Hello everyone!
I would like to learn how to handle different projects intelligence artificile (possibly with python).
Googling seems that the libraries are the most common Scikit learn and Pybrain.
These two libraries use very different approaches, the first and most statistical, the second with neural networks.
1) in addition to the statistical method and neural there are other methods?
2) what method it will ensure better results for supervised learning (classification, regression) and unsupervised?

In general my question is that I do not know where to start, there are several algorithms to do the same thing (for example, facial recognition or voice you can do both with statistical methods, both with neural networks) and do not know where to start studying.

thanks

What to study depends on what you want to accomplish, but I’d say a really good place to start is with early childhood development and neurology… 90% of learning about how to create AI is really learning how -your mind- functions.

“in addition to the statistical method and neural there are other methods?”
There are wayyyy simpler and easier to understand methods, if you’re just starting to learn about this stuff. For example, you could use ‘conditional-based learning’ (which most early chatbots used) where your program simply records every input to its output as a potential response.

Here’s a more detailed example of how this works:

—Session 1—-
User: Hello.
Bot: Hello.
User: How are you? //stored as a possible response to “Hello.”

—Session 2—
User: Hello.
Bot: How are you?
User: I am fine. //stored as a possible response to “How are you?”

It’s by far the easiest/simplest form of ‘artificial intelligence’ to start with as a learning basis, and requires no messy dictionaries or keyword tables.

As an end-note: make sure you understand exactly what an ‘algorithm’ is and how one works before attempting to use one.

Artificial intelligence is a huge field. You should start with determining what you want to do, otherwise you’d end up being a scholar in 33 different branches of AI, of which 30 that you’ll never use. But if that is what you want, you should take a course. e.g. this one (randomly picked from Google)

Iacopo Guarneri - Apr 30, 2015:

1) in addition to the statistical method and neural there are other methods?

Here is a diagram
Neural nets use statistical correlations too.
Expert systems and logic are a classic branch of AI that doesn’t use statistics but knowledge databases and inference. It has dependable results but is limited in flexibility and requires the most programming.
Another main method is genetic algorithms aka evolutionary AI, which randomly runs simulations to achieve a given goal, rewarding the simulations that get closest to the goal until the AI has literally stumbled upon the best method. It requires very little programming but sticks in the mud a lot.
Swarm intelligence is a branch of AI mostly used in robotics to get many small units to interact as a whole.
Reinforced learning can be applied to genetic algorithms, statistics, neural nets and traditional AI: Basically it is a method of telling the AI that it’s doing well or not by means of feedback or score. And then there’s “deep learning” and “machine learning”, which nowadays are mainly buzzwords for neural nets of various structures. Fuzzy logic and Turing Test shouldn’t be on the diagram in my opinion.