The machine learning model is a mathematical equation

what is a function

it’s important to keep in mind that a function is just a mapping from inputs to outputs, and these can take many forms.

function is intelligence which is static and from human

risk of deep learning

Just a few erroneous data points could seriously skew the model.

way to solve the problem

To avoid overfitting, it is important to have a large quantity of training data that is representative of the sample the model is expected to face.

why neural network so powerful

In summary, neural networks are powerful machine learning tools because of their ability to (in theory) learn any function -> note : function is intelligence

what is needed in deep learning

In practice, training an accurate model in a reasonable amount of time depends on many factors, such as optimizer, model architecture, quality of data, and many more

definition of deep learning

They are equations that can represent an extremely broad family of relationships between input and output, and where it is particularly easy to search through this family to find the relationship that describes the training data.

why don't call it deep searching

neural network is more than one function

it is more than a function. it maybe a mutiple function container , is more like a environment where needed function can be added . why it is so intelligenct probably is that it can put any function inside itself and make these functions interact properly and show more than one feature.

??e.g. a multivariate binary classification model is maybe composed by mutiple single univatiate binary classification model and a integraor which integate them together and work properly

the key problem of researching deep learning

how to make multiple function work together!!! is the most core reason why deep learning so intelligent

this is very similar to human learning

There are several different approaches, but one technique is to use deep neural networks to build a mapping from the observed state to an action.

some thought of gradient descent method

this kind of method is used for problems which has no closed-form expressions to solve . so it is also means this method is only used for finding partial extremum and it cannot be used to find global extremum. cause to find the global extremum is what only closed-form expression can do.

inplement of teaching

Here, we might set up a reward structure based on capturing pieces, or just have a single reward at the end of the game for winning.

use this method , AI will be trained in any shape that people want .

people can teach AI by "reward" either to play in a certain style or just to win by the best method AI learns by itself.

provement of thinking

As we add more hidden units, the model can approximate more complex functions.

some evidence to prove the piecewise function fitting

Deep networks can produce many more linear regions than shallow networks for a given number of parame- ters.

my thinking : it is a process of creating smaller fitting of piece-wise function for before layers