Spotting and solving everyday problems with machine learning | Session 

Ok Google. 😆

hahahha funny

Well, yah. Big yah.

What is "cool"? That software can detect bicycle in video and mark that second minute contains it? Or possible to find voice "hi dady"? Oh my, people who u are to be fascinated by such things... What's really shocking is that u believe this is "artificial intelligence"...

It's a deadly road be tough😂 n don't give up

exactly! same for me

😍

Totaly agree! Great communication skills.

How much was approximate cost according to you?

But the meaning was lost in translation.

PHEW

8,000

Probably. That is usually a best use case

The example in the video used a ML model from google. So it's basically just loading in the video, let the model do it's calculations and take the output. What the model does under the hood is very comlicated math - and also a lot nobody actually knows. In this example, it used some voice recognition to transcribe speech, propably image-recognition to identify certain objects and areas and then the word embedding to feed the search-engine. The word embedding is seperate because the words you write into the search are send through the same embedding algorithm - but obviously you wouldn't send those words through the image-recognition. Image recognition might be done on every image or once per second or maybe it identified a scale of "change" so that if the camera was filming the same spot for a minute, the model does only go over it once instead of 60 or so times. The reason this works is because Google used billions of images which were already tagged, have ML and AI scientists create a theoretical construct that should be able to learn identifying those -> and then have that construct do a massive ton of calculations and self-adjustments (the actual learning process) until it found the same tags in the images as provided. This training process might easily have taken days or even weeks. At the end, the model has "learned" to identify these things and can be used on data it has never seen before and hopefully find fitting tags as well. Just to reiterate, nobody knows what calculations the model is actually doing or why they work. Because during training it has finetuned millions, if not billions of parameters for calculations.

I m working on AI & ML. Plurals engine. Ur presentation is greatly inspiring. Pls. Communicate me how long really it wl take in India for its wide use.

All that can be done locally, but she's advocating the benefits of using cloud and google ML APIs in particular. I personally love to apply ML to my daily life tasks to make it better and easier. For example the file searching example, I've made that so that it would scan the locally stored files, since that would be faster and given the bad rep that Google has regarding user privacy, I can't chance uploading my personal files and media to google drive, also it'd be a time and space consuming task.