How do machines learn?

If you were trying to teach a machine how to learn on its own, where would you start?

Our ability to learn is a skill we often take for granted. When we try to pin down exactly what we mean by learning we see that it is not so simple.

Machine learning began with the grand aspiration of creating strong artificial intelligence (AI). Computer scientists realised early on that understanding and creating systems that can learn is core to what we think of as intelligence. Although it was soon realised how ambitious the goal of strong AI was, the field of machine learning found many other purposes and applications, with only a dedicated core continuing with the fundamental goal of artificial intelligence. We are now seeing many more machines that can learn, and the underpinning algorithms are being applied to big data analytics. We will meet some of these algorithms during this course.

Both machine learning and statistical inference are fundamentally involved with extracting information from a dataset and for this reason we see a huge overlap between the fields. Many machine learning (ML) methods developed by computer scientists have been adopted by statistics, and likewise the theory and methods developed by statisticians have assisted the development of ML methods.

Real applications in the real world

Applications of statistical and machine learning techniques are now common place.

They are being used in our phones and social media constantly, looking for ways to improve customer experience and advertising effectiveness. These include, among many other examples: neural networks for speech recognition, Naïve Bayes for spam filtering, decision trees for recommendation systems, bandit algorithms for targeting advertisements, convolutional neural nets for face recognition, regression and variable selection for classification of crops using remotely sensed data, Bayesian networks for analysing genomic data and many others.

Installing the software you will need
Download and install R: https://www.r-project.org/

References
https://www.futurelearn.com/courses/big-data-machine-learning/

The fields of statistical inference and machine learning can be associated with many seemingly disparate goals and applications ranging from personal assistants and music recommendations, to robotics and apocalyptic films.

However, many of these problems have a common objective. If we can identify the commonalities, we can identify the most appropriate methods to use. In this way, the maze of goals, methods and algorithms can be easily negotiated.

In this thread, we will focus on these commonalities and illustrate them with applications. In order to start this process, we need to understand a bit more about what machine learning and statistical inference are. As a first step in this understanding, it is useful to consider ten basic questions.

1. What is intelligence?
2. What is learning?
3. How do humans learn?
4. Is a chess computer intelligent?
5. What motivates a computer?
6. What do we mean by data analysis?
7. What is inference?
8. How can we judge whether a statistical or machine learning method or algorithm is good?
9. What are some of the properties of big data that are challenging for analysis?
10. What do machine learning and statistical inference have in common? And how do they differ?

There are many software packages that can be used for data analysis and machine learning, examples are: R, Java, Scala, Python, J and other tools, but I will be focusing on using R.

By the way, this thread will focus more on teaching you how to solve problems in machine learning using R. I don't enjoy talking, I prefer action - which is writing codes and getting desirable results.

Enough said, Let's go!


References
https://www.futurelearn.com/courses/big-data-machine-learning/

Wow... I will stay off your thread I guess.

sirRiddy:
Thank you for your input. No need to apologize at all for expressing your thoughts and critique
I was actually talking generally and not in terms of how Google works or how a specific
Machine Learning technique works, that is why I called it simplistic.
By the way, I do not think there is anything wrong with the 'bookish' definition or understanding of a concept.
Both research and academia work together to explore the frontier of knowledge.
I am glad you offered your take on this. Feel free to offer more insights and/or criticisms.

I'm sorry if I sounded rude or offensive. I hope my apology is accepted?

Now to the topic, knowing ML algorithms is not the problem but knowing when to apply them is.

I've worked with people who use Dimension Reduction techniques instead of Regularization, Regression instead of Decision trees and the list is endless.

There are online communities such as R bloggers, Stack overflow, and so on one can learn the latest trend from and don't forget to be involved on online competitions such as Kaggle and other hackathons.

