Projects
Kawasu
The Kawasu website is currently the main
project that is being worked on.
Kawasu: The Miniature Exchange
'Kawasu' is an anglophone translation of the Japanese word for Exchange.
The Japanese were the first people to use a Futures Exchange,
the Dojima Rice Exchange, in 1710.
The Kawasu project is an attempt to apply the way an exchange works, to any situation where multiple identical
items are bought and sold. The idea is that anyone can set up an exchange and trade on it via the Kawasu website.
Think: eBay for Commodified Items
The point is that an exchange allows real time price negotiation.
Buyers bid a price that they are willing to pay.
Sellers offer unit items at a price they are willing to sell for.
Both sides can then increase or decrease their bids until a match is made, and a trade is effected.
Why would anyone do this?
There are benefits for both sides.
Buyers will get the lowest price that they are willing to pay, and if they can afford to wait,
they can wait until the price decreases.
Sellers can instantly assess demand, and can guarantee stock turnover rates by satisying that
demand at a pace that suits them.
What's wrong with how things are now?
Currently, if you mass produce Widgets, you need to do some market research, and try to
guess demand for your Widgets. Then, you run some pricing models, and come up with a
unit cost for Widgets, then maybe a discount for a five-pack, and all that marketing stuff
that people love.
If you get the price wrong, you're in trouble. Too low, and you have missed a profit opportunity,
you will sell out of Widgets. Too high, and you will not sell enough Widgets and stock piles up in
warehouses.
An exchange would allow you to offer different amounts of widgets at different prices at the same time.
You could offer 10000 Widgets at 5 dollars each, and 2000 at 4.80, and 1000 at 4.50 .
You might find that your offer of 1000 at 4.50 goes like hot cakes, and people start taking Widgets at 4.80 .
You can adjust your offers in real time, and bump the price up, until the selling rate hits a level that
matches your production rate.
Alternatively, you might find that your Widgets go untouched at these prices. There are buyers around, but
they have offered 4.20 for 1000 Widgets. You can take that offer if you like, or you could edge your best
price downwards in an attempt to cajole the buyers upwards.
The end result is that trades occur when both sides agree on the price. It is a mechanism that has been
tried and proven in financial markets for centuries.
What about Clearing?
Exchanges need Clearing. Clearing is taking the promise to trade at the price agreed, and enacting the
transfer of goods for money. The time-lag between the agreement to trade, and the transfer of money, causes
the need for Clearing.
At the Kawasu site, you credit your account before you buy.
When you agree to buy an item, the money is deducted from your Kawasu account, and paid into the
seller's Kawasu account. We get the money side out of the way first.
Web Neural Net: A Trainable Neural Net via The Internet
A new project to deliver Neural Net functionality as a service.
You bring the neurons, we'll do the thinking...
A neuron is a simple thing.
It has two states, and a stimulation threshold to change between those states.
When the threshold is reached, the neuron fires, sends a weighted stimulating signal to any connected neurons,
and then re-charges.
As a web service, the client would define the states, and the threshold to trigger the state change.
The client would also define the expected output of the net.
A simple example would be an optical character reader.
It would take as input 64 neurons in an 8 by 8 grid.
One state would be for white pixels, and one for black.
Each neuron would trigger if the pixed it was attached to flipped from black to white.
The output of the net would be the character that the neural net thinks is closest to the ones
that it has been trained to recognise.
The point here is that the web service would offer a set of input neurons and a set of
output neurons. The meaning and application of the neurons would be defined by the client.
The neural net is abstracted and generalised.
The client would define the application. The web service would provide generalised
neural net functionality.
Early Days
This project is at the very earliest stages, but any feedback would be welcome.