Memory-aware dynamic spatially adaptive inference

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

• Giorgos Zampokas
• Monday 11 March 2024, 13:00-14:00
• Department of Electrical and Electronic Engineering, Room 909B.

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

In dynamic systems, the ability to adapt to varying computational costs is useful, since computational resources can fluctuate in real-time, therefore operating across a spectrum of FLOP budgets is an important advantage for deep learning models. Moreover, with real systems constrained by finite memory capacities, it becomes imperative to account for this limitation in model design. This talk aims to explore the multifaceted challenges inherent in balancing performance against FLO Ps and the number of parameters. Finally, leveraging the effectiveness of spatially adaptive methods in achieving favorable performance to computational cost trade-offs, a multi-target training scheme is proposed to achieve dynamic inference with negligible parameter number increase.

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
• Department of Electrical and Electronic Engineering, Room 909B

Note that ex-directory lists are not shown.