There is a thread similar to this here on nairaland but I hope this could be different by us sharing our codes and assisting one another instead of engaging in too much talk.

sirRiddy:
Machine Learning(ML), a way to do AI, is gaining traction now because of things like when you make a typo,
for instance, while searching in Google,
it gives you the message: "Did you mean..."? This is the result of one of Google's machine learning algorithms;
a system that detects what searches you make a couple seconds after making a certain search.
For example, suppose you were searching for Who is Okocha' on Google but accidentally typed 'Who is pkpcha'. After the search,
you'd probably realise you typed it wrong and you'd go back and search for 'Okocha' a couple of seconds later.
Google’s algorithm recognises that you searched for something a couple of seconds after searching something
else, and it keeps this in mind for future users who make a similar typing mistake. As a result, Google
'learns' to correct it for you. [3]
So, ML is an attempt to make machines learn from data themselves. A major domain where it is being explored is the
understanding of the human genome.
This is how it works, in a very simplistic way.
1. You have a dataset, usually large.
2. Your system 'learns' from those dataset
3. You have ML algorithms explore patterns in those datasets
4. This knowledge can be applied to an array of tasks like speech recognition, image recognition etc

Alexa and the rest were programmed and would fail in other domains. With ML algorithms, systems can be
applied to different array of tasks, made to learn from data and apply this knowledge to new domains
while trying to solve tasks intelligently.

References
1. https://en.wikipedia.org/wiki/Artificial_intelligence
2. https://en.wikipedia.org/wiki/DeepMind
3. http://www.wired.co.uk/article/machine-learning-ai-explained

I get your explanation bro but I think you only have the "bookish" understanding [ I'm sorry for saying this].

The steps you outlined are important but you need to understand many processes that goes on for them to be accomplished. First, ask yourself "how did Google get these datasets?"
Second, "which systems learn from these datasets?", what is a system? Are they series of codes, programs, or written algorithms?
Third, what are the ML algorithms that can explore this pattern? How do they explore this pattern?

Lastly, you are right about step 4 outlined above

Please how do I get a prepaid courier envelope to myself? it's one of the requirements for the long term student visa. Please I need suggestions, I will be leaving Lagos soon.

Good day everyone,

My application was rejected at the embassy today and I as advised to apply for long term student visa instead of Schegen visa even though I will be staying in Germany for one semester. I was given an appointment on the 12th of July because I'm on scholarship else it would have taken longer.

Thanks to everyone who cared enough to reply my earlier question.

eMidas:


Its still the same procedure. Get to Germany first then apply to Spain for the remaining semester after your first semester in Germany.
Did you win an Erasmus Mundus scholarship?

Thanks for replying, yes I did...

Good day house, I asked a question the last time and no one responded I'm not offended anyways.

My appointment with the embassy is slated tomorrow by 7:30am, I am applying for a Schegen visa since I would be spending my first semester in Germany and the remaining three semesters in Spain (Erasmus Mundus Exchange program). I would be grateful if anyone can provide me with useful tips.

Thank you.

Pedagogue:

I have said it before and I'll say it again. OAU Ile-Ife is not for the intellectually faint-hearted, neither is it for the socially over-committed. Rather, the Uni is a perfect fit for students that are academically thirsty, and ready to personally improve their understanding of life.

And Ivy colleges are for who?

Must we have it difficult before we can be successful?

If you like apply to 20 schools but are your referees willing to write 20 topnotch reference letters?

I want to use this medium to appreciate our own very first lady on this forum. She has been wonderful, helpful and ....[please borrow me adjectives that ends with 'ful']. I can't thank her enough for her immense support.

Lagosismyhome, don't worry from today henceforth Lagos only won't be your home, goodness, mercy, favor, God's protection, preservation, peace, progress, prosperity, purity, promotion will also be your home.

I say a very BIG THANK YOU, you've been great and you remind me of Proverbs 31.

From a grateful heart.

adey2014:
the software if i am really correct started with my set in the faculty of education. before 2007/2008. results were calculated manually by the course adviser. but it changed for us. if u ever had of prof temi busari.who died in education. she was the dean who introduced that software for results calculation.

Bro, I empathize with you and I know how it feels. In my final year in the University, a course i had a B was turned to a C. How do i know i had a B you may ask? The course was a 300 level course and the result was on my portal and i printed it out. With this printed copy i had an evidence which i use to complain to the department severally. To cut the long story short, guess what i was told? I was told i already had a first class degree why should i still bother over a grade that was changed and it makes little difference in my CGPA. Moreover, from my transcript i never had a C, so at least it shows the department is very strict with marks. Do you know the difference of 0.01 costs me from being the Best Graduating student in the school? The Best graduating student's CGPA was higher than mine with 0.01.

