Attempts to predict events is perhaps as old as the first agrarian civilizations. Knowing when to plant crops could make the difference between survival and starvation. Thankfully, prosperity in modern times has allowed us to move past subsistence to less urgent tasks. The ability to forecast everything from the weather to sporting event scores continues to elude absolute accuracy. Still, computer science is making strides each year in the pursuit of prediction accuracy. A relatively new tool being applied to this effort is that of machine learning. A variety of artificial intelligence tools and techniques are being applied to tasks such as demand forecasting and predictive analysis. These approaches are producing steadily improving results.
Foundational work in the 1940s and 50s, exploring nerve cells and how they work, launched artificial neural networks as a possible basis for artificial intelligence. Computer hardware limitations of that era slowed development in this field. Early processors were incapable of completing the number of calculations required within practical time frames. The development of more sophisticated algorithms and hardware improvements during the 1980s reignited interest in neural networks. Today, they are becoming a dominant force in predictive tasks. Weather forecasting, stock price performance and self-driving cars are a few applications benefiting from these technologies.
Numerical weather prediction (NWP) models have been the generally accepted method for forecasting weather patterns. These are models based on physical equations applied to current weather conditions to predict future conditions. By contrast, artificial neural networks use past weather observations to train or teach deep learning algorithms. Once trained, the network mimics dynamic climate conditions to predict future weather. While early results are promising, additional research is needed to establish reliability and accuracy in this field.
Stock Price Forecasting
The prediction of future stock market performance is a highly sought after domain of knowledge. Past efforts to forecast stock values have included statistical models and various mathematical tools. With the ability to tune various parameters and inputs, artificial neural networks are uniquely well suited to this type of task. Past stock performance information supplies ample training data for these learning algorithms as well.
Self-Driving Vehicle Forecasting
Two primary tasks required for autonomous vehicles are image recognition and path prediction for various objects near the vehicle. Image recognition is necessary to read and obey posted speed limits and other road signage. Another vital task is that of predicting the paths of other vehicles and pedestrians to avoid collisions. Recently, dramatic improvements in image recognition tasks have been achieved by artificial neural networks. Accurate predictions for collision avoidance are also achieved using these networks.
Predicting outcomes and events has been a powerful motivator enabling progress. The quest to see into the future has progressed from consultations with oracles to modern computer science and artificial intelligence. Data science, artificial neural networks and deep learning algorithms are making steady progress toward ever more accurate forecasts.