The process of selecting the correct data is partially automated, but still requires the intervention of data scientists. This is due to the presence of abnormal values or outliers in databases, which cannot always be processed automatically. Specialists must decide which of these anomalies should be removed and which should be left.
An example would be a bank analyzing data about customers and their mortgages. If a customer has a value of 100 in the column for "number of children," that is clearly an outlier and can be automatically removed. However, a value of 10 or 20 may be an anomaly, but it is still real and important to keep.
Large data sets may turn leads into sales with overseas chinese in worldwide data contain errors, so it is not always possible to fully trust the decisions of neural networks. It is important to avoid overfitting neural networks, where they overfit the available data, as this can reduce their ability to discover new, important decisions.
For example, a neural network trained to detect spam might be overtrained to the words "millionaire" and "inheritance," and if a spammer changes one of those words, it might not recognize the email as spam.
Practical application of neural networks in business
Bots Make Pizza Based on Information from Images
Researchers at the Massachusetts Institute of Technology (MTI), together with the Qatar Computing Research Institute (QCRI), have developed a neural network called PizzaGAN that learns to make pizza. The GAN in PizzaGAN stands for Generative Adversarial Network and is a type of neural network.
This neural network learns to make pizza by analyzing thousands of images of this food. After training, it is able to not only identify different toppings, but also determine the order of their layers on the pizza. The system can create step-by-step recipes based on a single image. Initial tests showed that the neural network correctly determines the order of the toppings in 88% of cases. Researchers suggest that this technology can be applied in other areas, not only in pizza making.
Read also!
"KPI for the commercial division: calculation methods and adaptation"
Read more
McDonald's restaurant chain offers menu adapted to current weather
McDonald's has acquired Israeli startup Dynamic Yield for $300 million. Dynamic Yield uses consumer data in retail through predictive technologies based on neural networks.
This will allow McDonald's to gain more information about its customers, especially those who order food from the drive-thru. The neural network will remember customers' preferences based on their purchases and use this information to predict their future orders.
Screenshot from the official Dynamic Yield website