This is just to give an insight on what is happening in our institutions. All said, i was asked to do my MSc in that same school, me? Only if they know the resentment i hold against them.

Bro, never give up, you don't have to prove anything to anyone and with God by your side your best is yet to come.

Grades alone cannot define your future but Grace can. Do your best, this American Dream must be achieved.

Sorry for the long epistle, I've never said all these to anyone.

Good day everyone.

Please i would be immensely grateful if anyone can provide me with first hand information.

I have been awarded QEM Erasmus Mundus Scholarship. My first semester would be in Universitat Bielefeld, I would be grateful if someone can assist me with valuable first hand information on Visa application.

Caveat: I have read enough information from the German embassy website and I'm not a lazy reader, I just need first hand information.

Thanks for taking your time and effort to provide a befitting response, I would be anticipating your prompt responses.

Thank you.

Please take time to pray for your family, there is nothing better than when Jesus is in your family. The joy is indescribable, the love is unquantifiable, despite the storms of life you will feel this satisfaction.

Why won't we talk about family when family is all we've got? ---- See you again by Wiz Khalifa

joeluv:
Thank you. I would do just that and I would keep in touch. I am a she. Lol.

Oh! my bad, thanks for the correction. I look forward to hearing from you soon.

God take me out of Nigeria fast..... I look forward to hearing interesting results from Facebook Data Warehouse.

Besides, the way these companies are investing in machine learning, deep learning and artificial intelligence is mind boggling and no Nigeria company (that I've heard of) has invested in these breaks my heart, later we would cry of brain-drain. If Google, Microsoft or Facebook try to even flash me not call ehn! i will dust my slippers and go with them forgetting any luggage in this country called Nigeria.

LagosismyHome:


if you have time , write it now and then GRE doesn't expire for next 5 years so at least you have it when you target 2019 application and then apply early

Alright ma, i will work on it with immediate effect. I will send a pm right away. Thanks.

LagosismyHome:


Dont let this your good cgpa waste at all... with this gpa the easiest way to have gotten funding is to Write GRE, score high and then apply for masters in the USA and not applying without GRE

Anyway since you still want to apply for phd in the US ... then best to take GRE. If u can study hard for it and write it before the master program start that would be better. Send me a pm if you need help in paying for GRE. You have a good gpa . You should have mentioned the above back in sept and i would have sorted you out based on this gpa for sure

Unfortunately, I would love to write but next year. 47,000 pounds is too much for me to ignore now (lol) but i would contact you anyways. Thanks very much sister, I really appreciate.

joeluv:
Congrats. That's a very good one. I see your post and see your very good with R. Am for fall 2017 in Applied economics and thr professor for my statistics course says been proficient with R would help a lot. Any suggestions on how I should go about learning this software. It's been a while I left school and I feel like a block head sometimes.

No need to feel like a block head bro, it will take time but you would master it eventually. To start with, you have to download and install R, Tinn R and R studio. Download R here: https://cran.r-project.org/bin/windows/base/ and R Studio here: https://www.rstudio.com/products/rstudio/download2/
You can download Tinn R (Just like a text pad for R) and search for tutorials on R online, I can help from there.

Best.

LagosismyHome:


Your gpa is quite high....did u write GRE ?. what school is this scholarship tied to

One can choose from any of:
Université Paris 1 Panthéon-Sorbonne, France
Universitat Autònoma de Barcelona, Spain
Universität Bielefeld, Germany
Università Ca' Foscari Venezia, Italy.

One can choose any of the four Universities but i chose Bielefeld for my first year and Universitat Autònoma de Barcelona, Spain for my second year.

