AI Concept of the week - Prediction
AI for Prediction
Ability to predict what is going to happen in future is always considered as a sign of knowledge, skill, and vision. Why it is important? Because we can be better prepared by appropriate actions before something happens. The typical examples of predictions are “is it going to rain tomorrow?”, “what will be the price of the stock of a company by end of year?”, “Is this customer going to buy the product he/she is browsing at? , “is this candidate going to win the election”?, and so on. Correct answers to these questions have huge benefits obviously. Experts make such predictions from their “experience” – what they observed in the past and what patterns they identified. The word “experience” has a deeper meaning. It involves remembering the historical information and analyzing the collected information to observe any patterns that help to extend the pattern to what is happening currently (current contextual data). Since computers are good at storing massive amounts of data and good at executing algorithms (step by step instructions) that can quickly identify patterns in data, prediction of an outcome or an event is a natural problem. These are often statistical, and AI based algorithms which are widely used for many predictive AI use cases.
Let us look at a very common use case – weather prediction. All of us have heard and seen weather predictions in TV and radio channels as well as on web portals. How do they come up with temperature, humidity, wind, and rain information for the next day? They use weather prediction models and look at the data such as wind speed, air pressure, clod conditions, air and water temperature, and similar data and use the model to calculate the next changes for the coming days. They collect this data sent by satellites, ground stations that sense wind speed, temperature, air pressure, etc. How are these models built?
The weather prediction models are often built using AI techniques. One such technique is called “deep learning algorithms”. Deep learning techniques employ artificial neural networks. In these systems, the conceptual processing nodes called “neurons” connected in a multi-layer network are fed data and cooperate to solve a problem, such as recognizing images, identifying patterns, classifying objects, and so on. The links between nodes that "cross-feed" the neuron in the next layer is designated with a "weight" indicating how much each of the input should influence the "neuron" it feeds. The neural net repeatedly adjusts the weights of these links between its neurons and sees if the resulting patterns of behavior are better at finding a solution. Over time, the network discovers which patterns are best at computing results. These computing nodes and connections between them mimic human brain, hence the name neural networks. The aspects of a neural net that change with learning, such as the nature and weight of the connections between neurons, are known as its parameters. A neural network is called “deep” if it has multiple layers of neurons. Once the model is "fully" trained, now new set of inputs can be fed to output a recognized pattern or binary classification,
In weather prediction, satellite images are first divided into granular grids and digitized to create a set of data. Hundreds of thousands of such image datasets are used to train the neural networks so that it learns how to recognize different weather conditions given a set of inputs. Such deep neural network models can then be used to feed a new set of images and can output what it can predict based on the several thousand patterns it has seen earlier during the "learning" phase.