The survival rule of product managers has changed after the epidemic!
Nowadays, social e-commerce products, marketing short video products, data security products, artificial intelligence products, and robot customer service products are all booming. In social e-commerce, machine learning is used to recommend accurate products or services based on social relationships, and short video robots are used to test Huang , Data sets generated based on big data use machine learning algorithms to train and mine potential risks, etc.
Most of these products imply a job demand, which points to a kind of product manager: AI product manager.
Everyone is a product manager, but if you want to be the product manager closest to the CEO, you must survive the epidemic first, and you must be able to choose the right product direction after the epidemic.
This article first introduces the demand for AI product manager positions in many popular industries, and then introduces the AI skills that AI product managers must understand in detail, and then writes what is a neural network? What is machine learning? Finally, a case is used to disassemble the overall case of how AI product managers use machine vision to recognize handwritten fonts in detail.
What are the necessary skills for AI product managers?
AI product managers are not direct AI+ product managers. AI product managers have their own unique skills. For example, the basic level AI product manager should have linear algebra, calculus, and statistical probability theory as a base, and then the application level is best Have computer-related knowledge, and then at the algorithm level, at least sort out the technical boundaries that can be supported by the algorithms in your own industry and your products. Wait. These skills are a must for AI product managers. Now let’s introduce the specifics. What is a neural network algorithm? What is a machine learning algorithm?
What is a neural network?
The underlying model of artificial intelligence is the "neural network". Many complex applications (such as pattern recognition, automatic control) and advanced models (such as deep learning, machine learning, machine vision, machine hearing) are based on it. To learn to be an artificial intelligence product manager, the technical skill that must be understood must be neural network, which must start from it and then gradually deepen.
Then the neural network is not mysterious, the essence of the neural network is to imitate human thinking. Human thinking lies in the biblical network of the human brain. The specific thinking process of human neurons is as follows:
1. The external signal stimulation is transformed into electrical signals through nerve endings, and the electrical signals are transduced to neurons;
2. Many neurons constitute the human nerve center;
3. The nerve center integrates various signals to make decisions and judgments;
4. The human body responds to external stimuli according to the instructions of the nerve center.
5. So artificial neurons are the core technology in artificial intelligence products, which are regarded as neural networks.
A neural network contains input data, and output results. The processing of numbers in the middle requires the weights assigned to the weights of different numbers, and the threshold for which results are suitable.
Therefore, we often see the structure diagram of the neural network as follows: x represents the input, layer a is the neuron, and h is the output.
What is machine learning?
Machine learning is a data analysis technique that allows computers to simulate a skill that humans and animals have naturally acquired: learning from experience. Machine learning algorithms "learn" information directly from data through calculations, instead of relying on predetermined models. When the number of samples available for learning increases, these algorithms can adaptively improve performance.
In addition, the principle of machine learning is the neural network mentioned above, which also includes the input of data, the calculation of the intermediate neuron layer, and the output of the final result.
But machine learning itself is divided into supervised (labeling the input numbers), unsupervised (not labeling the input data) and semi-supervised.
Therefore, it can be understood that machine learning is a specific application of neural networks.
Case: AI product manager practice machine vision recognition handwritten fonts
Below the author, LineLian, uses an example to actually operate a neural network, also called a perceptron. In this case, it is called MLPClassifier (Multi-layer Perceptron Classifier).
The demand for AI products is to use neural networks to recognize handwritten fonts.
The principle of this case is also applicable to the camera (electronic police) on the road to identify the license plate, and other practical product cases related to computer vision (CV).
The AI product process is: import data, train models, optimize models, heuristic understanding, etc. As shown below:
The author follows the process of training the neural network in the above figure. First select the classic data set. Students who need training can leave a message on the official account and perform the following training:
Step Zero: Introduce a suitable data set
The focus in this step is to find a suitable data set.
Step 1: Design the neural network parameters
The points that the product manager should understand are:
sloer means optimizing the digital weight of the neural network, relu means activating the neuron, alpha means the parameter term, hidden behind the neural network layer and the number of neurons, and random means randomly generated numbers , Verbose represents the printing process, learning_rate='invscaling', used to update the effective learning rate.
Step 2: Generate the model and save the model
Step 3: Screen the model
Step 4: Cross-validate the model
Step 5: Load the data and select the accuracy rate as the specific index for the verification of the verification set model
Test the accuracy of the model with data.
Step 6: Adjust the large parameters of the neural network learning model. Here, adjust the training times of the neural network.
Step 7: Continue to train the neural network to adjust the large parameters of the neural network learning model, here adjust the number of nodes in the middle layer of the neural network n
Step 8: Output the training results
Step 9: Output the training results of the overall needle on the test set data
The operation process of the case in this article is universal in the practice of AI products. The focus of AI product managers in the process of making AI products is not on training parameters and large parameters of the model, or hyperparameters. But understand the meaning of some key functions inside.
That is:
1. Know how many layers of neural network your machine learning uses;
2. Know the data quality of your data set and the data grouping situation. For example, this article divides the data set into training set data and cross-validation set data And the test set data, and the training error data and cross-validation error data collected in the middle;
3. Knowing the code may not be proficient in writing the code;
4. Understand the meaning of some core functions, but the AI product manager may not be able to transform the function;
5. In addition to the application of such a mature machine vision in the recognition of license plates, criminals, and nursing homes, what needs this algorithm model can also solve? For example: combine other models to make a new functional application.
It's not impossible to make a face-changing product like the Douyin picture below.
The impact of AI on products is not just a face change, nor is it just a robot customer service (Chatbots). In theory, AI can redefine products. In practice, AI requires AI product managers to find the nail needs with AI technology. Before I say, you and I have the opportunity to make famous AI products. After ten years of rain in the rivers and lakes, let’s talk about AI and the life of an AI product manager. Your forwarding is my motivation for future follow-up!