Skip to main content Skip to Footer

LATEST THINKING


Delivering improved insights with automated analytics

Accenture Labs’ model management framework simplifies analytics platform deployment at scale.

IDENTIFYING THE PROBLEM

Amid growth of artificial intelligence and the Internet of Things, businesses are building analytics platforms to derive fresh insights and improve performance.

Analytical models are at the heart of these platforms, applying machine learning techniques on data at massive scale and in real-time.

As these platforms gain wider adoption, we see a greater need to manage and test a wide range of models requiring different administration skills and other variables.

Read on for more on how Accenture Labs developed a model management framework that automates, simplifies and accelerates the model life cycle management.

DOWNLOAD THE FULL REPORT [PDF]

Analytical models must be trained, compared and monitored before deploying into production, requiring many steps to take place in order to operationalize a model’s lifecycle.

DIFFERENT APPROACHES

Even though a real-time data analytics system can deliver significant benefits, it still presents issues in the life cycle management and modeling process. There is no easy way to monitor, retrain and redeploy the models.

In general, data scientists collect the data they are interested in, prepare and stage the data, apply different machine learning techniques to find a best-of-class model, and continually tweak the parameters of the algorithm to refine the outcomes. Automating and operationalizing this process is difficult. Additional challenges with manual modeling include a lack of:

  • Resource expertise - Analytical modeling in a real-time, streaming architecture requires individuals with expertise in both data science and engineering.

  • Runtime abstraction - The need for skills in systems engineering for analytical modeling underscores the lack of run-time abstraction. Data scientists are generally comfortable with machine learning and statistical computing programming languages. However, they may not be trained in systems engineering to troubleshoot and operate a run-time environment.

  • Model support - When the number of model types and runtime environments increases, an abstraction for the environments becomes critical. Run-time environments cannot support all types of models.

  • Central repository - Currently, there is no standard method for comparing, sharing or viewing models created by other data scientists, which results in siloed analytics work.

OUR WAY FORWARD

As business relies more heavily on real-time big data analytics in the digital age, the ability to manage, train, deploy and share models that turn analytics into action-oriented outcomes is essential.

Accenture Labs has developed a flexible and extensible model management framework to help simplify the interfaces for analytical modeling at scale on a Lambda architecture.

In this framework, data scientists can more easily train and deploy analytical models in various run-time environments.

The framework abstracts the data engineering throughout the life cycle management and modeling process, reduces complexity of training and deployment, and shares models in a consumable way.

We apply advanced machine learning technology to provide a set of comprehensive abstractions for analytical modeling—one that is both easier to use and quicker to produce results.

The framework is unique from existing tools because of its life cycle management of models and its agnostic approach to run-time environments.

It can be extended to support additional environments, increasing the number of models that can be leveraged in the data pipeline.

Using this framework, an organization can create an ecosystem for harvesting analytical models, providing an application store-like experience for data scientists and business analysts to discover the best models and promote them for use.

SUGGESTED CONTENT