The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Open Source at Qualcomm AI Research with Jeff Gehlhaar and Zahra Koochak - #414
Sep 30, 2020
Jeff Gehlhaar, Vice President of Technology at Qualcomm, and Zahra Koochak, Staff Machine Learning Engineer, dive into the exciting world of mobile AI. They unveil advancements in the Snapdragon 865 chipset and discuss the AI Model Efficiency Toolkit designed for edge devices. The duo explores federated learning, enhancing privacy through on-device data processing. With insights into integrating tools like TensorFlow and the MLIR framework, they spotlight collaborations shaping the future of AI efficiency and open-source innovation at Qualcomm.
AI Snips
Chapters
Transcript
Episode notes
Qualcomm's Expanded AI Engine
- Qualcomm announced the Snapdragon 865 with an expanded AI Engine, focusing on on-device AI.
- This includes broader support for Android Neural Networks and Hexagon NN Direct for direct hardware access.
Snapchat's Use of Hexagon NN Direct
- Snapchat uses Hexagon NN Direct to accelerate its filters, demonstrating the technology's real-world impact.
- This hardware acceleration enables smoother, more realistic filters than previously possible.
Developer Access to AI Acceleration
- Developers can access Qualcomm's AI acceleration through various APIs, including Android Neural Networks and Hexagon NN Direct.
- The choice depends on the specific use case and developer preference.