Subscribe to DSC Newsletter

Machine Learning is all about Common Sense.

Machines are invented to solve problems and make life easier. It is widely accepted that common sense is a sense which is not so common :) So, don't you think this problem should also be addressed? 

Right. if machines are supposed to solve our problems then is there any machine that can solve our this problem? Or someone would say that can a machine ever solve this problem?  I'll say ! machine learning is all about to solve this issue. 

What is a common sense? According to Wikipedia "Common sense is a basic ability to perceiveunderstand, and judge things, which is shared by ("common to") nearly all people and can reasonably be expected of nearly all people without any need for debate"

Simple, easy to understand and widely accepted facts (common sense) are the building blocks of every complex structure. If common sense is persistent it becomes intelligence.

Humans are not persistent, most of the times we loose our common sense that results into a mistake. Sometime, due to some social norms we even don't care about common sense, literally. There are number of examples when people blindly followed some trend or buzz word and experienced a great lose while just using common sense they could have save themselves.  In human common mistakes are common due to not using common sense. Common sense is the collective heritage of human being that they have learnt by experience.

If someone does not like to play with fire, dose not go for swimming in cold weather, does not invest in business where lose is certain and takes like wise decisions then this is due to common sense. Common sense increases by experience. 

When we train a machine learning model we try to let it learn common facts in underlying data. Every feature contribute something, and acts like sensor (predictor/classifier) then model collects all these sense and tries to produce a common sense in general to predict or classify something on new data.

Whenever any machine learning model tries to become unnecessarily complex we say hay why are you loosing your common sense? do some generalization and this situation is called overfitting. Similarly whenever a model underfits we again suggest to follow common sense.

Common sense is infact average thinking. And if you are thinking that average is something which is not valuable then you should review your though. Actually here you are underfitting ;)

See these images

You might think that unique and striking facial features make a someone drop-dead gorgeous – but this image shows that on the most-part they do not! An attractive face is an ‘average’ face: I’d bet that you think the faces on the right are the most attractive – and these faces aren’t even real – the faces on the right are computer generated ‘averages’ of several faces. To read further on average face go to this link.

In nature every distribution tends to follow Normal Distribution. Hence, normal distribution is the common distribution and every distribution tends to follow common distribution.  So, being normal and following common sense persistently is something that is natural. 

"The bean machine, a device invented byFrancis Galton, can be called the first generator of normal random variables. This machine consists of a vertical board with interleaved rows of pins. Small balls are dropped from the top and thenbounce randomly left or right as they hit the pins. The balls are collected into bins at the bottom and settle down into a pattern resembling the Gaussian curve." - Wikipedia

Infact uncommon things are outliers and outliers are not common. No machine learning model can behave accurately until it get rids of these uncommon things. So, machine learning is all about training common sense, is't it?

I would like your comments on this and will try to see that what people are saying in common ;) 

Views: 3894

Comment

You need to be a member of Data Science Central to add comments!

Join Data Science Central

Comment by Dalila Benachenhou on January 28, 2016 at 10:20am

"No machine learning model can behave accurately until it get rids of these uncommon things. "  No totally true, tree based predictive models are known to be robust against outliers.  You can also reduce the effect of outliers by using L1 instead of L2-norm.  I think what you meant there is a level of amount of outliers that makes all machine learning models useless.

Comment by Richard Ordowich on January 25, 2016 at 9:17am

Data and algorithms are subjective and governed by human values, behaviors and norms. What is "common sense" to one person or group of people is different in another context.

Machine learning are algorithms designed by humans using selected (biased) data. Machine learning is pattern recognition. But what patterns are relevant and in which contexts? Correlations are not causation and correlations are subjective. 

The term "machine learning" is a misnomer. A machine cannot learn. Algorithms change based on the data but at its core, it is just discovering correlations that have been "programmed" in. The "learning" concept doesn't apply to machines, only to humans. Machines are unaffected by emotions, experience and are unwary of themselves or their backgrounds.  (understanding Computers and Cognition by Winograd and Flores) .

Comment by Troy Le on January 22, 2016 at 5:12pm
"...Infact uncommon things are outliers and outliers are not common. No machine learning model can behave accurately until it get rids of these uncommon things...". Unless of if the target is to predict these outliers, such as in fraud detection.

Videos

  • Add Videos
  • View All

© 2019   Data Science Central ®   Powered by

Badges  |  Report an Issue  |  Privacy Policy  |  Terms of Service