Financial solutions

Save up 60% of Cloud Computing costs!

Extracting value from the available data is becoming the key differentiator in the financial market. Systems using Artificial Intelligence (AI) can analyse much more data than humans, and therefore they are expected to react quicker and make better investment decisions based on more information.

The start-up AI Investments  uses AI in a self-improving platform for optimizing an investment portfolio in a complete trading solution for the global markets, including a diversified way of signalling transactions, determining market conditions, and managing exposure. Instead of predicting the prices of portfolio assets, the solution offers “diversified high probability transaction notifications” for exposure, risk, and position sizing management by applying deep CNN, Long Short-Term Memory networks, as well as an reinforcement learning algorithm based on Monte Carlo Tree Search with value network.

AI based investments portfolio optimization

The diagram above presents the application structure:

  1. There is a control component for supervising the training: This component is usually deployed in an active-passive configuration. It orchestrates the work of the whole system and executes the investment strategies.
  2. A communication and data bus based on Kafka is used for the whole solution. A set of transformation pipeline components. There are parallel pipelines, one for each broker, where the data is fetched, transformed, trained, and then the trained models are stored.
  3. Predictions components: These components are used to make a one-time prediction using a given model.

Requirements

Hence, the application is fully distributed, and both the preprocessing data step and the training step could be done in a distributed manner. The computing requirements for this application are:

find out more »

Business scenario

For the AI Investments application it is assumed that on average 10 models are trained in parallel, but at peak level up to 50 models could be trained, and the average number of models can reach 18 models during 120 hours per month.

find out more »

Results

There are many possible Virtual Machine (VM) types or flavours provided by the various Cloud Providers. Properly choosing the most optimal flavour for specific task requires additional calculation. This action is done automatically by MELODIC.

find out more »

More MELODIC users success stories

See how CE-Traffic uses MELODIC to process mobility data at scale.

Bitnami