Publications by authors named "Meriem Bettayeb"
Article Synopsis
- - The text discusses the rising use of transformer networks in AI but highlights issues with their computational complexity, particularly due to the self-attention mechanism, which limits performance similar to CNNs.
- - An innovative hardware accelerator using memristor-based in-memory computing is introduced to address memory bottlenecks in matrix-matrix multiplication (MatMul) operations, leading to a tenfold decrease in multiply-accumulate operations while maintaining high accuracy on the MNIST dataset.
- - Simulation results show efficient resource usage and performance metrics (e.g., latency of 15.52 seconds, 3 mJ energy consumption), indicating progress toward a transformer design optimized for edge devices capable of real-time processing.
