Embedding smarter decisions deep in the operational IT fabric
Deep learning is, without doubt, the hottest topic in machine learning, if not the hottest topic on the planet. This article assumes, you already know what deep learning is and you have decided that you want to use it – as part of your business process or to build smarter applications powered by “artificial intelligence”.
Training complex, deep neural network algorithms on big data is one challenge. What if we (i.e., our super-smart data scientists) have already mastered that task and built a model with one of the many deep learning tools available today? We now have a deep learning model that is able to deliver valuable results, if we could only apply it easily to new data where and when we need to.
Deep Deployment in action:
Real-time, on-demand execution of deep neural network recognizing objects in images taken with smartphone camera
This is where we encounter the main barrier to a broader adoption of deep learning:
Before we can reap in the benefits of our smart, deep learning algorithms, we must be able to deploy, integrate and execute deep learning models as part of operational IT systems. In other words, we need to make deep learning models more accessible to enterprise IT.
It requires a standards-based process that is applicable across (1) different deep learning tools as well as (2) different IT solutions, from cloud-based hosting to streaming analytics, from real-time scoring of single transactions to massively parallel batch processing on big data platforms like Hive, Spark and Storm.
Deep Deployment: Embedding deep learning models in business applications
To overcome this barrier to adoption, Zementis set out to deliver a solution for flexible, cross-platform deployment and execution of deep learning models. This is what we call “Deep Deployment”.
Simply said, our “deep deployment” process allows users to embed deep learning models tightly & deeply in their day-to-day business processes and operational IT solutions. On one hand, this reduces cost and complexity by leveraging the existing investments in IT infrastructure, on the other hand, it also enables developers to quickly build new composite applications that are powered by deep learning algorithms.
Zementis’ deep deployment is based on the Predictive Model Markup Language (PMML) industry standard and delivers the following key benefits:/h3>
- Multi-tool support: Build models in different deep learning tools
- Cross platform: Deploy models across different IT solutions
- Batch and real-time: Execute against data at rest or in streaming analytics
- Rapid time-to-insight: Deploy models in minutes, not months
- Reduced complexity: Minimize custom code, leverage existing infrastructure
As with all machine learning and predictive algorithms, deep learning models only become valuable, once they deliver better, smarter, more precise decisions.
Not ready, don’t have models, yet? Download our free white paper to learn how Standards-based Deployment of Predictive Analytics will make your data science process more effective.