Hi David, Thanks for starting this thread - I think you're right that it would be useful to share ideas coming from the various implementations.
Just to avoid confusion, Felix has done all the work on Comportex and ComportexViz; I can claim no credit for the huge amount of excellent work on that project. It's so well implemented that it makes more sense for me to contribute to Felix' project than to try and re-implement it. One thing that Clortex brings is the idea of using a high-performance persistent immutable database (Datomic) with independent peer processes operating on and visualising the state of the HTM model. This is the level at which I'm currently working - we'll likely take advantage of Felix' code to do the actual algorithmic part of the system. I have yet to deeply examine what Eric is doing with his Continuous HTM system, but it looks like he's making some real progress in applying it to some "traditional" machine learning applications. I'll take a really close look at his code shortly, as I would like to assess how closely his methods match up with HTM theory. One great thing about CHTM is that the code is really, really short and simple (especially compared with the Python and C++ of NuPIC and the Java of htm.java!). Regards, Fergal Byrne On Tue, Jan 6, 2015 at 5:27 PM, David Ragazzi <[email protected]> wrote: > Matt, > > > But it's probably out of date and very sparse on details. > > Thanks for this, Matt. I sent this message, just because the wiki is > outdated and a ML discussion could be interesting to people expose the > _why_ their implementation has something extra over Nupic and others. > > Aseem, > > > By implementation did u mean actual programming algorithm? > > Yes, I did. It's because algorithms is naturally the final product of a > cognitive model. Actually, you don't need enter into details in your > answer, you could simply mention why your implementation is interesting and > links to documentation (as you did). For example, Fergal and Eric have > their own implementations of HTM where they add other cognitive features > already present on deep learning networks and others. Usually they post > their major breakthroughs here or at gitter. > > Keith and Chandan, > > Yes, a benchmarking would be interesting! We could have our own > competition examples (or re-use other cognitive chalenge) to test what HTM > implementation is better to perform such example. > > Once an implementation excel over others we could check its design was the > decisive factor and whether it's enough closer to biology to discuss its > relevance to Nupic. :-D > > > > > > On 6 January 2015 at 14:45, Matthew Taylor <[email protected]> wrote: > >> We have a list of HTM implementations here: >> https://github.com/numenta/nupic/wiki/Other-HTM-CLA-projects >> >> But it's probably out of date and very sparse on details. If anyone >> wants to provide more details about individual projects (especially >> developments status, features), feel free to edit the wiki. >> --------- >> Matt Taylor >> OS Community Flag-Bearer >> Numenta >> >> >> On Tue, Jan 6, 2015 at 4:00 AM, David Ragazzi <[email protected]> >> wrote: >> > Hi guys, >> > >> > It is known that many here have implemented their own versions of HTM. >> Some >> > implementations have features that even Nupic implementation still >> doesn't >> > have. So I think is a healthy discussion you share the description of >> > implementations, and who knows we take the best from each one and >> replicate >> > these features on Nupic (if biologically plausible, of course). This >> would >> > accelerate the process of make Nupic closer to brain. >> > >> > Again, it would be interesting heard implementations that really have >> > something else than Nupic: I mean cognitive features like better >> > inference/learning, motor integration, or even performance improvements >> like >> > parallelism and others, and that are biologically designed (based on >> cells >> > archicteture, not on mathematical rules). >> > >> > Looking forward for heard you, >> > >> > Cheers, >> > -- >> > David Ragazzi >> > MSc in Sofware Engineer (University of Liverpool) >> > OS Community Commiter at Numenta.org >> > -- >> > "I think James Connolly, the Irish revolutionary, is right when he says >> that >> > the only prophets are those who make their future. So we're not >> > anticipating, we're working for it." >> >> > > > -- > David Ragazzi > MSc in Sofware Engineer (University of Liverpool) > OS Community Commiter at Numenta.org > -- > "I think James Connolly, the Irish revolutionary, is right when he says that > the only prophets are those who make their future. So we're not > anticipating, we're working for it." > -- Fergal Byrne, Brenter IT http://inbits.com - Better Living through Thoughtful Technology http://ie.linkedin.com/in/fergbyrne/ - https://github.com/fergalbyrne Founder of Clortex: HTM in Clojure - https://github.com/nupic-community/clortex Author, Real Machine Intelligence with Clortex and NuPIC Read for free or buy the book at https://leanpub.com/realsmartmachines Speaking on Clortex and HTM/CLA at euroClojure Krakow, June 2014: http://euroclojure.com/2014/ and at LambdaJam Chicago, July 2014: http://www.lambdajam.com e:[email protected] t:+353 83 4214179 Join the quest for Machine Intelligence at http://numenta.org Formerly of Adnet [email protected] http://www.adnet.ie
