The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)

Today we kick off our annual AI Rewind series joined by friend of the show Pablo Samuel Castro, a Staff Research Software Developer at Google Brain.

Pablo joined us earlier this year for a discussion about Music & AI, and his Geometric Perspective on Reinforcement Learning, as well our RL office hours during the inaugural TWIMLfest. In today’s conversation, we explore some of the latest and greatest RL advancements coming out of the major conferences this year, broken down into a few major themes, Metrics/Representations, Understanding and Evaluating Deep Reinforcement Learning, and RL in the Real World.

This was a very fun conversation, and we encourage you to check out all the great papers and other resources available on the show notes page.

Today we close out our NeurIPS series joined by Aravind Rajeswaran, a PhD Student in machine learning and robotics at the University of Washington.

At NeurIPS, Aravind presented his paper MOReL: Model-Based Offline Reinforcement Learning. In our conversation, we explore model-based reinforcement learning, and if models are a “prerequisite” to achieve something analogous to transfer learning. We also dig into MOReL and the recent progress in offline reinforcement learning, the differences in developing MOReL models and traditional RL models, and the theoretical results they’re seeing from this research.

The complete show notes for this episode can be found at twimlai.com/go/442

As we continue our NeurIPS 2020 series, we’re joined by friend-of-the-show Charles Isbell, Dean, John P. Imlay, Jr. Chair, and professor at the Georgia Tech College of Computing.

This year Charles gave an Invited Talk at this year’s conference, You Can’t Escape Hyperparameters and Latent Variables: Machine Learning as a Software Engineering Enterprise. In our conversation, we explore the success of the Georgia Tech Online Masters program in CS, which now has over 11k students enrolled, and the importance of making the education accessible to as many people as possible. We spend quite a bit speaking about the impact machine learning is beginning to have on the world, and how we should move from thinking of ourselves as compiler hackers, and begin to see the possibilities and opportunities that have been ignored.

We also touch on the fallout from Timnit Gebru being “resignated” and the importance of having diverse voices and different perspectives “in the room,” and what the future holds for machine learning as a discipline.

The complete show notes for this episode can be found at twimlai.com/go/441. 

Today we kick off our NeurIPS 2020 series joined by Taco Cohen, a Machine Learning Researcher at Qualcomm Technologies.

In our conversation with Taco, we discuss his current research in equivariant networks and video compression using generative models, as well as his paper “Natural Graph Networks,” which explores the concept of “naturality, a generalization of equivariance” which suggests that weaker constraints will allow for a “wider class of architectures.”

We also discuss some of Taco’s recent research on neural compression and a very interesting visual demo for equivariance CNNs that Taco and the Qualcomm team released during the conference.

The complete show notes for this episode can be found at twimlai.com/go/440.

Today we close out our re:Invent series joined by Edgar Bahilo Rodriguez, Lead Data Scientist in the industrial applications division of Siemens Energy.

Edgar spoke at this year's re:Invent conference about Productionizing R Workloads, and the resurrection of R for machine learning and productionalization. In our conversation with Edgar, we explore the fundamentals of building a strong machine learning infrastructure, and how they’re breaking down applications and using mixed technologies to build models.

We also discuss their industrial applications, including wind, power production management, managing systems intent on decreasing the environmental impact of pre-existing installations, and their extensive use of time-series forecasting across these use cases.

The complete show notes can be found at twimlai.com/go/439.

Today we continue our re:Invent series with Srivathsan Canchi, Head of Engineering for the Machine Learning Platform team at Intuit. 

As we teased earlier this week, one of the major announcements coming from AWS at re:Invent was the release of the SageMaker Feature Store. To our pleasant surprise, we came to learn that our friends at Intuit are the original architects of this offering and partnered with AWS to productize it at a much broader scale. In our conversation with Srivathsan, we explore the focus areas that are supported by the Intuit machine learning platform across various teams, including QuickBooks and Mint, Turbotax, and Credit Karma,  and his thoughts on why companies should be investing in feature stores. 

We also discuss why the concept of “feature store” has seemingly exploded in the last year, and how you know when your organization is ready to deploy one. Finally, we dig into the specifics of the feature store, including the popularity of graphQL and why they chose to include it in their pipelines, the similarities (and differences) between the two versions of the store, and much more!

The complete show notes for this episode can be found at twimlai.com/go/438.

Today we’re kicking off our annual re:invent series joined by Swami Sivasubramanian, VP of Artificial Intelligence, at AWS.

During re:Invent last week, Amazon made a ton of announcements on the machine learning front, including quite a few advancements to SageMaker. In this roundup conversation, we discuss the motivation for hosting the first-ever machine learning keynote at the conference, a bunch of details surrounding tools like Pipelines for workflow management, Clarify for bias detection, and JumpStart for easy to use algorithms and notebooks, and many more.

We also discuss the emphasis placed on DevOps and MLOps tools in these announcements, and how the tools are all interconnected. Finally, we briefly touch on the announcement of the AWS feature store, but be sure to check back later this week for a more in-depth discussion on that particular release!

The complete show notes for this episode can be found at twimlai.com/go/437.

Today we’re joined by Subarna Sinha, Machine Learning Engineering Leader at 23andMe.

23andMe handles a massive amount of genomic data every year from its core ancestry business but also uses that data for disease prediction, which is the core use case we discuss in our conversation.

Subarna talks us through an initial use case of creating an evaluation of polygenic scores, and how that led them to build an ML pipeline and platform. We talk through the tools and tech stack used for the operationalization of their platform, the use of synthetic data, the internal pushback that came along with the changes that were being made, and what’s next for her team and the platform.

The complete show notes for this episode can be found at twimlai.com/go/436.

Today we’re joined by Daan Odijk, Data Science Manager at RTL.

In our conversation with Daan, we explore the RTL MLOps journey, and their need to put platform infrastructure in place for ad optimization and forecasting, personalization, and content understanding use cases. Daan walks us through some of the challenges on both the modeling and engineering sides of building the platform, as well as the inherent challenges of video applications.

Finally, we discuss the current state of their platform, and the benefits they’ve seen from having this infrastructure in place, and why using building a custom platform was worth the investment.

The complete show notes for this episode can be found at twimlai.com/go/435. 

Today we’re joined by Peter Mattson, General Chair at MLPerf, a Staff Engineer at Google, and President of MLCommons. 

In our conversation with Peter, we discuss MLCommons and MLPerf, the former an open engineering group with the goal of accelerating machine learning innovation, and the latter a set of standardized Machine Learning speed benchmarks used to measure things like model training speed, throughput speed for inference. 

We explore the target user for the MLPerf benchmarks, the need for benchmarks in the ethics, bias, fairness space, and how they’re approaching this through the "People’s Speech" datasets. We also walk through the MLCommons best practices of getting a model into production, why it's so difficult, and how MLCube can make the process easier for researchers and developers.

The complete show notes page for this episode can be found at twimlai.com/go/434.