Incremental Learning Demo, page-192

Thank you uiux for making this program, it has been a fun day experimenting with it! I tried it at the grocers and on some plants and animals outdoors. Some things I've experienced:

1. Easy enough to distinguish between very different objects with 1-shot learning, but with similar objects ie. orange and mandarin, grass and she-oaks (both with green and brown leaves) it became a challenge. This is where more learning would be needed to improve accuracy I suppose.

2. Fast or constantly moving objects were a challenge, ie. In trying to identify some birds.

3. When I set it to learn an image of multiple objects ie. A tray of oranges, it could also identify the single object ie. One orange correctly. So it's not just single object learning with clear backgrounds

Real life application - might be for example in quadrat sampling for biodiversity - being able to identify X species within a given space, and maybe with some extra programming to quantify the numbers as well.

ID of species - Sometimes for work I use a plant recognition app, where I can take a picture of a plant that I do not know, and the app will suggest the closest matches from its database. Its servers have been down recently, thank goodness for backup servers, but what if those failed too? What if I had no Internet connection in a remote location to use the app? I can see the Akida advantages in these situations, if this had on-device learning. Something equivalent, but based on another sense could be similar to FrogID (sound recognition).

Just thought I'd share my layman's user experience with it