Beschleunigung von Convolutional Neural Networks mit programmierbarer Logik - Einzelansicht

- Introduction to Convolutional Neural Networks, which are widely used for image classification and object detection

- Implementing CNN inference on a resource constrained CPU and analyzing bottlenecks

- Introduction to state of the art dedicated ML accelerators in form of student presentations

- Usage of different design styles and abstraction models

- Project in simulation and synthesis, co-designing an accelerator for parts of the workload and integrating it using the Pynq Z2 SoC.

- Evaluation of key performance metrics and power consumption of different implementations

- The course will consist of a weekly lecture and an accompanying lab

- The lecture will teach the required techniques for the design and evaluation for both the software and hardware implementation.

- The lab consists of two tasks and a project, which will require students to put the lecture content into practice

- The students will work in small groups, presenting and discussing their results during the lecture once after each task and once after the project.

- A weekly tutoring hour will be offered for students to discuss problems (if hybrid a virtual tutoring session will also be offered)