Data scientists get a new solution to help automate machine learning
|Richard Harris in Artificial Intelligence Thursday, September 5, 2019|
Aible announces world's first fully automated machine learning AI platform for data scientists and developers.
Foster City-based Aible has launched Aible Advanced, its new solution for data scientists and developers that automates the machine learning process from data-to-impact.
Aible Advanced empowers data scientists with a platform that advances beyond legacy semi-AutoML solutions such as DataRobot to enable data scientists to achieve business impact quickly and efficiently. Aible Advanced automates all the repetitive and boring parts of the end-to-end data science and model deployment process so data scientists and developers can focus on the parts where they add the most value.
The Only AI Platform Purpose-Built to Understand Business Objectives
In March, Aible launched the only Artificial Intelligence/Machine Learning solution for business people that understands business objectives and constraints, then crafts a custom prediction model that maximizes business impact. Forrester called it the “best choice for pure businesspeople” in The Forrester New Wave™: Automation-Focused Machine Learning Solutions Q2 2019 – The Nine Providers That Matter Most And How They Stack Up by Kjell Carlsson, Ph.D. and Mike Gualtieri et al. published 28 May, 2019.
Aible Advanced is the new complementary solution for data scientists and developers that takes what makes Aible unique and helps organizations adopt it at scale.
“Business users have the business domain knowledge; data scientists have the modeling expertise,” said Arijit Sengupta, Aible’s Founder and CEO. “Aible Advanced, in conjunction with Aible, lets each type of user seamlessly contribute their unique skills and knowledge to generate the best predictive model for their unique business reality.”
The Ten Steps of True AutoML
- Requirement Gathering whereAutoML solution learns the business objectives and business realities of the use case
- Blueprints where it helps users get started by showing them examples of good predictors for their use case
- Data Recipes where it helps gather the training data from existing enterprise applications and data repositories
- Data Enhancement where it cleans the data, prepares it for machine learning and even creates derived variables that can improve the quality of the predictive model
- Model Customization where it ensure the model is trained to maximize the actual business objectives of the use case while respecting the business constraints
- Hyperparameter Tuning where it trains various types of models and tries out many settings for each model type to create the best model
- Model Selection where it recommends the model that best optimizes the business objectives and conducts evaluations like sensitivity, what-if and scenario analysis to ensure the best model is chosen
- Model Deployment where it starts running the model in the customer’s environment
- Prediction Writeback where it writes predictions back to the enterprise applications the business users use
- Monitoring where it observes the actual business outcomes and compares it to the predicted outcomes to determine how well the predictive model is doing and whether it needs to be retrained
AutoML solutions today barely cover half of these ten steps, while Aible Advanced automatically conducts all of them. That is why Aible calls Aible Advanced the first true AutoML solution. In fact, all other AutoML solutions miss the most important steps, Requirement Gathering and Model Customization, which enable Aible Advanced to focus on delivering the unique needs of each use case, rather than focusing on maximizing simplistic measures like accuracy.
If, for instance, the benefit of successfully pursuing a sales opportunity is $100 and the cost of pursuing it and failing is $1, then you might be willing to pursue 98 sales opportunities and fail at each as long as you successfully sold one deal. Such an approach would be highly inaccurate based on simplistic accuracy measures like Log Loss, but it would create significant business benefits. Forrester independently stated in their New Wave report that , “Unique among AutoML vendors, Aible gets that a model that maximizes accuracy almost never maximizes business impact."
Aible is offering a 30-minute AutoML challenge: in less than 30 minutes, Aible will create a predictive model based on users’ proprietary data in their own AWS account and deploy the model also in their own AWS account without ever moving the data beyond the security of the user's account. If Aible fails to do that, users can keep the model free of charge. No other AutoML vendor offers such a challenge because they actually use expensive consulting to do several of the 10 steps of a true AutoML offering.
The combination of Aible Advanced for data scientists and Aible for business people allows experts to scale themselves by enforcing best practices while easily soliciting business user input to maximize business impact. Aible Advanced users can solicit feedback from Aible users or even delegate specific steps of the end-to-end process to specific business people using Aible. Business users can ensure the solution meets their business objectives and respects their business constraints. Data scientists can ensure the rigor of the end-to-end process because they can see every aspect of the model training process in Python code in notebooks.