Last week, Sama exhibited and presented at the venerable Embedded Vision Summit in Santa Clara, California.
Embedded Vision Alliance has hosted EBS since 2012, bringing together technology providers across industries who are enabling innovative and practical applications for computer vision.
This year for the first time, I represented Sama at the talks and presented a technical talk about Training Data Strategy: Avoiding Bias and Legal and Ethical Sourcing Considerations. Take a look at my related blog post for more details.
AI at the Edge
On the exhibition floor, the hottest topic was AI at the edge and edge computing. AI at the edge brings AI into small, powerful computer chips located directly in devices, such as phones.
With the computing happening in the device itself, AI devices no longer need to send data back to the cloud. AI at the edge enables remote and distributed applications, such as using a cell phone to diagnose crop diseases on rural farms.
For Sama, AI at the edge means new, flexible applications for computer vision that might require new training data strategies and solutions.
High quality training data annotation continues to be a high priority need for companies developing AI and computer vision. The challenge remains: high quality data at a reasonable cost and timeline.
Blending highly trained workforces with annotation automation and auto-labeling can make significant gains in efficiency. Sama’s expert, experienced workforce combined with our technology continues to be an offering that is well-positioned to help address these problems with real-world data.
At the Sama booth, we had many great conversations about data capture. Attendees who stopped by our booth also asked questions about semantic segmentation and smarter video interpolation.
Though computer vision has advanced in many ways, data capture remains a challenge – particularly as algorithms become more sophisticated and require bespoke data sourcing.
That’s one of the many reasons why Sama plans to expand our data capture offering.
Towards the end of the two-day event, I presented on training data strategy in the Technical Insights II Track.
In my presentation, I discussed approaches to training data strategy for avoiding data bias and considering legal and ethical sourcing issues. Data can be biased if it does not represent reality accurately: it is missing examples of use cases, or doesn’t have enough examples of use cases (even if a few are present).
Biased data can result in poor algorithm performance if an algorithm simply doesn’t work as designed. Even worse, biased data can cause problems if say, a facial recognition algorithm identifies the wrong person as a criminal, or an autonomous vehicle algorithm fails to detect the presence of a cyclist.
I also discussed legal and ethical sourcing considerations: knowing regional laws and considering privacy and other gray areas when sourcing data. We’ll update this blog post to link to the video of my presentation when EBS posts it. Also be sure to check out my related blog post for more details.
2019 Women in Vision Reception
We were invited to attend the first Women in Vision Reception ever held at EBS. It was inspiring to hear the exciting work in which other women are involved. We hope to attend future receptions!
As always, Sama continuously seeks to be ahead of the training data market. We are always researching the latest technologies, applications and business cases to be able to offer the most comprehensive training data solutions and strategies.
Our time at the Embedded Vision Summit 2019 gave us an inside look into the computer vision industries’ current needs and challenges. We can’t wait to see what’s next!