Does AI Model Lifecycle Management matter?
Artificial intelligence (AI) is becoming ubiquitous in many areas, from the edge to the enterprise. So, how do you use AI? Do you just feed data to a predictor? The answer is “no.”
In fact, during the infusion of AI, we need to collect data, train the data, build a model, deploy the model, and run the predictor. The pipeline to using AI is longer than one might expect, given that there are several elements (seen in Figure 1 as described in the Google article “MLOps: Continuous delivery and automation pipelines in machine learning”):
In enterprise, the critical role of AI requires a well-defined and robust methodology and platform, and a business may even fail if its methodology and platform are not up to par. For example, if fraud detection makes bad decisions, a business will be negatively affected. In the long pipeline for AI, response time, quality, fairness, explainability, and other elements must be managed as part of the whole lifecycle. It is impossible to manage them individually.
Therefore, what we call “AI Model Lifecycle Management” manages the complicated AI pipeline and helps ensure the necessary results in enterprise. We will detail AI Model Lifecycle Management in a series of blog entries. In addition, we will show how the IBM Cloud Pak® for Data can help AI Model Lifecycle Management.
We expect these blog entries to be of interest to the following people:
- Data science and AI leaders: To better understand how to increase returns on data science and AI investments.
- Data scientists: To better appreciate how data science activities can leverage/integrate with DevOps tools/processes, and to more deeply understand IBM’s strategy for end-to-end AI Model Lifecycle Management.
- DevOps engineers: To better understand the AI development process, its associated complexities, and how it can integrate with DevOps.
What is AI Model Lifecycle Management?
Let us think about what is necessary for AI Model Lifecycle Management. The first requirement is a set of components for the whole pipeline. The document “The AI Ladder – Demystifying AI Challenges” explains how to introduce AI into enterprise and clearly outlines four steps in the pipeline:
- Collect: Make data simple and accessible.
- Organize: Create a business-ready analytics foundation.
- Analyze: Build and scale AI with trust and transparency.
- Infuse: Operationalize AI throughout a business.
Figure 2 shows these four steps with their relationships and operational examples:
Another requirement is data governance of the whole pipeline. Quality is essential in enterprise, and explainability and fairness are growing increasingly essential. During the whole pipelining, data governance for AI Model Lifecycle Management should monitor and give feedback regarding quality, fairness, and explainability.
How tools help AI Model Lifecycle Management
As we have seen, AI Model Lifecycle Management is not easy. It is impossible to do it manually. Therefore, the necessary tools should have the following features to effectively support AI Model Lifecycle Management in a cloud:
- Ease of model training and deployment
- Model deployment and training at scale
- Monitoring data governance, quality, and compliance
- Visualization of the whole pipeline
- Rich connectors to data sources
One example of these tools is the IBM Cloud Pak for Data. IBM Cloud Pak for Data is a multicloud data and AI platform with end-to-end tools for enterprise-grade AI Model Lifecycle Management, ModelOps. It helps organizations improve their overall throughput of data science activities and achieve faster time to value from their AI initiatives. The Cloud Pak for Data includes the following key capabilities:
- Model development and training tools, including AutoAI and no-code, drag and drop capabilities, and support for a rich set of commonly used open source libraries and frameworks.
- Model deployment tools to scale deployed models in production for modern apps and meet performance requirements.
- Model monitoring and management tools to deliver trusted AI.
- Data virtualization capabilities to significantly increase the AI throughput of data science teams by helping data scientists efficiently access the broad set of data sources of an enterprise across a hybrid multicloud environment, without having to copy data.
- DataOps to meet data governance, quality, and compliance requirements.
- Complete data services, with a rich set of data connectors and scalable multicloud data integration capabilities to enable efficient extract, transform, and load (ETL) operations from a variety of data sources.
Stay tuned for the next blog entries
Future blog entries will detail the following phases in AI Model Lifecycle Management:
- AI Model Lifecycle Management: Collect Phase
- AI Model Lifecycle Management: Organize Phase
- AI Model Lifecycle Management: Build Phase
- AI Model Lifecycle Management: Deploy Phase
- AI Model Lifecycle Management: Monitor Phase (Technical Perspective)
- AI Model Lifecycle Management: Monitor Phase (Customer Perspective)
This post will be updated with links as they become available. For a deeper dive into the subject, see our white paper.