“if you want to go fast, go alone. If you want to go far, go together”

For the past few years, the banking industry has been exploring the benefits of incorporating machine learning and artificial intelligence into their industry. In this regard, there have been significant advances in security, customer behavior recognition, sentiment analysis, trading recommendation systems, etc…
But after a few years, the industry has seen that the data that they have can only take them so far. Challenges like data quality, data tagging, pattern correlation vs causation, etc.… hit a wall when you only have so much data. When training an AI, engineers need more or different data that the banking entity currently has. And accessing external data sources often proves an impossible task due to security or privacy issues.
It is due to this necessity that federated learning has come into the light, since it permits different players to cooperate and share training data without compromising the privacy or security of their datasets. Not only that, but the extracted intelligence can be used for more than one industry player, encouraging cooperation between competitors in certain fields or different type of industries when their interests meet. 
Federated learning also provides systems to adequately reward those players that contribute with higher or better quality data to the model training and to reject the entities whose data is of no value to the model that is being trained.

Imagen 1.png 124 KB

If configured to do so, in the above image the model-server, implemented by a neutral entity, is capable of addressing the quality and quantity of the data that each player is bringing into the equation, and reducing the impact or, in extreme cases, flat out rejecting the data of one of the federated entities. Either by creating a new model from scratch or by re-training a model to include a new banking industry's intelligence, this approach ensures that the prediction quality will always be rewarding those who contribute the most. 
Both in supervised and unsupervised approaches, it is clear that this technique can have a great value for future endeavors in machine learning. It will be through cooperation and not competition that the most powerful and advanced AI will be created in the future, simply because this approach will ensure access to the most varied and higher quality data, and in Fintep we are embracing this change with enthusiasm. As an African proverb says: “if you want to go fast, go alone. If you want to go far, go together”