November 5, 2020
A new paper was presented by Gábor Lóki (Department of Software Engineering) at this year’s ICCE-Asia conference. The paper, titled “Memory-aware Performance Optimization of Tensor Programs for Embedded Devices” was written in collaboration with our partners at Samsung (Samsung Research UK & Samsung Electronics (Korea)).
The conference, the Fifth International Conference On Consumer Electronics (ICCE) was held both in-person (in Busan, Republic of Korea) and online, with livestreams accommodating those who could not participate in the sessions because of the pandemic.
The paper tackles issues of memory consumption in resource constrained embedded devices. The techniques are based on the Tensor Virtual Machine, but with the added improvements it results in better performance both in terms of memory footprint and inference time.
You can read more about the conference on its official website.