Ask HN: Simple projects to implement machine learning

I used lua to create a simple AI. This AI can learn words and store them in a dictionary. It can link one word with another, having some resemblance or relationship. Also I implemented a "emotional link" between them; one word can be either in a GOOD or BAD relationship with another. Like the word "rat" will have a BAD relationship with "cat". It is based on meaning of words, rather than acting like a mere processor of vocabulary crawler! This AI can communicate with me.. asks me questions that it generates from meaningful concepts. Ofcourse, it is very basic right now and makes errors. But it will grow. The bottomline is that you must implement "from the roots". Perception and recognition are built over these roots. Hope you got on idea what I am attempting to build. One thing most required right now is adding a solid grammer engine to it. If I were not a lazy programmer, I would have done it already. :D

I built an image interpretation app once that was really interesting from a machine learning perspective. Using image magick I extracted shapes and based on the layout I was able to feed that into a "composition" algorithm that judged whether the image layout was good or not. I also implemented a "plant detection" feature to find out whether a plant was in the image or not. I think images is an area that could still benefit a lot from machine learning, i.e. facial recognition, image search (extracting meaning from images). thats just some of my ideas. machine learning is awesome I wish you all the best!!

How about creating a recommendation engine for HN articles? It should recommend articles upvoted by users that have similar likes/dislikes as me.

I would also like to browse HN articles/comments on basis of topics. For example, right now if I want to find HN posts related to "machine learning" I use HN Search. Can you make it better by using machine learning techniques so that I can get all relevant HN articles on a given topic sortable on basis of upvotes and time?

How about analyzing a database of approx 400k car parts requests? Analyze key words, year-make-model, geographic location, etc?

I'm currently doing this to generate a chloropleth map for PartsLine.com. https://skitch.com/tzmartin/eydmm/partsline.com-rfq-to-fips-...

Why not sign-up to kaggle and do some of their challenges, that way you can also benchmark yourself against others.