%0 Conference Proceedings
%T Low-Overhead Early-Stopping Policies for Efficient Random Forests Inference on Microcontrollers
%+ Politecnico di Torino = Polytechnic of Turin (Polito)
%+ Alma Mater Studiorum Università di Bologna = University of Bologna (UNIBO)
%+ Eidgenössische Technische Hochschule - Swiss Federal Institute of Technology  [Zürich] (ETH Zürich)
%A Daghero, Francesco
%A Burrello, Alessio
%A Xie, Chen
%A Benini, Luca
%A Calimera, Andrea
%A Macii, Enrico
%A Poncino, Massimo
%A Jahier Pagliari, Daniele
%< avec comité de lecture
%@ 978-3-031-16817-8
%( IFIP Advances in Information and Communication Technology
%B 29th IFIP/IEEE International Conference on Very Large Scale Integration - System on a Chip (VLSI-SoC)
%C Singapore, Singapore
%Y Victor Grimblatt
%Y Chip Hong Chang
%Y Ricardo Reis
%Y Anupam Chattopadhyay
%Y Andrea Calimera
%I Springer Nature Switzerland
%3 VLSI-SoC: Technology Advancement on SoC Design
%V AICT-661
%P 25-47
%8 2021-10-04
%D 2021
%R 10.1007/978-3-031-16818-5_2
%K Machine learning
%K TinyML
%K Adaptive inference
%K Dynamic inference
%K Energy-efficiency
%K Random forests
%K Microcontrollers
%Z Computer Science [cs]Conference papers
%X Random Forests (RFs) are popular Machine Learning models for edge computing, due to their lightweight nature and high accuracy on several common tasks. Large RFs however, still have significant energy costs, a serious concern for battery-operated ultra-low-power devices. Following the adaptive (or dynamic) inference paradigm, we introduce a hardware-friendly early stopping policy for RF-based classifiers, halting the execution as soon as a sufficient prediction confidence is achieved. We benchmark our approach on three state-of-the-art datasets relative to different embedded classification tasks, and deploy our models on a single core RISC-V microcontroller. We achieve an energy reduction ranging from 18% to more than 91%, with an accuracy drop lower than 0.5%. Additionally, we compare our approach with other early-stopping policies, showing that we outperform them.
%G English
%Z TC 10
%Z WG 10.5
%2 https://inria.hal.science/hal-04419563/document
%2 https://inria.hal.science/hal-04419563/file/538103_1_En_2_Chapter.pdf
%L hal-04419563
%U https://inria.hal.science/hal-04419563
%~ IFIP
%~ IFIP-AICT
%~ IFIP-TC
%~ IFIP-WG
%~ IFIP-VLSISOC
%~ IFIP-TC10
%~ IFIP-WG10-5
%~ IFIP-AICT-661