What is incremental learning?
Incremental learning is a new approach ACI is taking to ensure our machine learning models perform to the highest quality, for longer than other current methods on the market. It was developed by ACI’s Data Science team and was the subject of a patent filing in January.
How is it different from self-learning models on the market today?
Our incremental learning algorithm allows our models to adjust to new behaviors without the need to re-learn everything they already know. We feed the algorithm with new data every 24 hours, and it spots new behaviors as they happen. If the change in behavior is enough to cause a degradation in performance, the model will adapt to what it has recently learned. This ensures that model performance quality lasts for longer periods of time, without degradation.
In contrast, the majority of self-learning models still need to be retrained on large amounts of historical data if fraud behaviors change, to maintain performance. While our algorithm can revert to learning over a longer period of time if necessary, our incremental learning approach reduces the need to train models over very large data sets, or to continuously re-deploy to production and still maintain performance and reliability in predicting fraudulent behaviors.
It’s like the difference between a pianist learning to play new tunes incrementally, while still remembering the old ones – instead of having to re-learn all the old tunes with each new one.
What are the key advantages of incremental learning?
Our customers are protected without sacrifice of performance, even when changes in fraudulent behavior are evident.
- It reduces the number of model deployments required for our customers.
- The process is automated and does not require additional resourcing by our customers.
- Most importantly, it performs better than traditional machine learning methods.