0
$\begingroup$

this might be a silly question but I guess the answer comes with experience in this field. I'm just wondering if today, with the internet overflowing with data and specifically with images (maybe not tagged), are there a lot of examples for learning tasks (specifically ones that takes images as inputs) without a proper database for those tasks?

I'm asking because my experience in this field is limited and I only tried to work with some more standard learning tasks that have a proper size database and I'm wondering if the need for data (or tagged data) is still an issue. So, if you can give some examples for tasks that still require a proper database of images, that will be great. Thanks.

$\endgroup$
3
$\begingroup$

A naive observer might be tempted to think that with the vast amounts of data being collected, there's enough to answer all learning tasks. But this is misleading at least for the following reasons:

  1. There's no such thing as "all learning". Humanity has not yet perceived what could be the limits of knowledge. With the boldly stated ambitions of AI to match human intelligence, this also transfers to the lack of limits to what could machine learning aim to learn.

  2. One feature that a lot of data scientists downplay, but that comes as heritage from traditional science fields, is that sound research starts with identifying what you want to achieve and focus data collection on that objective. In other words, in an ideal data analysis setup, the analyst would have a say in how fit-for-purpose data is collected.

  3. A big part of machine learning is preprocessing data in order to clean, normalise or annotate it. This could also be done to pre-existing data as a way to adapt it for certain - previously unenvisaged - purposes. If the data is not collected with attention to how it is going to be used, it would need more preprocessing, and thus - generally speaking - would be less reliable due to the accumulation of steps where noise could be introduced.

These three arguments are only a scratch on the surface of the topic, but they come to show that even if it is sometimes useful, the widespread data collection does not always respond to all questions one might want to ask. Even if in machine learning it sounds a bit more complicated, it all boils down to this amazing quote from Alice in Wonderland:

“Alice: Would you tell me, please, which way I ought to go from here?

The Cheshire Cat: That depends a good deal on where you want to get to.

Alice: I don't much care where.

The Cheshire Cat: Then it doesn't much matter which way you go.

Alice: ...So long as I get somewhere.

The Cheshire Cat: Oh, you're sure to do that, if only you walk long enough.”

Examples:

  • Comprehending information from random receipts or invoices.
  • Automatically judging football situations.
  • Telling Google (think captcha) if that thing on the image is a street sign, a bridge or a number.
|improve this answer|||||
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.