Reinventing AI Research & Development: Part VIII
In This Article
This piece covers the work accomplished across two rotations from late 2021 to Q1 2022. From engaging in a new project selection process to producing podcast episodes featuring our past rotation projects, we have continued to grow and transform AI R&D at World Wide Technology.
This year kicked off with a new method of selecting AI R&D projects. The R&D group hosted a brainstorming session, led by David Hall, in January of 2022, which produced many creative ideas – five of which were further developed and presented to the project selection panel by David Hall and Essak Seddiq. These ideas ranged from identifying deep fakes to using data for sales forecasting, with the next targeted project being a data analytics use case around improving energy consumption. This project selection process was a success in that it generated momentum in kick-starting an idea into a project while also adding promising projects to a backlog.
Data Analytics for Smart Energy Consumption
After a groundswell of support for this ESG project after our Project Selection process, the AI R&D rotation began a project examining electricity consumption at WWT's data centers. Data centers are typically large consumers of energy, their energy usage is rarely optimized, and data center energy growth could become exponential.
The R&D project team used a hypothesis-based approach to define heuristics and relevant data sets to identify dynamic throttling and optimize peak load as the two primary levers available to reduce energy consumption.
Figure 1. The current data center utilization in Figure 1 highlights the need to consolidate the unused half of our equipment. Future work will be to get WWT to a place where we can handle peak load while using the minimum amount of equipment, all without sacrificing performance.
Figure 2. Identifying the direct correlation between chilled inflow water temperature and fan speed illustrates how decreasing water temperature saves electricity by dynamically throttling fan speed when possible.
The authors proposed a general framework, debias-GAN, to address possible bias in AI and Machine Learning (ML) algorithms by explicitly augmenting a training dataset for NLP models with underrepresented instances synthesized by a pretrained sequence generating model.
The authors trained a generator model to synthesize ethnicity oblivious tweets through a GAN via policy gradient. This is a solid example of utilizing AI to fix biased input data that correlates conversational tweets and user ethnicity. Likewise, the same approach can also be applied in real life tasks to combat biases in AI, especially those that result from biased training data.
WWT's Data Science team and Supply Chain Process Improvement team worked together on a Smart Receiving solution that leverages computer vision technologies in aiding operators in the material validation process at the receipt.
Figure 3. A diagram depicting the process proposed by the authors using OCR and a matching algorithm to automate the workflow
During this past R&D rotation, the MLOps tiger team produced two high-quality pieces centered around helping organizations and leaders understand MLOps.
Finding the right cloud provider to support your MLOps goals can be challenging. This article walks through the different tools currently available from Google Cloud, AWS and Azure.
Curious about MLOps and how the field is evolving? Read the article above for the latest developments that we have observed from working with our partners.
We had several guest speakers on our bi-weekly forums, including members of Soniox, the world's first self-learning AI tool for automatic speech recognition, as well as Plainsight, a platform for turn-key vision AI.
Looking forward, we are excited to continue to strengthen and expand on our AI/ML related work to create new value for future clients. The publication team will continue to publish more exciting articles and papers, so be sure to follow our Data Analytics and AI topic area to hear the latest from the team.