Recruit the first machine Learning Demo
In 1958, Frank Rosen Bluet Demonstrate the remarkable thing for reporters in Washington, DC his “Peripperron“You can see the cards on them and tell them where the shape is. It was noteworthy that the system was not clearly programmed how to do it – how to do it by looking at the examples.
In traditional systems, a programmer thinks about the inputs of a program and comes with data structures and algorithms to solve the problem and create valuable results. Human programmer is the star of the show.
The machine learning system does not require any programmer How To solve a problem. With Under supervision machine learningOf the system Inputs and outpats Are addicted to Train The system is a fundamental change in the building system so that they can recognize the samples in fresh inputs and predict the output. Learn The show is the star.
These Code playback Recapped the Rosen Bluettic experience using modern object -based programming techniques. You will see two solutions to the same problem. The first uses traditional programming where I come with the algorithm to solve the problem. The second machine is a very easy use of learning Learns How to solve the problem.
Issue
Each card has a rectangular shape on its left or right. In the original Demo of Rosen Bluet, the cards are photographed and then it has been converted into 20×20 pixel images. In this program, I imitate cards and photos. The program’s job is to look at and predict new cards.
The traditional point of view
Code playback A solution starts with which I have come out. I thought about the system input, 400 pixels from one icon, and the algorithm to count the active pixels on all sides. The one who has more pixels is the aspect I ‘predict’ is the form of the form.
It works well. All 500 test cards are predicted. But I had to tell the computer exactly what to do.
The machine learning point
Then comes the prescription, this machine was one of the first examples of learning. It is given many different inputs (400 pixel images) and output (labels associated with each card, stating that the shape is continuing). Peripperone takes input and output and learns how to predict where the shape is.
As a programmer, I don’t tell how to do it. I compiled the program so that she could learn from what she sees during training.
My paracepteron uses 400 weights (one for each pixel position). Initially, all weights are set at zero. The Press Pracetrone predicts the price of each pixel by multiplying the value of its relevant weight, then offers a summary of everything. If the money is negative, it will predict the left. Otherwise, it will make the right prediction.
Pressuprone trains on labeled examples. When it makes a false prediction, it learns by adjusting the relevant weight. After training, it does a really good job of predicting the test card.
Will you learn
Code playback Runs through you:
Card representation and making pixel data
Implement the non -AI solution (so you can see the traditional approach)
Creating a prescription class with weight
Understanding the forecast method (the combination of products))
Seeing how training updates weight based on errors
Seeing the weight of the weight from random values to see the samples
Understanding why some initial points fail and how to fix them
You will see the original weight values after training. Some samples are clear. Something is amazing. Playback challenges you to understand and explain these patterns.
Why does it make a difference
This machine is one of the easiest example of learning that you can study. Perfection is basically the same neuron. Modern nerve networks stack many neurons in layers. They can learn too much complex samples, but the basic idea is the same. This is a good way to start your journey if you want to know about nervous networks, machine learning, and AI.
Interactive learning
View the full code playback here
Playback includes some challenges by asking some questions along the way. Can you know how much training examples are actually needed? When does the prescription errors leave during training? The code is available to you to download and experience, which will allow you to recover a piece of AI history. The Rosen Bluetiyyyyyyyyyyyy is the place where it all started.
The code playback medium is different from traditional videos or lessons. You are guided through a code with a story and you get to see my thinking process. If you like this format, I have more free content on my site, Playback press. Feel free to share your comments, questions, or feedback via email: mark@playbackpress.com.