Optimal resource usage in ultra-low-power sensor interfaces through context- and resource-cost-aware machine learning

Abstract This paper introduces an approach that combines machine learning and adaptive hardware to improve the efficiency of ultra-low-power sensor interfaces. Adaptive feature extraction circuits are assisted by hardware embedded training to dynamically activate only the most relevant features. This selection is done in a context- and power cost-aware manner, through modification of the C4.5 algorithm. As proof-of-principle, a Voice Activity Detector illustrates the context-dependent relevance of features, demonstrating average circuit power savings of 70%, without accuracy loss. The RECAS database developed for experimenting with this context- and dynamic resource-cost-aware training is presented and made open-source for the research community.

[1]  Refet Firat Yazicioglu,et al.  A Configurable and Low-Power Mixed Signal SoC for Portable ECG Monitoring Applications , 2011, IEEE Transactions on Biomedical Circuits and Systems.

[2]  Nobutaka Ito,et al.  The Diverse Environments Multi-channel Acoustic Noise Database (DEMAND): A database of multichannel environmental noise recordings , 2013 .

[3]  Rich Caruana,et al.  Do Deep Nets Really Need to be Deep? , 2013, NIPS.

[4]  Yi Hu,et al.  Subjective comparison and evaluation of speech enhancement algorithms , 2007, Speech Commun..

[5]  Anantha Chandrakasan,et al.  An 8-channel scalable EEG acquisition SoC with fully integrated patient-specific seizure classification and recording processor , 2012, 2012 IEEE International Solid-State Circuits Conference.

[6]  T.Y. Lin,et al.  Anomaly detection , 1994, Proceedings New Security Paradigms Workshop.

[7]  Andreas Krause,et al.  SenSay: a context-aware mobile phone , 2003, Seventh IEEE International Symposium on Wearable Computers, 2003. Proceedings..

[8]  Qiang Yang,et al.  Decision trees with minimal costs , 2004, ICML.

[9]  Kun Li,et al.  User-Centric Indoor Air Quality Monitoring on Mobile Devices , 2013, AI Mag..

[10]  Juan Manuel Górriz,et al.  Speech/non-speech discrimination based on contextual information integrated bispectrum LRT , 2006, IEEE Signal Processing Letters.

[11]  Marian Verhelst,et al.  Ultra-low-power voice-activity-detector through context- and resource-cost-aware feature selection in decision trees , 2014, 2014 IEEE International Workshop on Machine Learning for Signal Processing (MLSP).

[12]  Felix Naumann,et al.  Data fusion , 2009, CSUR.

[13]  Mario Konijnenburg,et al.  A voltage-scalable biomedical signal processor running ECG using 13pJ/cycle at 1MHz and 0.4V , 2011, 2011 IEEE International Solid-State Circuits Conference.

[14]  George Kurian,et al.  Self-aware computing in the Angstrom processor , 2012, DAC Design Automation Conference 2012.

[15]  VARUN CHANDOLA,et al.  Anomaly detection: A survey , 2009, CSUR.

[16]  Marian Verhelst,et al.  24.2 Context-aware hierarchical information-sensing in a 6μW 90nm CMOS voice activity detector , 2015, 2015 IEEE International Solid-State Circuits Conference - (ISSCC) Digest of Technical Papers.

[17]  Matt J. Kusner,et al.  Cost-Sensitive Tree of Classifiers , 2012, ICML.

[18]  Robert Puers,et al.  The BladderPill: An in-body system logging bladder pressure , 2010 .

[19]  James Tschanz,et al.  A 2.3nJ/frame Voice Activity Detector based audio front-end for context-aware System-on-Chip applications in 32nm CMOS , 2012, Proceedings of the IEEE 2012 Custom Integrated Circuits Conference.

[20]  Guang-Zhong Yang,et al.  Body sensor networks , 2006 .

[21]  Marian Verhelst,et al.  A reconfigurable, 0.13µm CMOS 110pJ/pulse, fully integrated IR-UWB receiver for communication and sub-cm ranging , 2009, 2009 IEEE International Solid-State Circuits Conference - Digest of Technical Papers.

[22]  James Tschanz,et al.  A 2.3 nJ/Frame Voice Activity Detector-Based Audio Front-End for Context-Aware System-On-Chip Applications in 32-nm CMOS , 2013, IEEE Journal of Solid-State Circuits.

[23]  Marian Verhelst,et al.  Context- and cost-aware feature selection in ultra-low-power sensor interfaces , 2014, ESANN.

[24]  Wolfgang Rosenstiel,et al.  Generic Self-Adaptation to Reduce Design Effort for System-on-Chip , 2009, 2009 Third IEEE International Conference on Self-Adaptive and Self-Organizing Systems.

[25]  Chao Li,et al.  SolarCore: Solar energy driven multi-core architecture power management , 2011, 2011 IEEE 17th International Symposium on High Performance Computer Architecture.

[26]  Mustafa Cenk Gursoy,et al.  Energy-aware and robust task (re)assignment in embedded smart camera networks , 2013, 2013 10th IEEE International Conference on Advanced Video and Signal Based Surveillance.

[27]  Yanqing Zhang,et al.  Body Sensor Networks: A Holistic Approach From Silicon to Users , 2012, Proceedings of the IEEE.

[28]  Refet Firat Yazicioglu,et al.  18.3 A multi-parameter signal-acquisition SoC for connected personal health applications , 2014, 2014 IEEE International Solid-State Circuits Conference Digest of Technical Papers (ISSCC).

[29]  Ian H. Witten,et al.  One-Class Classification by Combining Density and Class Probability Estimation , 2008, ECML/PKDD.