Artificial Intelligence

Model Testing Strategies

noreply@blogger.com (Vikas Goyal) — Tue, 30 Jun 2015 00:54:00 +0000

For supervised learning models the learning efficiency of model largely depends on the quality of training data.

As a thumb rule, models should be trained using as large & varied dataset as possible. But in case of limited data here are some of the training / testing strategies which can be used.

Train & Test using same data

This is most simple but least recommended. As most of the time we will see excellent performance on test data but there is a huge risk of model not performing well for unseen data. This can be a case of over fitting.

Split Data

This involve splitting the available in some proportions e.g. 80:20 for training and testing. This ensuring that we don’t have over fitted model. The challenge here remains which 80% to take for training. To improve learning we can do multiple splitting.

Cross Validation

Cross Validation is the most recommended splitting technique where the data is split into k equal parts (k-fold). Usually k=10. The model is then trained and tested 10 (k) times but with different training and testing data each time ( 80:20) to ensure each dataset takes part in both training and testing but not during the same run.

Machine Learning–Selecting the right algorithm

noreply@blogger.com (Vikas Goyal) — Mon, 29 Jun 2015 09:34:00 +0000

Microsoft Azure Machine Learning team has published an extremely useful decision tree for identifying the right algorithm to use for your machine learning problem. Its released as a cheat sheet poster. It can be used even if you do not plan to use Azure ML as your ML platform.

Go grab it from here.

Statistical Learning vs Machine Learning

noreply@blogger.com (Vikas Goyal) — Sun, 28 Jun 2015 10:33:00 +0000

The comparison between the above two is really interesting. While they look so different yet they are so same. Both deal with prediction. While it is called machine learning mostly of the community of software engineers, it is called Statistical Learning by statisticians. Can we call Machine learning as marketing name for Statistical Learning ?

Free book on Machine Learning

noreply@blogger.com (Vikas Goyal) — Sat, 27 Jun 2015 14:43:00 +0000

Microsoft has published free eBook on Azure Machine Learning which is a service based offering on Machine Learning from Microsoft.

Grab it from here :

* Singularity Summit 2010

noreply@blogger.com (Vikas Goyal) — Tue, 22 Jun 2010 16:50:00 +0000

http://www.singularitysummit.com/summit

How advanced will AI be in 2100 ?

noreply@blogger.com (Vikas Goyal) — Sat, 06 Mar 2010 19:51:00 +0000

There is an interesting article published on silicon.com website which publishes thoughts on above subject from leading AI experts. Here are snapshots :

Hugh Loebner : by 2100 we'll have human equivalent or superior machines.

Ray Kurzweil : That's past the Singularity.

Eric Horvitz : I expect to see a lot of autonomous operation in robots.

Kevin Warwick : ..we will have gone through the Singularity..

Read the complete article here .

What is the ‘Turing Test’ ?

noreply@blogger.com (Vikas Goyal) — Thu, 22 Oct 2009 16:21:00 +0000

The Turing Test, proposed by Alan Turing (1950) in his paper ‘Computing Machinery and Intelligence’, was designed to provide a satisfactory operational definition of intelligence.

It proceeds as follows: a human judge engages in a natural language conversation with one human and one machine, each of which tries to appear human. All participants are placed in isolated locations. If the judge cannot reliably tell the machine from the human, the machine is said to have passed the test. In order to test the machine's intelligence rather than its ability to render words into audio, the conversation is limited to a text-only channel such as a computer keyboard and screen.

http://upload.wikimedia.org/wikipedia/commons/thumb/e/e4/Turing_Test_version_3.png/180px-Turing_Test_version_3.png

Loebner Prize : Its an annual competition which gives awards to applications which come close to passing Turing Test as much as possible. It started in 1990. Loebner Prize Website.

The Alan Turing Year : June 23, 2012, is the centenary of Alan Turing’s birth in London. 2012 will be celebrated with lot of events around Artificial Intelligence. Alan Turing Year website.

History of Artificial Intelligence

noreply@blogger.com (Vikas Goyal) — Mon, 12 Oct 2009 17:33:00 +0000

Before 20th Century

Started to appear in fiction
Aristotle invented syllogistic logic, the first formal deductive reasoning system.
Descartes proposed that bodies of animals are nothing more than complex machines.
Pascal created the first mechanical digital calculating machine (1642).

From 20th Century

1943 - McCulloch & Pitts: Boolean circuit model of brain, laying foundations for neural networks.
1950 – A. M. Turing. Introduction of Turing Test as a way of operationalizing a test of intelligent behavior.
1956 - Dartmouth meeting: "Artificial Intelligence“ name adopted.
1952 – 1962 - Samuel's checkers program
1957 - The General Problem Solver (GPS) demonstrated by Newell, Shaw & Simon.
1958 - John McCarthy (MIT) invented the Lisp language
1962 - First industrial robot company, Unimation, founded.
1966 - Ross Quillian demonstrated semantic nets.
1968 - Marvin Minsky & Seymour Papert publish Perceptrons, demonstrating limits of simple neural nets.
1974 – MYCIN – rule based system
Mid 1980s - Neural Networks become widely used with the Backpropagation algorithm
1997 - Deep Blue defeated the reigning world chess champion Garry Kasparov

Subdisciplines of Artificial Intelligence

noreply@blogger.com (Vikas Goyal) — Fri, 09 Oct 2009 04:55:00 +0000

Machine Learning & Neural Networks : How can a system learn from experience, or from past case data.

Distributed AI : Collaborative AI systems.

Problem Solving, Search & Decision Making : How can an AI system go through all the possibilities in a systematic manner when looking for solutions to complex problems. Searching for the right or optimal option to achieve goal. Rational & Uncertain reasoning like Fuzzy logic.

