EuroClojure meeting?

[Daniel Slutsky]

Dear friends,
this mail is sent to the incanter and numerical-clojure groups.

Maybe several of you are about to visit EuroClojure this week.

I thought it could be nice if some of us try to meet before/after the
conference.

We may discuss our experiences in using clojure for statistical computing / scientific computing / data analysis.
Probably most of us have various ideas for future development. It
would be nice to make plans (or at least fantasize) together.

What do you think?
(Would any of you like to meet?
What topics would you like to discuss?
What time of meeting would you prefer? (25.6? 28.6? other?))

~~~~

(to present my personal interest:
In the last year I have had some (difficult) experience in
trying to insert Clojure as a tool into an office working on some data
mining tasks. There was some partial success. Clojure has made several of us think in new ways.
Still, most team members prefer other tools in tasks where core.matrix could be a delight.
Unfortunately, most of the work could not be open
sourced. Now I am entering a new job, where I hope to finally be able to contribute to the community.)

[Mike Anderson]

Unfortunately I can't make EuroClojure - but will be in London over the next three weeks so if anyone is in the area would be good to meet up!

--
You received this message because you are subscribed to the Google Groups "Numerical Clojure" group.
To unsubscribe from this group and stop receiving emails from it, send an email to numerical-cloj...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Bruce Durling]

Mike,

There is a Clojure Dojo on Tuesday 24th in London

https://www.eventbrite.co.uk/e/june-2014-london-clojure-dojo-at-thoughtworks-tickets-12000494803

and talks on 1st July at skills Matter

https://skillsmatter.com/meetups/6430-reactive-web-apps-and-minecraft-movies

cheers,
Bruce

--
@otfrom | CTO & co-founder @MastodonC | mastodonc.com
See recent coverage of us in the Economist http://econ.st/WeTd2i and
the Financial Times http://on.ft.com/T154BA

[Mike Anderson]

Thanks Bruce! I'll try to make both if I am able.

What's the topic for the Dojo - is it covering numerics / Incanter this time around?

[Bruce Durling]

Mike,

Usually voted on by the people attending unless someone has prepared
something. It usually runs as per this blog post here:

http://otfrom.wordpress.com/2012/07/04/how-to-run-a-london-clojure-dojo-in-20ish-easy-steps/

Though it varies from MC to MC and what the group wants. They aren't
rules so much as guidelines.

cheers,
Bruce

[Fergal Byrne]

Hi Daniel,

I'm speaking at EuroClojure, about a new implementation of Jeff Hawkins' Hierarchical Temporal Memory (HTM) in Clojure, called Clortex. I'd love to meet up at any stage over the few days. I'll be in Krakow from tomorrow afternoon until Saturday, so we should get a chance.

See you there,

Fergal Byrne

Fergal Byrne, Brenter IT

Author, Real Machine Intelligence with Clortex and NuPIC 

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

http://inbits.com - Better Living through Thoughtful Technology

http://ie.linkedin.com/in/fergbyrne/ - https://github.com/fergalbyrne

e:fergalby...@gmail.com t:+353 83 4214179

Join the quest for Machine Intelligence at http://numenta.org

Formerly of Adnet edi...@adnet.ie http://www.adnet.ie

[Daniel Slutsky]

Hi Fegral, your talk sounds exciting.

I can meet at any time from tomorrow (June 25) 16:00 to June 30.

If you wish to meet a person who has not heard your lecture yet but is curious about it, let us meet tomorrow.
If you wish to meet a person who has already heard your lecture, let us meet afterwards.

Thanks, Daniel Slutsky

[Mike Anderson]

Hi Fergal,

Clortex looks very interesting! I notice that you have an implementation of encoders / decoders that is a common requirement of many machine learning solutions (including my own neural network kit - I presented some simple coders and decoders in my talk at the 2012 Conj).

As a result I was thinking of building a multi-function encoder/decoder library using protocols (on top of core.matrix) so that people can avoid re-inventing this particular wheel 

At its core there are just two abstractions:

encoder: Arbitrary value -> Vector

decoder: Vector -> Arbitrary value

But then there are higher order abstractions / combinators as well, e.g.

{Map of encoders} -> encoder for maps

[Vector of coders] -> encoder for fixed-length vectors/sequences

"reduce-style" encoder -> encoder for variable-length vectors / sequences

As well as a library of "primitive" codes for Strings, Numbers, Characters etc.

Possibly this could be linked to something like Prismatic's schema for validation / structure definition etc.

Any interest in collaborating on something like this?

  Mike.

[Fergal Byrne]

Hi Mike,

Your talk was one of those which convinced me that Clojure was the language to use first (I started with Elixir, I'm hoping Jonny Winn and I can still get that project rolling), so I'm familiar with your project. (In fact I was looking at it this morning again to steal ideas! This playlist [1] is a kind of record of talks I've watched for insight and inspiration during the project, several of them are yours).

The guys over in Numenta have been working on encoders like this for almost 10 years now; I believe encoders are hugely important for the more brain-faithful end of the neural-net spectrum. Because of the importance of time in HTM, we also have specific temporal encoders which have evolved over the years.

HTM encoders are pure binary (every input value gets mapped to a vector of bits, each with "semantic" meaning), so that they fit in with the paradigm of Sparse Distributed Representations (SDRs), a key feature of HTM. So, for example, a scalar value like temperature or server load would typically be encoded using 128-512 bits per field, whereas you'd more likely use the scalar value directly as input to a traditional NN (subject to scaling and mean perhaps).

I noticed in your talk that your demo encoded numbers into binary prior to input; if this is your chosen paradigm then HTM-style encoders would be a perfect match, and of course we should pool resources on that.

I'll get in touch with you next week, let's see what we can do.

Regards

Fergal

[1] https://www.youtube.com/playlist?list=PLrgvR0h3ott4SCTIeXQ0zvtiMQA1jxj1V