Nhờ sửa bài giúp

Process of a Machine Learning Project Machine Learning technology has been used in our daily life widely, and one of the most popular applications is handwriting recognition.

This function can be used for word input of our cellphones and also can be used as number input in Banks.

It's a typical machine learning application and commonly referred as a student project, because it is not very complicated and can help beginners to understand the algorithms.

In the following text, a process of the project about hand writing recognition with neural network algorithm will be introduced.

Following the steps and achieve the milestones will help accomplish the project.

First, we need powerful computers to build our programming environment, while in the latter stage of this project we need to do lot's of calculation, and it can last for days.

So if we want to deliver our project in time, we need to consider the calculation ability of the computer.

After choosing the hardware environment, we should think about our development environment software.

This can be decided by personal hobbies, and remember to install the python-software of the same version in the team.

Of this way can share our code without some syntax problems.

After we successfully build our integrated development environment, we need to understand the concept of the neural network algorithm and figure out how can we implement it in our code.

Now, a smart way to do this, is to be relaxed and be concentrated to read some relevant research papers about neural network algorithm in handwriting recognition.

Since many researchers and scholars have already achieved some significant results about this, we need to learn from them, rather than just learn from the book.

And for beginners like us, it's really a short cut to learn from our seniors.

After this is done, we known the basic idea of how the algorithm works.

And now we can divide our work in more details and assign different parts to different team members.

The first part of the implementation is feature extraction.

What's is feature extraction and why we need this?

While we can get the dataset, which our program needs to learn form, from the internet, but the neural network algorithm cannot directly be trained with the graphics data.

Using the feature extraction to acquire the important information i.e. the features from these graphics datasets and changing them into matrixes and vectors.

With these this features, we can train our neural network model.

Next part is the implementation of the neural network model and train it with the features we got before.

Before action, we need to think about how many layers should we use and how many perceptron in each layer.

To do this we can read the papers again and search something relevant in internet and read some suggestions from others.

While after reading paper we find out, that basically more layers and more perceptrons can give a more reliable prediction result but more than two layers will slow down the training part significantly, and the results will hardly be much better than a two-layers-model in our circumstance.

So we can try the two-layers-model for further observation.

We implement the algorithm in the code and train it with all the data we get, then observe the results.

After the implementation of the algorithm and training, we come to the last part of our project, which is testing and improvement.

But how can we improve our performance after we fixed the neural network structure?

The trick is that we divided the training set into a big part and a small part.

Using the big part to train the neural network model and using the small part to test the results.

And after this, stop training the model, when the prediction-performance with the test dataset start to fail.

And we should divide the dataset in different ways and try several times.

At last we can get a neural network model with relative better trained parameters.

Finally, we should finish the most part of this project by now.

Then gathering all the material from the beginning to the end.

And writing a summary of this project to prepare for communication with professor about what we have done in the project and how the results looks like.

Doing successful machine learning project takes time, patience and cooperation.

Once we accomplish all the milestones before the deadlines, we can find out, that the process of the project is full of delight and not that hard.

As our experience grows, we will go deeper in the ocean of machine learning and maybe one day, we can make our own products and attribute it to our society.