Ask HN: Is machine learning useful outside the few very common applications? | Hacker News Reader

Ask HN: Is machine learning useful outside the few very common applications?

2 pointscplat1 comment

In theory, we can use machine learning to do a lot of cool things. However, I've seen a dearth of material on how real world problems are formulated in terms of machine learning problems.

Computer Vision is rapidly advancing as far as deep learning is concerned, and we can generally see a lot of applications there (medical, autonomous, etc.)

NLP has a different story. For example, I was researching how a support desk uses machine learning. They use that to classify tickets automatically to the right department. Useful? Somewhat. Revolutionary? I don't think so. [this was from MonkeyLearn's website, btw]

What are the kind of people I can connect with who are doing a lot of applicative work when it comes to machine learning and AI? I mean, focusing on real-world problems and then formulating them as an ML problem? Or are there resources that I could read?

Thanks!

I can't provide any resources to help you learn or about who to contact, just an interesting story I heard.

I think it was on Reddit, but a guy had hooked up all of the lights in his house to a ML box, that he trained based on his family's actions for a few weeks and then let it take over. The idea being that he has motion sensor data and the data for when lights are turning on and off feeding into this machine and making decisions on when to turn lights on and off, with a few rules; don't leave lights on indefinitely, etc.

He noted how at times the "AI" would turn on lights for a room he wasn't in on Saturday, but M-F he was in that room at that time. It also would make incorrect assumptions about which lights to turn on based on a variety of other factors. In those cases, he would manually turn the light off and the machine would note that it was incorrect.

Towards the end he said that the false positives were getting infrequent, maybe 1 or 2 day, whereas in the beginning they were occurred more often. I think there might have even been a calendar integration so the house knew when guests were coming over and that the light patterns were different.

I think there is a lot of room for growth in ML for this type of residential application. For example, a garden that could water itself, but also check the weather forecast and delay watering until after a storm has cleared the area, all while checking soil sensors and periodic photos of the garden to see if the plants are the correct size/color.

I'm really excited to see what uses will be possible once computing power makes ML trivial.

Reply on news.ycombinator.com