Christine Long

Christine Long

Software Engineering Manager (Machine Learning Engineering Team in Reality Labs) at Meta

Christine Long is a highly experienced engineering manager with a strong background in machine learning and deep learning. She currently serves as the Engineering Manager of EMG Engineering & Research Team in Reality Lab at Meta, where she leads a team of talented engineers working on building cutting-edge on-device AI models for wearable devices. The team has released multiple bigger, faster, and more power/compute-efficient models that can coexist within the same device, with the flexibility to adapt to various hardware systems. Prior to joining Meta, Christine held leadership roles at other public companies such as ICE. With over 10 years of experience in the field, Christine has developed a deep expertise in machine learning, deep learning and software development. She has worked on various projects involving LLM, SaaS and Embedded Systems, and has a proven track record of delivering high-quality products and solutions. She is known for her technical expertise, leadership skills, and ability to inspire and motivate her teams to achieve their best work.

All Sessions by Christine Long

Day 3 10/31/2024
2:00 pm - 2:30 pm

Wearable AI in Meta: On Device ML with Neural Interface System

<span class="etn-schedule-location"> <span class="firstfocus">Deep Learning</span> <span class="firstfocus">All Levels</span> </span>

Meta developed an EMG (electromyography) wristband integrated with smart glasses at Meta Reality Labs. This talk will walk through various challenges that we encountered when developing each generation of this wristband, what is the optimal solution that we eventually implemented to bring this product to market, and what lesson we learned that can be applied to future on-device machine learning development. One of the key challenges with implementing deep neural network models on a device like the wristband lies in its limited computational resources, hardware, power and thermal constraints compared to more powerful computing platforms like laptops or servers. This necessitates the use of lightweight yet powerful models that can perform efficiently within these constraints while still delivering acceptable accuracy levels for various applications such as gesture recognition or activity detection. To address these challenges, we applied preexisting techniques but also invented a few novel techniques to optimize machine learning models specifically for on-device deployment. Additionally, a close collaboration between hardware engineers, software developers, and researchers worked together to optimize the end-to-end system for power efficiency and performance, ensuring seamless integration of on-device machine learning capabilities into the wearable products. Moreover, developers need to consider the impact of on-device training and adaptation due to varying user profiles and environmental conditions. Multiple techniques for personalization and privacy have been compared via carefully designed benchmarking systems. Together with the consideration of hardware limitation, the best approach has been employed to improve model performance across diverse user populations while adhering to strict data privacy requirements. Another crucial aspect involves devising efficient data collection and preprocessing strategies tailored for the resource-constrained nature of wearable devices like this wristband. This involves leveraging smart data sampling techniques, compressive sensing, and customized signal processing algorithms to minimize the amount of data needed for accurate model training and inference. I will also talk about challenges when we integrate this wristband with other Meta wearable devices, and how we resolved those challenges.

Day 3 10/31/2024
2:00 pm - 2:30 pm

Wearable AI in Meta: On Device ML with Neural Interface System

<span class="etn-schedule-location"> <span class="firstfocus">Deep Learning</span> <span class="firstfocus">All Levels</span> </span>

Meta developed an EMG (electromyography) wristband integrated with smart glasses at Meta Reality Labs. This talk will walk through various challenges that we encountered when developing each generation of this wristband, what is the optimal solution that we eventually implemented to bring this product to market, and what lesson we learned that can be applied to future on-device machine learning development. One of the key challenges with implementing deep neural network models on a device like the wristband lies in its limited computational resources, hardware, power and thermal constraints compared to more powerful computing platforms like laptops or servers. This necessitates the use of lightweight yet powerful models that can perform efficiently within these constraints while still delivering acceptable accuracy levels for various applications such as gesture recognition or activity detection. To address these challenges, we applied preexisting techniques but also invented a few novel techniques to optimize machine learning models specifically for on-device deployment. Additionally, a close collaboration between hardware engineers, software developers, and researchers worked together to optimize the end-to-end system for power efficiency and performance, ensuring seamless integration of on-device machine learning capabilities into the wearable products. Moreover, developers need to consider the impact of on-device training and adaptation due to varying user profiles and environmental conditions. Multiple techniques for personalization and privacy have been compared via carefully designed benchmarking systems. Together with the consideration of hardware limitation, the best approach has been employed to improve model performance across diverse user populations while adhering to strict data privacy requirements. Another crucial aspect involves devising efficient data collection and preprocessing strategies tailored for the resource-constrained nature of wearable devices like this wristband. This involves leveraging smart data sampling techniques, compressive sensing, and customized signal processing algorithms to minimize the amount of data needed for accurate model training and inference. I will also talk about challenges when we integrate this wristband with other Meta wearable devices, and how we resolved those challenges.

Open Data Science

 

 

 

Open Data Science
One Broadway
Cambridge, MA 02142
info@odsc.com

Privacy Settings
We use cookies to enhance your experience while using our website. If you are using our Services via a browser you can restrict, block or remove cookies through your web browser settings. We also use content and scripts from third parties that may use tracking technologies. You can selectively provide your consent below to allow such third party embeds. For complete information about the cookies we use, data we collect and how we process them, please check our Privacy Policy
Youtube
Consent to display content from - Youtube
Vimeo
Consent to display content from - Vimeo
Google Maps
Consent to display content from - Google