In This White Paper

As previous AI R&D rotations have laid a solid groundwork for subsequent developments, the R&D program is currently in a steady state running smoothly. The last rotation fully concluded the University Machine Learning (ML) project and the MRI Radiomics on-rotation project, updated the project selection process with a new scoring matrix to help debias selections. With all the progress made by the last operations team, over the past few months we have focused on maintaining and maturing the program's processes, as well as identifying opportunities for increased innovation.

This article spotlights our primary objectives from the past few months, including:

  • Developing a new project on utilizing AI to detect and mitigate biases by training a debias-GAN model using a large Twitter dataset.
  • Publishing several new articles and white papers onto the platform.
  • Introducing a guest speaker series with Reid Blackman, a former philosophy professor currently focused on ethics in AI, and Sriram Rajaraman, a Principal Engineer focused on building products for Google Search Ads.
  • Creating future plans as we move into Q2 2021.

Project focus during our time on the AI R&D Ops rotation

Detecting bias in AI

The application of AI (artificial intelligence) in sensitive areas, including recruitment, medical referral and diagnosis, and criminal justice, has stirred a debate on the algorithmic bias in AI against certain social groups. While bias in AI has several contributing factors, the bias in data lies at the heart of the issue.

For the latest AI R&D on-rotation project, we dug into biased data, one of the central issues at the heart of AI work, and explored the potential of utilizing AI techniques to detect and mitigate this issue. We leveraged a large Twitter dataset to train a debias-GAN model to generate tweets. With the generated and actual tweets, we detected biases in the text and researched methods to reduce biases and eventually balance the dataset. 

Throughout this project, we aimed to educate ourselves on common sources of biases, identify AI approaches to addressing biased data, and ultimately bring more attention to the issue. Figure 1 below visualizes the debias-GAN model, one core component of the project.

Debiasing tweet generation model using adversarial training with policy gradient
Figure 1: Debiasing tweet generation model using adversarial training with policy gradient

The current data science on-rotation team recently has wrapped up our work in this space, so stay tuned for our white paper that details the research methodology and findings.

New publications

The previously formalized peer review process has been busier than ever. We have reviewers with knowledge in a wide variety of areas including business strategy, software engineering, ML infrastructure engineering, data science and human-centered design. The following papers and articles were all recently published to the WWT platform.

R&D articles

The majority of mines in the world have been in operation for more than a century and, therefore, have lagged in adopting benefits from the Artificial Intelligence (AI) industry. This article shows how AI can help such a capital-intensive industry.

MLOps articles

During this past R&D rotation, the previously established MLOps tiger team has been producing various high-quality content focusing on helping organizations and leaders understand MLOps. These articles range from helping leaders identify whether their organizations are ready to take on MLOps to help realize their business goals, to explaining the core benefits that MLOps can provide.

Customize a strategy to build scalable machine learning in your organization and accelerate the realization of your business goals.

The promise of a data science revolution felt like it was right there, within your grasp. You recruited a smart team, established sophisticated governance and constructed a common repository for your data. Even with all these investments, you still find it hard to put models into production and continuously update and monitor them.

Adopting MLOps platforms gives organizations the tools they need to manage the machine learning (ML) lifecycle.

Whether artificial intelligence (AI) can deliver business value to organizations, regardless of the size or industry, is no longer a doubt. Now, the pressing question is: how can organizations most efficiently integrate AI into their processes to deliver that value?

White papers

This paper leverages deep learning to match MRI tumor images to scans with similar features with the goal of providing clinicians a tool to increase accuracy and speed to diagnosis.

Deploying the Kubeflow MLOps platform in AWS to enable our Data Science team to create end-to-end ML workflows for automated delivery of machine-learning models.

Guest speakers

The R&D program invited our first guest speakers during this rotation with Reid Blackman and Sriram Rajaraman, to talk to our group about ethics and AI at large organizations.

Reid is a former philosophy professor who taught, researched and published articles with a concentration in ethics and free will. During his time as a professor, he started focusing on the social impacts of AI and realized this was his calling. He then went on to fulfill this mission in 2018, when he moved on from academia and founded Virtue Consultants, a consulting firm focused on driving ethics and ethical risk mitigation into company culture and the development, deployment and procurement of emerging technology products.  He is also a Senior Advisor to E&Y and sits on their AI Advisory Board.

Sriram is a Principal Engineer at Google and has spent the best part of the past decade focusing on building products for Google Search Ads. While not a data scientist himself, he has been working side-by-side with data scientists building and improving his products. 

It was wonderful learning about the work revolving AI outside of WWT and we hope to continue the speaker series in the future.

What's next?

With the on-rotation project on bias in AI wrapped up, we're excited to announce that WWT's project selection panel has chosen AIOps as the topic for next rotation's project. The AIOps on-rotation project will focus on developing a generalizable end-to-end machine learning framework that can address AIOps in a cohesive manner. In addition, there will be two ongoing projects starting within the next R&D rotation, which we will unveil once the team has kicked off.

We hope to continue strengthening our collaboration efforts to help identify additional clients interested in AI/ML-related work. 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.