What is deep learning? The concept of deep learning comes from the research of artificial neural networks. A multilayer perceptron with multiple hidden layers is a deep learning structure. Deep learning forms a more abstract high-level representation attribute category or feature by combining low-level features to discover distributed feature representations of data. Perhaps such a professional explanation may not be well understood by everyone, and driverless driving is even more half-understood. I have obtained some professional knowledge from the three-year experience of researching deep learning algorithms to extract some principles that are easy for the public to understand.
Explain artificial intelligence well and understandably
Artificial intelligence started to catch on. When everyone started talking about artificial intelligence, perhaps it originated from the “Go Challenge”. The world’s top masters were eventually defeated by a robot “dog”. Its deadly magic weapon is to imitate the working principle of human “deep learning”. How do robots learn like humans? Can it really be used in autonomous driving to serve humans? Do you think and judge like people?
Firstly, we start the dissection from the deep learning brain . The deep learning brain is a pair of complex calculation formulas. Input conditions are given from one end of the formula, and the output result will be obtained after the calculation. Many parameters covered in the formula can be understood as the result of deep learning network training through a large number of samples. The left side of the figure below is the input condition, which can be the data of a picture, such as the front camera of an autonomous car; it can also be some physical quantities, such as the distance information of the obstacle in front of it measured by some distance sensors. So what is the output? The output layer is the right side of the figure below. The result obtained through the calculation of the multi-layer neural network should be the driving decision, such as brake control, throttle control and steering wheel angle.
Professional explanation of the core algorithm convolution operation
Convolutional Neural Network (CNN) is a feed-forward neural network. Its artificial neurons can respond to a part of the surrounding units in the coverage area and have excellent performance for large-scale image processing. This algorithm is mainly used to process image operations. CNN is mainly used to identify displacement, scaling and other forms of distortion invariant two-dimensional graphics. Since the feature detection layer of CNN learns through training data, when using CNN, it avoids the displayed feature extraction, and implicitly learns from the training data; in addition, because the neuron weights on the same feature mapping surface are the same, So the network can learn in parallel, which is also a major advantage of convolutional networks over networks that connect neurons to each other. Convolutional neural network has unique advantages in speech recognition and image processing with its special structure of local weight sharing. Its layout is closer to the actual biological neural network. Weight sharing reduces the complexity of the network, especially multi-dimensional. The feature that the image of the input vector can be directly input to the network avoids the complexity of data reconstruction in the process of feature extraction and classification.
Understand and explain the core algorithm in seconds
In fact, the convolution calculation is to extract a certain range of feature data on the image through matrix operations, and use the features to calculate the final result. This process can be explained by a dynamic picture. In fact, all graphic data is composed of pixel data, and each pixel data can be an integer, which can be used for calculation.
So, how can deep learning be used for autonomous driving? Autonomous driving requires a car to recognize things that appear in front of the car and make decisions like a human brain. The deep learning network is equivalent to the human brain. It collects the image of the camera installed in the front of the car, and proposes the characteristics of the image through the convolutional neural network, and obtains several output quantities through model calculation, such as acceleration, deceleration, braking, Information such as the angle of the steering wheel. This is a simple explanation as a deep learning technology otaku.
However, deep learning or artificial intelligence technology cannot achieve 100%. This is a crucial issue. This is one of the reasons why deep learning has not been applied. Because any driver is unwilling to give his life to an event that cannot be 100% probabilistic.