I’m going to try to do this every month, you can find older editions of this elsewhere in the blog, the numbering continues from the previous installment.
A system that manages a database of netblocks belonging to ISPs and the major search engines and blocks accesses from everywhere else. Included in the blocklists would be zombie nodes and other machines known to be used for things like comment spam and so on.
The difference with a regular ‘blacklist’ is that this would pre-emptively take out all ranges belonging to hosting providers.
A three dimensional display built up from strings of LEDs attached to a central post.
The central post is connected to a motor and an encoder to sense it’s position (or a stepper to turn it), when spinning ‘at speed’ the led strings will extend horizontally, at rest they’ll hang down. Because the angle of the post is known you can compute where each LED is in space so you can light it up at the right colour/intensity to generate a three dimensional image. Because of the speed of rotation the wires themselves will not be visible, they’ll just dim the overal picture a bit. When at rest the strings of LEDs hang down and so then display will then occupy very little space.
The display will be viewable from any angle and will not require trickery to offset for viewer height.
(inspired by: http://www.seekway.com.cn/e/3d/h32/video.htm and http://www.youtube.com/watch?v=iDTMsF9CAE0 )
A really dumb image classification system based on the observation that many images are quite alike and that classification can be attacked in the same way that you can find the ‘distance’ between two strings.
Transforming one image in to another image using a series of operations (for instance, copying a pixel, increasing or decreasing the value of a pixel and so on) will yield an ‘edit distance’, and images that are more alike will have a shorter edit distance.
To figure out what’s on an image find the image most closely related to it and present the the tags used to describe it, allow the user to correct the tags and store the image + corrected tags as a new datapoint.
Obviously this would be extremely CPU intensive, how effective this would be depends on how many sample images you’d need to store with proper tags to make it work well enough. Sample images could be stored at a reduced resolution (say 64x48) if that does not affect the accuracy, and checking an image to test against all the images already stored could be done in parallel on GPUs for increased efficiency, the problem is ‘embarrassingly parallel’.
It’s not a very elegant solution to the problem, especially not because it still requires human intervention to allow for corrections.
Shoppert is a very simple cart shaped bot that folds up it’s carriage so you can stow it in the car. When you take it out of the car it unfolds the legs and powers up. You then simply walk away towards the store you wish to visit, and shoppert will follow you. It calculates its distance to you by monitoring your cell phone bluetooth signal strength. You shop as usual, you can dictate shoppert to ‘stay put’ or ‘follow’ using the cell phone as a remote control, the default is to simply stay right behind you ready to receive the goods you pluck from the shelves.
When you’re done shopping you simply push shoppert back in to the back of your car to take it and the contents home.
Inspired by our weekly trip to the local mall here where we have to get a shopping cart, shop, transfer all the goods to the cashier, pack them in to the cart again, pack the goods from the cart in to the car, go home, then finally transfer the goods from the car in to the house. With shoppert you’d save a bunch of time on handling the goods.