

Generative AI at the Edge with Vinesh Sukumar - #623
Apr 3, 2023
Vinesh Sukumar, senior director at Qualcomm Technologies, dives into the future of AI at the edge. He discusses the unique AI needs for mobile and automotive platforms, emphasizing the shift towards text-based inputs and generative content. Vinesh highlights the challenges and innovations surrounding ML Ops and synthetic data use. The conversation includes insights on advanced models like GPT-4 and strategies for optimizing performance in edge computing. His expertise sheds light on the rapid advancements and exciting opportunities in the evolving AI landscape.
AI Snips
Chapters
Transcript
Episode notes
Evolving AI Use Cases
- AI use cases are evolving beyond image and video processing to text, linguistics, and commerce.
- This shift impacts hardware architectures, moving from convolution-heavy designs to support transformers and generative content.
Hardware Optimization Challenges
- Optimizing hardware for specific architectures like CNNs can improve efficiency.
- However, shifts in dominant architectures require adapting optimization strategies to maintain performance across new and old models.
System Design for AI
- Consider key performance indicators (KPIs) like latency, performance, and power efficiency when designing AI systems.
- Optimize hardware for data types, compute resources, memory, and bandwidth to meet these KPIs.