AI for AI: IBM debuts AutoAI in Watson Studio
The democratization of AI is driving a significant acceleration in demand for machine learning and data science skills across all industries. To further accelerate this trend and bridge the skills gap, creation of AI itself is being automated and is transforming the way businesses and data science teams operate. According to McKinsey Global Institute, nearly half of work tasks could soon be automated in the global workforce. For data scientists, this means a reduction in manual tasks and an acceleration of time to deployment.
“Almost 85% [of company executives] believe AI will allow their companies to obtain or sustain a competitive advantage.” – MITSloan Management Review
AI is fast moving beyond solving only specific tasks and dependence on manually-crafted features such as laborious feature-engineering steps. Now, AI learning is adaptive, using automatically-constructed features and neural network architectures that can optimize the model for your use case, delivering time-to-value and unprecedented value for enterprises. AI for AI is here.
AI for AI: The revolution
Today’s machine learning models are rapidly becoming highly complex, involving labor-intensive data preparation and feature engineering. As a result, enterprises are quickly deploying sophisticated neural network architectures with tens of millions of parameters. Consistent breakthroughs from researchers produce new machine learning methods and new architectures for neural networks designed to solve unique problems.
Faced with these complex challenges, your team’s process for getting the most from AI involves designing, optimizing, and governing models.
Design. You need highly skilled data scientists to design optimized neural networks, a time-consuming and always evolving task. Data science engineers used to get by leveraging CNNs or RNNs, but now they need an arsenal of architectures to deliver state of the art performance. Enterprises are hungry to exploit this explosion of innovation while leveraging learnings of their data science teams—which experiments worked, which failed and the characteristics of each.
Optimize. When working with structured data, your team faces challenges around finding optimal machine learning pipelines. These pipelines include data preparation, feature engineering, hyper parameter optimization (HPO) and ensembling. This can take your data science team weeks or even months, requiring manual experimentation at each stage. The tasks of feature engineering and hyper-parameter optimization (HPO) are especially challenging for inexperienced data scientists who lack the expertise to ascertain best transformations for feature extraction needed to improve model performance.
Govern. The final challenge is governance of AI models which is fast becoming a requisite for successfully deploying AI for businesses. Governing AI includes monitoring all aspects of AI in deployment, including performance, anomalies, decision fairness and explainability. If any bias is detected, de-biasing algorithms can correct and ensure that enterprises have a trust in their AI. Using advanced monitoring and corrective algorithms and techniques for explainability, governing AI models are critical in opening up black box AI models that enterprises have been hesitant in deploying.
AI for AI makes it possible to automate the end-to-end data science and AI process, allowing your business to take the next steps in complementing human-led expertise and innovation with machine-generated insights.
IBM is pioneering the development of AI. Instead of developing a standalone AutoML tool, we chose to add the feature into IBM Watson Studio, our multi-modal data science platform. We productized IBM Research’s unparalleled innovation in automating AI lifecycle management as a new capability: AutoAI.
AutoAI in IBM Watson Studio makes it possible for you to:
- Automate your AI lifecycle management
- Enable one-click deployment with Watson Machine Learning
- Build better models faster and go live using the skill sets you have
- Scale experimentation and deployment processes
- Monitor and de-bias AI outcomes with Watson OpenScale
- Increase trust and transparency in AI/ML development
AI designing AI
NeuNetS in IBM Watson Studio synthesizes a neural network and trains it on your unstructured training data, like images or text, without you having to design or write code by hand.
- Reach model accuracy within minutes/hours instead of days/weeks
- Scale your AI training with our cluster of GPUs
- Ensure consistency and repeatability
AI optimizing AI
AutoAI automatically prepares data, applies algorithms, and attempts to build model pipelines best suited for your data and use case.
- Citizen data scientists learn the process of optimizing AI pipelines
- Experienced data scientists reach production level accuracy more quickly
- Deploy machine learning pipelines, including all data transformations, with a single click using Watson Machine Learning
AI governing AI
- Automate and operationalize AI lifecycle in business applications
- Ensure AI models are unbiased, understandable and explainable
- Allow auditing of models for business transactions
The road ahead: democratize and industrialize AI
IBM is focused on simplifying the complex, time-intensive and costly AI development process, making it easily accessible for all. By relieving data scientists of manual planning, management, and monitoring tasks, we hope to empower them with more time dedicated to developing powerful, predictive models. Teams assisted by AutoAI can extract greater value from AI investments and invent new business models by focusing on the differentiating parts of their business.
Collaborate with us in our open source community initiatives as well as co-development and partnership initiatives with IBM clients and partners.