I was curious about what others are going to use their knowledge of Tiny ML for. I personally, was thinking of starting a business. Is there anyone who would be interested in this?
Great question! I’m currently researching TinyML and my goal is to apply the technology for application in conservation and environmental science. (like this: http://arribada.org/)
Will your business use TinyML to increase efficiency or will something TinyML related be your product?
That is really cool Cbanbury. Conservation and Environmental science is definitely using Tiny ML for a good purpose.
I made a kind of proof-of-concept device called ML gloves which used analog flex sensor data. This was before I knew anything about Tiny ML. It was really supposed to be a way of typing on any hard surface or in the air. I was able to type around 10 keyboard keys using the device. I actually found out about Tiny ML from a magazine that called it edge computing.
Anyway, I would like to make a company around the idea of not having to take a large keyboard around, and being able to type almost anywhere. A lot of devices on the market use gestures to type. But I wanted mine to be able to not break the current QWERTY typing pattern. This way those that can touch type will still be able to use the device, with little to no learning curve.
I like the idea Xen. I would like to part of it if you decide to move forward with it.
I also thought I’d mention that I really liked the course on anomoaly detection. To me it was an interesting way of using an NN to detect such a thing. What kind of network and sensors do you think you’ll work with the most Cbanbury? For example a CNN or dense? Will you use a microphone or camera, like the Arducam?
That is great to know pmoise. It would be nice to have some help. It would be difficult for me to do all the data gathering and validation myself. If you would like to be a partner in my company. Please, feel free to email me at firstname.lastname@example.org. Tell me a bit about yourself and what motivates you to want to work on such a project.
That’s an interesting application! Keep us posted on the forum on your progress.
I’m glad you liked the anomaly detection section (I actually created those assignments). I agree is a unique way to apply NNs.
I generally work with CNNs and with audio and visual inputs. I mostly focus on system-wide optimizations though which aren’t specific to a sensor.
I was thinking of using tinyML for making rotating equipment more reliable (anomaly detection) by understanding when and how they are suffering some potentially dangerous upsets - apart from usage of typical bearing temp, vibration readings etc. Was thinking of using microphone and vibration sensors.
I’m not sure whether such a project may better fit: in the company where I currently work, or as a separate business? I will give it thoughts.
Xen: very interesting idea - if you start a business I may also be interested. Have you thought of using sensors which are less of a burden than those within gloves? I see ultrasound sensors are becoming smaller and smaller - maybe one wristband per wrist- battery and tinyML powered?
That is cool that you are interested in anomaly detection Andrea. As far as whether a company or starting a business would be better, I’m not sure.
I was going to use a camera this time around. The limitations of the flex sensors is that they aren’t sensitive enough to catch home-row data on the keyboard. However, the strength would be that they could work in low or no light environments, as opposed to a camera. There are some flex senors that might be sensitive enough, but having to put on two typing gloves each time may become tedious for users. I have seen the big guys, including amazon and some others release ML powered typing devices. But they all seem to use gesturing to type. I think the data for pressing each individual key is unique enough to have an image based Tiny ML dataset for it.
The ultrasound sensor sounds like a really good idea, pun intended. I hadn’t thought about using that sensor. The problem with that would be that it could make collecting the data more difficult. After all, how many people have access to an ultrasound sensor. But it may work better than the two other methods I mentioned. Plus that way, maybe the demographic may not matter as much, only the size and shape of the hands
Wow, that is awesome that you made those assignments Cbanbury. The anomaly detection section was the most interesting to me. I thought that if I didn’t already have this idea in my mind of what I’d like to do with Tiny ML, I’d do anomaly detection. Nice work!
Before I started learning about ML, I thought CNN was just a news network.
I was curious if anyone has stumbled across the use of TinyML in underwater or submerged applications. There are interesting uses for TinyML in the conservation of oceans, lakes and rivers. However there are two primary challenges: 1) obtaining waterproof housings and 2) data collection in remote areas.
I live in Florida. A number of salt and brackish waterways have been impacted by the introduction of cyanobacteria from fresh water systems. Early detection of this algae can help improve environmental efforts.
Please do share any links you find @camcollins as that’s an area of interest to me as well.
Yes, Arribada looks like an interesting first mover in the use of TinyML in conversation technology. The WWF is also moving in this direction (Conservation Technology | WWF).
My company is very involved in COVID testing and vaccination efforts going on the USA, so unfortunately I will be sidetracked until the pandemic is under better control.