Anastasia AI has standardized on SigOpt, an Intel Company, to manage experimentation and hyperparameter optimization as part of its AI platform. SigOpt empowers Anastasia to efficiently build AI models for its mid-market customers, ranging across a wide variety of modeling use cases, including an innovative approach to time-series predictions.
What’s New
Anastasia AI is an artificial intelligence (AI) company that develops world-class AI applications for customers that cut across many industries. Anastasia has become an expert in building end-to-end machine learning platforms that allow orchestration of the different functional areas of a customer’s business and make these model development processes more efficient, scalable, and reproducible. As part of its machine learning platform, Anastasia has integrated SigOpt as the best-in-class solution for experimentation. Across all customer engagements, Anastasia relies on SigOpt to design experiments, explore model behavior, and optimize model hyperparameters. Working in conjunction with SigOpt, Anastasia’s AI team builds models that solve customer business problems with state of the art results.
One example of this type of AI application is time-series forecasting, a complex machine learning problem that touches many industries. By applying SigOpt, Anastasia has developed a time-series forecasting platform capable of producing world-class results and customizing these models for each customer. Customers like Arauco use the Anastasia time-series forecasting platform to increase the accuracy of their demand predictions for ~113,000 products by 15%.
“Time series forecasting is a worthwhile application for most businesses, but it might be too expensive in terms of time, personnel, and resources for some mid-sized enterprises to build their capability. Anastasia AI bolstered our in-house AI expertise with SigOpt to build a world-class time-series prediction platform to fulfill this need. We will continue applying SigOpt to optimize and personalize our platform for each customer. As a result, any company can gain access to time-series forecasting in an affordable and efficient way.”
–Pablo Zegers, Ph.D., Vice President of Product, Anastasia AI
Why it Matters
This innovative approach holds three essential lessons for efficient AI model development, the real-world impact of models on business outcomes and the democratization of AI applications.
First, Anastasia couples AI researchers with SigOpt’s best-in-class experimentation software to bolster the model development process. The combination of domain expertise and software automation makes model development more efficient and scalable. SigOpt’s sample-efficient approach to hyperparameter optimization makes it possible to customize these state-of-the-art models for specific customers and, in the case of time-series forecasting, even for specific products at SKU level. These minor adjustments allow the model to deliver high-quality predictions that reflect each customer’s circumstances at whatever scale is needed. As a result, the AI performs better while offering a more significant impact on the business.
Second, this type of centralized service is a critical path to democratizing access to world-class AI. There are plenty of companies who could benefit from AI, but for whom AI is not – and should not be – a core competency. Anastasia’s mission is to become the AI partner for these companies to profit from it while continuing to focus on their core operational strengths.
Third, implementing AI in the correct business processes offers measurable impacts on a wide range of industries. In time-series forecasting, Anastasia has worked with numerous industries to drive measurable impact in bottom-line outcomes. As mentioned, Anastasia caused a 15% boost in ~113,000 SKUs product demand prediction accuracy for Arauco. These direct benefits are often accompanied by indirect benefits, such as productivity gains that boost the team’s capacity.
How it Works
SigOpt runs in any cloud and is designed to be entirely agnostic to modeling framework, task, library, or problem. SigOpt’s API-enabled software can be integrated into any machine learning platform or specific modeling notebook with a simple pip install and a few lines of code. As Anastasia iterates through training and tuning runs, SigOpt automatically logs metadata on these runs back to a web dashboard. When Anastasia launches a hyperparameter optimization job, SigOpt manages this process by using either its proprietary optimizer or any third-party optimizer instead. As a result, Anastasia can streamline the workflow, compare architectures and scale their model development processes.
“Anastasia AI team relies on SigOpt to design experiments, explore model behavior and optimize model hyperparameters, which, in the process, gives us novel insights on model performance. By combining our expertise with SigOpt, we were able to ask the right questions during experimentation and develop world-class models. At the same time, we can also rely on SigOpt to automatically select the best performing configuration of hyperparameters to individualize our model for each problem.”
– Pablo Zegers, Vice President of Product, Anastasia AI
Anastasia AI then converts these AI models into business applications capable of addressing a real-world problem for their customers. Making this possible often requires additional software, platform, and machine learning engineering to ensure the application is correctly deployed and that it is delivering results with business impact. “Most companies need AI but cannot implement it. Anastasia’s mission is to make it possible for all companies to benefit from AI in this new world,” said Zegers.
Take Action
If you have a business challenge that you would like to tackle with AI, reach out to Anastasia AI at anastasia.ai/contact. If you want to apply SigOpt to your next modeling project, you can access it for free at sigopt.com/signup.