What is data poisoning and is there an antidote

[rakhirhif8963]

Spyros Potamitis
Spyros Potamitis
Data poisoning can become a threat to infrastructure if machine learning (ML) is not properly managed. Spyros Potamitis, senior data scientist in the global technology practice at SAS, discusses the consequences for Information Age and how to avoid them.

More and more organizations are turning to ML models to advance their artificial intelligence (AI) technologies. However, another trend could pose a threat to the reliability of these systems: data poisoning. The key to successfully countering this is not simply fixing the problem after it occurs. To protect valuable data from it, companies must fully understand the severity of the threat, why data poisoning occurs, and how to protect against it throughout the entire AI system development process.

Back to Basics of MO
Before we get into what data poisoning is, it’s worth dominican republic mobile database back to how machine learning models work. They are trained to make predictions by being fed historical data. The very use of this data implies an expected outcome and the characteristics that determine whether it will be achieved. This data “teaches” the model to learn from the past. It can then use this knowledge to predict the future. Generally, if more data is available to train a model, its predictions will be more accurate and consistent.

AI systems that include ML models are usually developed by experienced data scientists. They carefully examine and analyze them, remove outliers, and perform several integrity and validity checks before, during, and after the model development process. This means that, as much as possible, the data used for training actually matches the goals that the developers want to achieve.