Five ideas for accelerating your data science and AI projects in a post-pandemic world

By | 5 minute read | November 12, 2020

In the age of COVID-19, business as usual just isn’t an option. Workforce resiliency, enterprise agility and cost reduction have all become top of mind for business leaders. Digital transformation through AI can play a key role in bolstering a COVID business recovery plan. From reallocating resources to increasing operational efficiency, the case for digital transformation and AI is growing stronger. According to “Covid-19 and the future of business” by IBM Institute of Business, 66 percent of surveyed organizations say they have completed digital transformation initiatives that previously met resistance. If the pandemic has shifted priorities, it has also made AI-powered digital transformation more justifiable.

But how do you make sure AI is trusted? What’s needed to manage model performance despite invalidated patterns caused by the pandemic? What can you do to fill data science skills gaps and realize value from your AI investments? IBM Watson® Studio on IBM Cloud Pak® for Data automates how your business can build, run and manage AI models at scale. It can reduce costs by simplifying AI lifecycle management and speed innovation with a flexible multicloud architecture and open source tools.

To discuss five ideas for accelerating your data science projects and how the latest version of IBM Watson Studio can support innovative and cost-effective digital transformation, I sat down with Thomas Schaeck, Distinguished Engineer, IBM Data and AI.

Idea 1: Bring together diverse skill sets for cross-team collaboration 

Developing AI can take a village. Data science and AI teams have a lot to learn from application development and DevOps teams in operationalizing the process from ideation to results at an accelerated pace. Cross-collaboration is key. “By harnessing the power of talent with cross-discipline backgrounds,” says Schaeck, “organizations can accelerate learning, increase productivity and bring models to production faster on a unified environment.”

To speed AI development, data scientists and developers can use AutoAI, a model development tool within Watson Studio, to automate data preparation, feature engineering and hyperparameter optimization. The latest launch increases the size of data sets and enables the multiple input data sets to generate model pipelines and train candidate models. Schaeck says, “We added SDK support for AutoAI (available as a tech preview) to generate Python Notebooks or Scripts in addition to the current user interface.” This works from Python notebooks or scripts running directly in Projects on IBM Cloud Pak for Data, as well as from notebooks or Python scripts running in desktop IDEs like VS Code and PyCharm and connecting remotely to IBM Cloud Pak for Data APIs. With Anaconda Repository for IBM Cloud Pak for Data, teams can also accelerate open source innovation and policy control with enterprise-grade package management.

Learn more: Explore operating models in our analyst webinar DevOps for AI.”

Idea 2: Increase cost savings and efficiencies by enabling machine learning without locational constraints

Many locational factors that determine predictions and optimizations have changed as work arrangements and consumption patterns have shifted during COVID. AI can help reduce costs and meet regulations by dynamically reallocating resources and automating logistical operations across both physical locations and multiple clouds.

Federated learning, available as a tech preview in IBM Cloud Pak for Data, helps increase training accuracy by using data across clouds while meeting data privacy, security and other regulations. Schaeck outlines how you can train an algorithm at the edge without moving data. Use cases include:

  • Fraud detection through improved access to training data for cross-border and multiple parties
  • Medical diagnostics and drug discovery by sharing patient data, research data and other ecosystem data as training data
  • Smart manufacturing to use data in robots and other devices across factories for model training

Discover industry leaders: Download The Forrester Wave™: Multimodal Predictive Analytics and Machine Learning, Q3 2020

Idea 3: Expand your deep learning projects on an integrated data and AI platform

Deep learning is now more widely adopted than ever. Inspired by the human brain, deep learning uses artificial neural networks to develop advanced pattern recognition and improve model training and inference. It crunches through large and complex data, including unstructured data from speech, audio, video, or handwritten notes.

A data and AI platform that scales compute, people and apps dynamically can enable faster deep learning implementation. It provides a multitenant, multicloud architecture that helps automate AI lifecycles and speed time to results.

IBM Cloud Pak for Data breaks down data siloes and optimizes deployment without disruptions to AI training. Furthermore, says Schaeck, with recent advancements in deep learning on Watson Machine Learning Accelerator, a capability in Watson Studio, organizations can choose more mature deep learning patterns, saving costs and speeding digital transformation with transparency.

Dig deeper: Check out this infographic, ‘Accelerate deep learning workloads on IBM Cloud Pak for Data’

Idea 4: Predict and optimize your business with broader contributors

Prediction and optimization models have been indispensable in forecasting volatile demands for services, goods and other purchases. For example, AI can help calibrate the opening of specific market segments and optimize staffing, logistics and other resources to support them.

A unified data and AI platform can save time and costs by combining prediction and optimization on a single hybrid multicloud architecture. It can also empower non-coders to accelerate model development with visual data science tool SPSS Modeler.

IBM Cloud Pak for Data offers decision optimization and visual data science as part of IBM Watson Studio Premium. The platform can rapidly transform any participants into contributors, diversifying their skills while increasing productivity. With version 3.5, the interface becomes more user-friendly for operational researchers. SPSS Modeler adds extension nodes, new charts and streamlined integration to accelerate data preparation and model development. 

Learn more: Join our live 3-part data science webinar series.

Idea 5: Measure the ROI of model monitoring and increased production 

“Investments in explainable AI are prerequisites for scaling AI with trust and transparency,” says Schaeck. By describing an AI model, its expected impact and potential biases, explainable AI capabilities can help organizations manage regulatory risks and, crucially during COVID, link model accuracy with monetary value. As one data scientist in financial services noted in a commissioned Forrester study1, with explainable AI, “our models are now more accurate, which means we can better forecast our required cash reserve requirements. A 1% improvement in accuracy frees up millions of dollars for us to lend or invest.”

IBM Cloud Pak for Data makes model monitoring more interactive with shareable “what if” analysis and improved understandability​ with role-based access and visibility. By unifying data and AI services, the platform helps end users better understand model decisions and detect potential fairness issues due to unseen data correlations.

Measure outcomes: Read New Technology: The Projected Total Economic Impact™ Of Explainable AI And Model Monitoring In IBM Cloud Pak For Data

As the pandemic impacts the business landscape, companies that innovate with AI-powered digital transformation stand to weather the storm. You can be one of them.

1 New Technology: The Projected Total Economic Impact™ Of Explainable AI And Model Monitoring In IBM Cloud Pak For Data, Commissioned By IBM, August 2020

Most Popular Articles