October 23, 2020
Categorized: Artificial Intelligence
Share this post:
For most organisations, making real-time decisions can be challenging, primarily because the required data resides in silos; on-premise systems, public clouds, and private clouds. As a result, when the pandemic hit us, most organisations weren’t prepared to dynamically respond to changes in their customers’ preferences and circumstances. For instance, many grocery stores in Australia and New Zealand struggled to optimise offers and personalise product catalogues in their online platforms to meet demand and supply imbalance during the pandemic’s early stages.
Real-time decision making using all the data at hand – not just historical data – has been a key enabler of companies whose business models rely on recommendation models; businesses like Google, Amazon, Netflix, and Spotify. For these internet giants and born-in-the-cloud companies, gaining real-time insights from data using AI is second nature.
No matter how advanced the organisations’ capabilities are, the unexpected changes in our daily circumstances can quickly outdate the models we have built however. Now more than ever, we need to ask how we can develop and deploy Artificial Intelligence (AI) models that fare better when circumstances change rapidly.
With the recent advances in AI (e.g. online learning, reinforcement learning) and the advent of Hybrid Cloud, large enterprises and governments don’t need to leave real-time decision making on the back burner anymore. It’s for that reason Ambiata and IBM got together and conducted a Proof of Concept (PoC) to determine the benefits of running Ambiata’s continuous intelligence system, Atmosphere, on IBM OpenPOWER systems (AC922).
Why do Ambiata’s Atmosphere and IBM OpenPOWER systems work well together?
Ambiata’s Atmosphere software is a real-time decision-making engine that learns continuously from historic and online data to support decision making based on contextual information. When deployed on Public Clouds, Atmosphere requires all the necessary historic data to be moved to the cloud. For many organisations with most of their historic and operational data on-prem in distributed systems, this move is costly and challenging.
The AI in Atmosphere is a combination of deep neural networks (DNN) and complex statistical models for the uncertainty that benefit from a GPU and CPU mixed architecture like that available in the IBM Power System AC922. Given the in situ extraction need and IBM’s presence, Ambiata and IBM embarked on this proof of concept to explore the benefits of running Atmosphere in a hybrid deployment model – where the real-time data extraction from enterprise data would take place on-prem and the online element could run on either Public Cloud or on-prem.
In this PoC we compared the performance for the above real-time decision-making setup using simulated data on an IBM Power System AC922 against a baseline (AWS P3.2xlarge). The results from the PoC demonstrated that with Large Model Support enabled, the POWER architecture could efficiently support large models that exceed the GPU’s raw capacity leading to a significant reduction in the execution time compared to the baseline. You can find more details about the PoC and the results we obtained in the detailed write up here.
Reinforcement Learning (RL) based Contextual Bandits systems such as Atmosphere can be the personalised recommendation engines within an organization’s AI software suite. These continuous intelligence services can personalise the user experience of websites and applications, personalise offers, route customer complaints, optimise sales funnels, and recommend news articles – a well-known example is the Netflix app, where almost every aspect is personalised on the fly.
Given the certainty of uncertainty in the post-COVID-19 world, if your organisation is finding it challenging to make a decision under changing circumstances, Hybrid Cloud solutions such as Atmosphere and IBM Power can guide you with real-time decision making under uncertainty. Enterprises should also consider activating their data in-situ where possible – take the compute and AI to the data using accelerated hardware such as IBM Power Systems.
You can contact one of the authors of this blog for further information to support your data activation journey.
Stephen Hardy, CEO, Ambiata
Min Kim, Senior Data Scientist, Ambiata
Nirandika Wanigasekara, Data Scientist, IBM Systems
Wijay Wijayakumaran, Chief Architect, AI, IBM Systems