Knowledge Representation : How do we represent knowledge about the world in a formal manner that can be manipulated in a sound and efficient manner?

Natural Language : How can a system communicate in a natural language such as English.

What is Artificial Intelligence ?

noreply@blogger.com (Vikas Goyal) — Mon, 05 Oct 2009 15:39:00 +0000

Artificial Intelligence or AI is the field of science which deals with making computational systems like computers intelligent. The intelligence can be similar to human intelligence or beyond that.

The Intelligence here means ability to :

learn and solve problems
act rationally
understand
infer , etc.

It is artificial because it has been created by humans and not nature. The intelligence which humans posses is natural.

Will it be possible for humans to posses artificial intelligence ? Yes, I believe in future.

Following are some of the attributes of systems having Artificial Intelligence:

Natural Language Processing : Ability to understand instructions and communicate in a natural language like English.
Knowledge Representation : Ability to understand knowledge and store it for further use.
Automated Reasoning : Ability to use the stored knowledge to solve problems.
Learning : Ability to expand knowledge based on new situations and thus act based on experience.

Dharmendra Modha : Neuroscience and Computing

noreply@blogger.com (Vikas Goyal) — Fri, 02 Oct 2009 17:13:00 +0000

4. What sparked your interest in engineering computing systems?

Modha: Neuroscience has experienced a veritable revolution furnishing a tremendous amount of quantitative data about the structure and dynamics of the brain. At the same time, supercomputers with exponentially increasing computation, communication and memory capacities are becoming available. As I became aware of the possibilities engendered by a convergence of these two trends, I realized that the time is now ripe for a bottom-up approach to understanding how the mind arises from the brain. As a first step in this quest, at IBM, we have recently built a near-real-time mammalian scale simulator (C2) that can integrate—in a single computational platform—the wealth of neuroscience. C2 is akin to a linear accelerator, an electron microscope, and is kind of a wind-tunnel for the mind.

5. How will your current project to design a computer similar to the human brain change the everyday computing experience?

Modha: While we have algorithms and computers to deal with structured data (for example, age, salary, etc.) and semi-structured data (for example, text and web pages), no mechanisms exist that parallel the brain’s uncanny ability to act in a context-dependent fashion while integrating ambiguous information across different senses (for example, sight, hearing, touch, taste, and smell) and coordinating multiple motor modalities. Success of cognitive computing will allow us to man and mine the boundary between digital and physical worlds where raw sensory information abounds. Imagine, for example, instrumenting the world’s oceans with temperature, pressure, wave height, humidity and turbidity sensors, and imagine streaming this information in real-time to a cognitive computer that may be able to detect spatiotemporal correlations, much like we can pick out a face in a crowd. We can profoundly transform the productivity and security of the society.

Read complete Point of View

Artificial Intelligence and Singularity

noreply@blogger.com (Vikas Goyal) — Fri, 02 Oct 2009 12:55:00 +0000

The Singularity is Near Book by Ray Kurzweil

“The Singularity is an era in which our intelligence will become increasingly nonbiological and trillions of times more powerful than it is today—the dawning of a new civilization that will enable us to transcend our biological limitations and amplify our creativity.”

About Ray : “Ray Kurzweil is one of the world’s leading inventors, thinkers, and futurists, with a twenty-year track record of accurate predictions. Kurzweil was selected as one of the top entrepreneurs by Inc. magazine, which describe him as the "rightful heir to Thomas Edison.”

Movie : The Singularity is Near based on above book to be released in late 2009

The Singularity Institute for Artificial Intelligence

The Singularity Summit 2009

Microsoft Research Project : Singularity

“Singularity is a research project focused on the construction of dependable systems through innovation in the areas of systems, languages, and tools. We are building a research operating system prototype (called Singularity), extending programming languages, and developing new techniques and tools for specifying and verifying program behavior.”

Artificial Intelligence : Is interest decreasing ?

noreply@blogger.com (Vikas Goyal) — Fri, 02 Oct 2009 07:05:00 +0000

While the field of AI shows lot of potential in terms of solving lot of complex problems still the following graphs from Google Trend show that interest seems to be decreasing in general public and the field is still limited to universities and researchers.

The search volume index has decreased while the news reference volume has increased.

Below graph shows another interesting fact

Most of the searches are coming from developing countries where AI is not so popular among universities.

What could be the reason for decrease in above graphs ? Any comments ? Is it still a fiction ?

The Singularity Summit 2009

noreply@blogger.com (Vikas Goyal) — Fri, 02 Oct 2009 05:06:00 +0000

“The Singularity is the technological creation of smarter-than-human intelligence”.

Singularity Overview

The Singularity Institute for Artificial Intelligence

Why we need Artificial Intelligence ?

noreply@blogger.com (Vikas Goyal) — Thu, 01 Oct 2009 14:40:00 +0000

Ok. so here we are today .. its 1/Oct/2009. This is second post on this blog and the first one was on 11/Oct/2006. There is a gap of three years between two posts. Insane me. Exactly that’s why we need artificial intelligence because it is supposed to be sane or rational.

Now that I have come back, I promise to be as regular as possible.

I will be starting from basics with AI in general. Later I will start getting into details about subfields of AI like Neural Networks, Fuzzy Systems, Expert Systems, etc.

If you have any feedback, content to be shared on this blog.. pls do contact me.

Applied Artificial Intelligence

noreply@blogger.com (Vikas Goyal) — Wed, 11 Oct 2006 10:10:00 +0000

This blog is dedicated to publish all the work i do in the field of Artificial Intelligence.
Be it Neural Networks , Fuzzy ... etc.