f-CNNx: Deploying Multiple CNNs in Complex AI Systems

Add to your list(s) Download to your calendar using vCal

• Stelios I. Venieris (Imperial College London)
• Monday 25 June 2018, 16:00-17:00
• Level 9 Seminar Room, EEE Dept..

If you have a question about this talk, please contact George A Constantinides.

In the construction of complex AI systems, deep neural network models are used as building blocks of a larger application. Nevertheless, deploying multiple models on a target platform poses a number of challenges. With each model trained for a different subtask of the system, the workload size and performance constraints vary accordingly. Moreover, the different models compete for the same pool of resources and hence resource allocation between models becomes a critical factor. In this talk, we will present f-CNNx, a toolflow whose goal is to automate the mapping of multiple convolutional neural networks (CNNs) on a target FPGA platform while meeting the require performance for each model. To generate an optimised multi-CNN design, f-CNNx introduces a highly-parametrised multi-CNN hardware architecture and a tunable memory access policy in order to explore a wide range of resource and bandwidth allocations. Furthermore, the toolflow incorporates the application-level importance of each model by means of multiobjective cost function in order to generate a hardware design that meets the target application’s requirements. Overall, f-CNNx overcomes the limitations of competing platforms by achieving up to 6.8x gains in performance-per-Watt over highly optimised embedded GPU designs in multi-CNN settings.

This talk is part of the CAS Talks series.

This talk is included in these lists:

• Andrea Picciau's list
• CAS Talks
• Circuits and Systems Group: Internal Seminars
• Level 9 Seminar Room, EEE Dept.

Note that ex-directory lists are not shown.