While Artificial Intelligence (AI) is about machines doing tasks that typically require human intelligence, and Machine Learning (ML) about machines learning by experience and acquiring skills without human involvement, Deep Learning (DL) is about mimicking how the human brain works.
DL, a part of ML methods based on Artificial Neural Networks (ANN or just NN), was introduced in the 1980s. It became popular recently as large amounts of data and computing power are already available. It matters because of its accuracy, achieving recognition accuracy at high levels that sometimes may exceed human-level performance.
How ML and DL are different?
ML uses algorithms to parse data, learn from it, and make decisions based on what it has learned, while DL on the other hand structures algorithms in layers to create an ANN that can learn and make intelligent decisions on its own.
When choosing between ML and DL, considering the availability of a high-performance GPU and lots of labeled data is a must. If not available, it may make more sense to use ML instead of DL.
Are DL and NN different?
The difference between NN and DL lies in the depth of the model. DL is a phrase used for complex NN. The complexity is attributed by elaborate patterns of how information can flow throughout the model.
What is a Neuron?
A Neuron, a basic working unit of the brain specialized in transmitting information, has a Body, branches (Dendrites) that are a receiver of signals, and Axon that is a transmitter of signals. Dendrites of a neuron are connected to Axon of other neurons. The neuron is by itself useless. Neurons work together to make a result.
How NN work?
Let’s consider an example of evaluating a house based on different variables. Each variable(s) is(are) impacting a hidden neuron that is transmitting output for a prediction later on. Input values (independent variable) have weights, allowing NN to learn which signal is important and which is not. Activation Function is being applied on the hidden layer according to the weights to predict output value by passing the signal on.
The model learns to estimate the true answer based on the given inputs it has been fed. You create a facility for the program to figure out and learn on its own. Ex: how to distinguish between dogs and cats by telling how each looks like.
Moreover, below another photo showing the input values and their weights, being used to predict an output value to be compared to the real one. It is about updating the weight for the predicted output value to become closer to the real one, by minimizing the cost function C, which is equal to the sum of half of the difference between output value and real value squared (C=1/2∑(output value – real value)². Weights to be adjusted until C is minimal, and the difference between the predicted value and the real one is also minimal.
What is a Convolution Neural Networks (CNN)? How different are they from ANN?
In mathematics, convolution is a mathematical operation on two functions to produce a third function that expresses how the shape of one is modified by the other.
Convolution Neural Networks CNN, part of ANN, are being applied for a variety of learning problems. They are quite effective for image classification problems. CNN may be about self-driving cars recognizing stop signs, FB auto-tagging photos by facial recognition, medical centers detecting cancer cells, Image colorization, etc.
Both CNN and ANN have learn-able weights and biases, and in both networks, the neurons receive some input, perform a dot product and follow it up.
Computer Engineer • Entrepreneur • Blogger