EETimes AI Everywhere - Lowering TCO of Transformers and Feature Extraction for Edge AI Solutions

Presented by Daniel Michek, Senior Manager at Altera, an Intel Company
Edge computing is experiencing a paradigm shift as artificial intelligence (AI) migrates from data centers to the network edge, closer to the data generation. Explore the role of Altera FPGAs in this transition, highlighting their AI capabilities. FPGA push-button IP generation from AI frameworks such as TensorFlow or PyTorch, together with high-performance silicon, enables the development of size, weight, and power-optimized custom platforms for AI inference, a capability referred to as ‘FPGAi’. Shifting AI algorithms from GPU-based architectures to FPGA-based edge implementations has compelling benefits; reduced latency, enhanced privacy, decreased bandwidth requirements due to local data processing, and reduced data transmitted for downstream AI processing. You’ll learn about the practical aspects of FPGA deployment, including using CNNs and Transformers for efficient feature extraction and data processing directly at the data source, real-world examples, and benchmark results.
This session was part of the AI Everywhere 2024 virtual event presented by EE Times. The full virtual conference is now available to view now on demand by visiting aieverywhere.e...