I didn't write the GRE (i had planned to after paying application fee to some of the universities)

Good day everyone,

I want to use this medium to appreciate everyone here that helped in one way or the other the last time i visited here. Unfortunately, my American dream has been postponed to 2019. I have been awarded the European Union Erasmus Mundus Scholarship to study Quantitative Economics (MSc) and i hope to pursue the American dream in 2019 for PhD of course. Lol. Here is my brief profile:

BSc: Statistics
CGPA: 3.81/4.00
Referees: A Professor and an assistant lecturer
Research experience: Served as an Undergraduate Research Assistant (during undergrad days) and currently a Graduate Assistant
Publication: Published my undergraduate final year thesis, unpublished article which i placed on Academia.ed (don't know of it both count anyways )

Rejection List
1. Berlin Mathematical Society
2. Purdue University
3. African Institute for Mathematical Sciences (Application submitted late and I was advised to re-submit for Fall 2017, so it count as 'rejection' as well)
4. University of New South Wales (Application submitted late hence lab of my potential supervisor was filled, so it counts as rejection as well)

Admitted but no funding
Universitat Bielefeld (Mathematical Modeling in Engineering); Waiting List on funding Number 36
Austin Peay State University

To those that might be interested, the Consortium place much emphasis on your Motivational statement, this I'm sure of.

Thank you all!

To Lagosismyhome who advised me not to give up on Fall 2017 even when i couldn't pay for TOEFL. Your husband is a lucky man i must confess.

Again I say a big Thank you.

Maybe i would name my son 'Carson' as well, it seems they are all intelligent

One and that's the rabbit.

Bro, can I send you a PM?

I think I would give it a try.
Thanks bro!

Chai! Different people and different problems. When I'm thinking of how to pay WES evaluation fee of $205 as a serving Corper now and get graduate assistantship, another person is thinking of when he will start NYSC.

God is great...

Hello Mr. Anthropino,

Thanks for taking your time to reach out, I've been going through the articles on Statistical pattern recognition and Neural networks from a Statistical perspective; I've found it to be an interesting read though at a slower pace than I anticipated [I guess my mathematical knowledge are starting to betray me after a long time from academics] but I'm optimistic I would understand it hook line and sinker. I will get back to you as soon as I'm done with it. Cheers!

Have a beatific day!

Regards.

anthropino:
Hi guys,

This is a nice job @neayho ---you are the big boss. If you permit me to humbly make few suggestions:

1 ) It is highly recommended to use Jupyter notebook and Anaconda: (to provide neat code and explanations)

http://jupyter.readthedocs.io/en/latest/install.html

One can install Anaconda as the package manager :
>> https://www.continuum.io/downloads

With the above installations one can use both python packages and R.
The R packages can be installed ib the conda environment from the link >>https://conda.io/docs/r-with-conda.html

But if you dont like python we can use RStudio the equivalent of jupyter notebook:
>> http://rmarkdown.rstudio.com/r_notebooks.html

2) Provided here is a repository of examples of jupyter notebook: (link below) from online communities
>> http://nb.bianp.net/sort/views/


3) Why R is very good and the work perfect , one can also try out python libraries:
>> Numpy http://www.numpy.org/
>> Pandas http://pandas.pydata.org/
>> Matplotlib http://matplotlib.org/
and some other ones.


4) Why not consider these libraries for visualisation in your preprocessing (python):
>> Seaborn (https://seaborn.pydata.org/)
>> (http://bokeh.pydata.org/en/latest/)
>> D3JS (the best for interactive visualisation not pythonic https://d3js.org/)


These are just additional standard libraries common in every day job and these are the ones I could remember --accept my ignorance.

Lastly I admire the treatment of missing data. But missing data is a very heated topic among intellectuals. I humbly invite you to read this paper:
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1839993/

and read the book by Roderic and Rubin " Statistical Analysis with missing data"

I hope this helps!

Thank you very much for this! To start with, I hope i can take my time to learn Python as i heard it's a great software though I know it won't be easy to venture outside my comfort zone (R), can you recommend a book for a beginner in data science using python?

Besides, i learnt from Nathan Yau that D3js is a great tool for data visualisation but one has to have a domain knowledge of Javascript, hence that pissed me off.

Pertaining to the paper sir, i read it and was impressed with the application of various software to solve the problem of missing data though I'm familiar with virtually all the R packages listed there to solve the problem of missing data, I'm impressed with the application of Stata, I guess there's much for me to learn this year and I'm grateful you cared to contribute your quota.

Thank you very much for this boss, I hope you can shed more light on your research interests and provide journals for me to read so i don't become rusty. Again, I say THANK YOU.

I look forward to reading your reply soon.