Привет, Антон. Можно тебя в нагулую попросить помощи по статье Dense Associative Memories and Deep Learning Дмитрия Котова? Что-то читаю и не врубаюсь про что она. Или может что посоветуешь куда выкурить.
Это - arxiv.org/pdf/1606.01164.pdf ? Боюсь не читал, не смотрел. Это надо сидеть пол дня в контекст врубаться. Она к тому же 10 летней давности, а значит определяется тогдашним состоянием науки. С тех пор много чего другими подходами решается.
I think powerful boards to process images is dea. The future will be low specs boards to send pics to the LLMS cloud platforms like qrok to analyze accurately the data like items sorting
It's not possible:) 1) Communication networks limitation. Not possible to stream every camera in cloud. Not possible to stream from every location. 2) Price for clowd recognition. Too expensve gor single image recognition. 3) Latency. Local embedding extraction vs fully cloud processing - ua-cam.com/video/TitZV6k8zfA/v-deo.htmlsi=15PSg0I1ThkIMq5q&t=1038 4) Privacy. Noone want to stream in cloud privat information.
Thanks for the answer but I see pretty viable a rpi5 with camera as items sorter connected to a stable wifi to grow, no issues about the privacy for this case study
@@annwang5530 1) What is the price of processing one hour of video with any big LLM?:) How many cameras can one server support? 2) Good Wi-Fi is super rare for real applications, such as security and transportation. How many cameras can you support in one network? What will you do if there is no wired connection to the WiFi hub? 3) Any images of persons or vehicles that are going on the server is a privacy issue.
Hi thanks again. Please let me explain my application takes a single compressed image from a single camera, not video stream, then it might have like 200kbs of a jpeg per 15 seconds at most, cheap, to the LLMS cloud, pretty feasible if the WiFi is ok.
@@annwang5530 Defenetly. It can work in some applications. But your initial statement was: "The future will be low specs boards to send pics to the LLMS cloud platforms like qrok to analyze accurately the data like items sorting" But for 95% of tasks, it's not the correct solution. And it will be incorrect in the future. I expect more good embedded devices with pre-trained networks in them.
![Как вкатиться в embedded и что там ловить. [RuCTF 2019]](/img/n.gif)
very great review, waiting for that board gift video 👍
Привет, Антон. Можно тебя в нагулую попросить помощи по статье Dense Associative Memories and Deep Learning Дмитрия Котова? Что-то читаю и не врубаюсь про что она. Или может что посоветуешь куда выкурить.
Это - arxiv.org/pdf/1606.01164.pdf ?
Боюсь не читал, не смотрел. Это надо сидеть пол дня в контекст врубаться. Она к тому же 10 летней давности, а значит определяется тогдашним состоянием науки. С тех пор много чего другими подходами решается.
@@AntonMaltsev да, она, это по учебе дали на разбор. и я думал она на слуху как-то, ага, значит уже ушла в прошлое.. понял.
I think powerful boards to process images is dea. The future will be low specs boards to send pics to the LLMS cloud platforms like qrok to analyze accurately the data like items sorting
It's not possible:)
1) Communication networks limitation. Not possible to stream every camera in cloud. Not possible to stream from every location.
2) Price for clowd recognition. Too expensve gor single image recognition.
3) Latency. Local embedding extraction vs fully cloud processing - ua-cam.com/video/TitZV6k8zfA/v-deo.htmlsi=15PSg0I1ThkIMq5q&t=1038
4) Privacy. Noone want to stream in cloud privat information.
Thanks for the answer but I see pretty viable a rpi5 with camera as items sorter connected to a stable wifi to grow, no issues about the privacy for this case study
@@annwang5530
1) What is the price of processing one hour of video with any big LLM?:) How many cameras can one server support?
2) Good Wi-Fi is super rare for real applications, such as security and transportation. How many cameras can you support in one network? What will you do if there is no wired connection to the WiFi hub?
3) Any images of persons or vehicles that are going on the server is a privacy issue.
Hi thanks again. Please let me explain my application takes a single compressed image from a single camera, not video stream, then it might have like 200kbs of a jpeg per 15 seconds at most, cheap, to the LLMS cloud, pretty feasible if the WiFi is ok.
@@annwang5530 Defenetly. It can work in some applications. But your initial statement was:
"The future will be low specs boards to send pics to the LLMS cloud platforms like qrok to analyze accurately the data like items sorting"
But for 95% of tasks, it's not the correct solution. And it will be incorrect in the future. I expect more good embedded devices with pre-trained networks in them.
Антон, поработайте над своим произношением пожалуйста. Ваша речь звучит как в несмешных шутках про русских.
там всем до лампочки до акцента если грамматика правильная
Всё нормально.