM can be trusted:http://www.cs.nyu.edu/pipermail/fom/2010-August/014981.html
A more metered response comes from Dick Lipton and Ken Regan: Issues In The Proof That P≠NP I will read the paper but given the responses so far my expectations are low for this proof. Is it true that P != NP? Harvey Friedman claims it is so. I've never been clear on the matter myself but in terms of the limits of current computational models the question is an important one. Note that I place emphasis on the word "current." A proof will be important in the development of future computational models if one can demonstrate why it is so in one case and not the other.Here's a link to the guy, Stephen Cook, that got us into this mess:
Stephen Cook And here is a link to his paper that started it all: The complexity of theorem-proving proceduresUPDATE:
Deolalikar responds to critics:
Deolalikar Responds To Issues About His P≠NP Proof
This posting by Richard Lipton provides some useful comments on a strategy for assessing proofs.
But no-one is buying into spending a lot of time on the question:
And now the ultimate indignity
FINAL UPDATE:
Hopefully this is the last we will hear of it except in the annals of foolhardy Internet career moves:
