Skip to content

Recapping the AI, Machine Learning and Data Science Meetup — Nov 2, 2023

We just wrapped up the Nov 2, 2023 , AI, Machine Learning and Data Science Meetup and if you missed it or want to revisit it, here’s a recap! In this blog post you’ll find the playback recordings, highlights from the presentations and Q&A, as well as the upcoming Meetup schedule so that you can join us at a future event. 

First, Thanks for Voting for Your Favorite Charity!

In lieu of swag, we gave Meetup attendees the opportunity to help guide Voxel51’s donations to charitable causes. The charity that received the highest number of votes this month was The Coalition for Rainforest Nation (CfRN), an organization working on the responsible stewardship of the world’s last great rainforests to achieve environmental and social sustainability.. We are sending this event’s charitable donation of $200 to CfRN on behalf of the Meetup attendees!

Missed the Meetup? No problem. Here are playbacks and talk abstracts from the event.

Foundation Models for Electronic Health Records (EHRs)

Hospitals generate an average of 50 petabytes of data per year. Unfortunately, almost none of this data is used to improve patient care, and it shows; despite spending over $4 trillion on healthcare annually, the US ranks dead last in health outcomes among high-income countries. Foundation models — large-scale AI models trained on large amounts of unlabeled data — offer a promising approach for harnessing this data to improve the efficacy of our healthcare system. In this talk, I will motivate the need for developing foundation models for electronic health records (EHRs), highlight some initial work in the space, and outline key unsolved challenges and opportunities for ML researchers hoping to make a meaningful impact with their work.

Speaker: Michael Wornow is a computer science PhD student at Stanford University advised by Nigam Shah and Chris Re. He is supported by an NSF Graduate Research Fellowship and Stanford HAI Graduate Fellowship, and his research focus is on developing and operationalizing foundation models for health systems.


  • “In regards to wearables, since data is expected to be sequential, is it treated as temporal in the EHR model? If yes, how is it processed?”
  • Can these models estimate a probability of an event that has not happened yet?
  • How effective are embeddings in preserving the diverse range of information found in EHR records? Especially considering that these records consist of not just categorical values, but also numerical and textual data?
  • You mentioned that codes within a visit can be in any order, but for training foundation models, are all the codes within a visit ordered in a certain way, before training?
  • Do these models need to be trained from scratch for each hospital record structure? Is there a way to perform transfer learning?

Resource links

Exploring the Two Headed Classifier Use Case

Let’s examine some practical applications of computer vision tasks. Although the classic classification problem may appear straightforward at first, in the real-world we’ll likely encounter numerous constraints, such as the model’s speed, size, and its ability to operate on mobile devices. Additionally, multiple tasks may need to be performed, and it may not always be advisable to employ a separate model for each task. Whenever possible, it is preferable to optimize the system’s architecture and employ fewer models, while still maintaining accuracy. Therefore, when considering all of these constraints and optimizations, the task suddenly becomes more complex. In this talk we’ll work through an example classification problem with several classes that may not appear visually similar. We’ll see how a two-headed model can assist us in this challenge.

Speaker: Argo Saakyan is a computer vision engineer with 5+ years of experience in data science. Argo has deployed numerous models to production and worked with classification/detection/segmentation tasks in both research and optimized deployment sides.


  • “It seems as though the approach is to separate the two tasks by separating them in space. We take our image that was mapped to a 1d classifying vector, and instead feed it to 2 separate heads. Alternatively, I could imagine having an output tensor with 2 dimensions, 1 dimension for each task. In this sense the separation is from orthogonal axes in the output tensor, as opposed to heads in the architecture. Was an approach like this considered, and if so, how did it compare?”

Resource links

Using AI to Test Software, Techniques and Tools

AI has innovated the way lot of people work. One of these ways is software testing. From unit testing, integration testing, and end to end testing, AI can help developers test better throughout the testing pyramid. During this session we will dive into a variety of methods and tools developers can use AI to help them test better.

Speaker: Justin Trugman is the founder of, an AI powered developer tool focusing on helping developers improve their test coverage. Before, Justin was an early engineer and the VP of Software Development at the startup Caregility, leading the engineering teams developing their revolutionary telehealth solution for a customer base of over 1000 hospitals around the globe. Justin was also previously at Loon, an Alphabet subsidiary and writes for his newsletter, Bug Driven Development.


  • “What are the main issues with having AI doing the testing as opposed to having the developer conduct the testing?”

Join the Computer Vision Meetup!

Computer Vision Meetup membership has grown to over 6,000 members in just one year! The goal of the Meetups is to bring together communities of data scientists, machine learning engineers, and open source enthusiasts who want to share and expand their knowledge of computer vision and complementary technologies.

What’s Next?

Up next on Nov 15 at 9 AM Pacific we have “FiftyOne Plugins Workshop: Authoring Data-Centric AI Applications”

Register for the Zoom here. You can find a complete schedule of upcoming Meetups and workshops on the Voxel51 Events page.

Get Involved!

There are a lot of ways to get involved in the Computer Vision Meetups. Reach out if you identify with any of these:

  • You’d like to speak at an upcoming Meetup
  • You have a physical meeting space in one of the Meetup locations and would like to make it available for a Meetup
  • You’d like to co-organize a Meetup
  • You’d like to co-sponsor a Meetup

Reach out to Meetup co-organizer Jimmy Guerrero on or ping me over LinkedIn to discuss how to get you plugged in.

The Computer Vision Meetup network is sponsored by Voxel51, the company behind the open source FiftyOne computer vision toolset. FiftyOne enables data science teams to improve the performance of their computer vision models by helping them curate high quality datasets, evaluate models, find mistakes, visualize embeddings, and get to production faster. It’s easy to get started, in just a few minutes.