Thread by @Sauain

Human Bias in Machine Learning !!

So let’s take an example first to learn what is Human bias

🧵👇

Close your eyes after reading this line and think what a shoe looks like…

Okay stop and let see shoes

Did you picture this?

or this?

or this?

We do not know why but each of us is biased toward one shoe than another.

Imagine you are teaching a computer to recognize a shoe and unintentionally you are making it bias towards one of the shoe then another.

That’s how bias happens in machine learning.

Machine learning helping us in our daily life :-

- Translations

- Suggestions

- Even understand what we say to it.

In traditional programming programmer codes to solve these problems but with machine learning we teach computers to solve these problems and to learn by data.

It is not that if something is based on data it is neutral, there’s a lot of chance of our human biases to extract that data too.

It is "impossible" to remove human biases completely.

So our human biases became part of the technology we made in many ways.

𝗧𝗵𝗿𝗲𝗲 𝗧𝘆𝗽𝗲𝘀 𝗼𝗳 𝗠𝗟 𝗛𝘂𝗺𝗮𝗻 𝗕𝗶𝗮𝘀𝗲𝘀

- Interaction Bias
- Latent Bias
- Selection Bias

𝗜𝗻𝘁𝗲𝗿𝗮𝗰𝘁𝗶𝗼𝗻 𝗯𝗶𝗮𝘀

As it is done in the recent game, many people are asked to draw a shoe for the computer and most people draw the first image and make computer biased to recognize the third shoe image.

𝗟𝗮𝘁𝗲𝗻𝘁 𝗕𝗶𝗮𝘀

For example, if we are training computer how a physicist looks like from the image of a past physicist, as most of the past physicists are men it makes computer biased to recognize female physicists.

𝗦𝗲𝗹𝗲𝗰𝘁𝗶𝗼𝗻 𝗕𝗶𝗮𝘀

It is based on the selection of data you take to train the model.

For example, you are training a model to detect a human face you have to take the photos of people around the globe otherwise the model will be biased towards one.

If you learn something from this thread then give a RT to first tweet of this thread 🧵