4Ds for Traders & Investors - Deep Learning

In our last post, we briefly introduced Deep Pipes, the Smart Data Fusion component of our next generation 4D architecture for AI-driven trading & investment.

Smart Data right from the start

Data strategies which just focus on the fusion of plenty of (external) data silos into a single enterprise whatever data lake, store, or warehouse get stuck at the data level, and never reach the information or knowledge level.

Replicating huge amounts of data every day from one location to another with some basic transformations in between neither creates information nor knowledge. We are convinced that we must make data smart right from the start.

That’s why we built Deep Pipes to bring the full spectrum of machine intelligence to the first mile of each data journey.

Deep learning must be re-envisioned

Deep learning has its permanent place in the processing of complex non-linear data. Non-traditional or alternative data in finance is just that. This is obvious. Why write a post about deep learning?

In this post, we do not discuss innovative language models like GPT-3 from OpenAI, or its open-source cousin GPT-J, and its application in finance. Although there are interesting use cases for modern natural language processing, we want to point to a more fundamental problem.

AI models and applications can be brilliant but are suffering from a critical disadvantage: They are artisanal handcrafted products, and this kind of “production” process is a severe bottleneck for volatile environments.

And even after an AI model is trained successfully, a plethora of different technologies and libraries prevent a predict as you train experience in business. Have you ever tried to change a model configuration, while “its” inference application is live?

When we talk about artificial intelligence, we always take a business boots on the ground perspective and annoy our audience with the same question over and over again:

What must be done to take AI to an industrialized technology?

This is crystal clear: Developing brilliant algorithms or models which can never be adopted by a broad range of smaller and midsize companies is the wrong action to answer this question.

Declarative AI

We certainly do not claim that we found the holy grail: But we are convinced that turning AI into a declarative technology (as was done with business rule processing and database queries a long time ago) is a step into the right direction.

That’s we built Declarative AI, a declarative Deep Learning component on top of Intel’s Analytics Zoo as one of the building blocks of our 4D architecture for traders and investors. It perfectly fits in the Apache Spark ecosystem, and complements the finance domain with code-free deep learning.

Whether it is earth observation data brought to Apache Spark with RasterFrames or global broadcast, print or online events & news with GDELTFrames:

They all have one thing in common: They can be connected to Declarative AI with ease - either on the first or the last mile of their data journey.?

Just a reminder at the end of this post: Declarative AI is just one the D's in our 4D architecture for traders & investors.

Stay tuned and next, we briefly sketch, our Deep Graph component.