- "Hybrid" Machine Learning (ML plus Stochastic Optimization)
- Knowledge Gradient (KG) approach
- Adaptive Machine Learning
- Knowledge Gradient
"Hybrid" Machine Learning (ML plus Stochastic Optimization)
ORA belongs to a class of hybrid Artificial Intelligence (AI) solutions that are at a mid-point between, on one hand, rules-based algorithms (classical “knowledge engineering”), and – on the other hand - pure “deep learning” (i.e., AI systems that purposefully avoid incorporating a priori knowledge).
While also relying on deep learning, ORA nevertheless uses as a starting point stochastic optimization models co-developed by RoomSage with the team of researchers from Princeton University in the US.
The models allow ORA to learn the optimal bidding policies for sponsored search auctions, focusing on hourly variations in auction frequency, click propensity, bidding competition, and revenue originating from advertisements.
ORA relies on the concepts of Adaptive (or “closed loop”) Machine Learning (AML).
Particularly relevant in a fast-changing competitive environment, the AML-based algorithms avoid the “ageing” of the learned solutions, while giving more weight to more recent data.
As a result, ORA algorithms can be “smarter” than more traditional deep learning tools while operating on much smaller datasets
ORA relies on the Knowledge Gradient (KG) approach
– a way to squeeze information in a most efficient way out of a given data set.
Knowledge Gradient takes into consideration the marginal value of information derived from a data sample (i.e., it constantly trades off the value of such information vs. the cost of acquiring it).
Knowledge Gradient can be compared to a type of cartography technique where one can construct a path leading through the supplied data sets, and whose ultimate end point represents the most optimal ad campaign strategy. The result is that ORA reaches optimal performance levels in much shorter timeframe than other learning systems.
operations per second
sources of data read every minute
times faster than human
times more accurate than human
TB of data and still growing
We live in the age of the Fourth Industrial Revolution
After steam, electricity, and computers, now the change is driven by Artificial Intelligence. The most sophisticated algorithms ever created by the human race drive cars safer than humans, pilot drones, recognize objects in images, play chess much better than any world chess champion in history, translate between any languages, protect us from spam, help us find the content we like on YouTube or Facebook, help doctors diagnose cancer and soon AI will help judges decide if an arrested person should be held in custody.
Should your company use Artificial Intelligence?
To answer this question think if you can imagine a contemporary company not using electricity.
Where is the limit?
Artificial Intelligence can be applied to any task where data is involved, and if there is a lot of data the AI has a huge advantage over humans.