On-device deep neural network (DNN) training holds the potential to enable a rich set of privacy-aware and infrastructure-independent personalized mobile applications. However, despite advancements in mobile hardware, locally training a complex DNN is still a nontrivial task given its resource demands. In this work, we show that the limited memory resources on mobile devices are the main constraint and propose Sage as a framework for efficiently optimizing memory resources for on-device DNN training. Specifically, Sage configures a flexible computation graph for DNN gradient evaluation and reduces the memory footprint of the graph using operator- and graph-level optimizations. In run-time, Sage employs a hybrid of gradient checkpointing and micro-batching techniques to dynamically adjust its memory use to the available system memory budget. Using implementation on off-the-shelf smartphones, we show that Sage enables local training of complex DNN models by reducing memory use by more than 20-fold compared to a baseline approach. We also show that Sage successfully adapts to run-time memory budget variations, and evaluate its energy consumption to show Sage's practical applicability.

The recent breakthroughs in machine learning (ML) and deep learning (DL) have catalyzed the design and development of various intelligent systems over wide application domains. While most existing machine learning models require large memory and computing power, efforts have been made to deploy some models on resource-constrained devices as well. A majority of the early application systems focused on exploiting the inference capabilities of ML and DL models, where data captured from different mobile and embedded sensing components are processed through these models for application goals such as classification and segmentation. More recently, the concept of exploiting the mobile and embedded computing resources for ML/DL model training has gained attention, as such capabilities allow (i) the training of models via local data without the need to share data over wireless links, thus enabling privacy-preserving computation by design, (ii) model personalization and environment adaptation, and (iii) deployment of accurate models in remote and hardly accessible locations without stable internet connectivity. This work summarizes and analyzes state-of-the-art systems research that allows such on-device model training capabilities and provides a survey of on-device training from a systems perspective.

Bluetooth Low Energy (BLE) and the iBeacons have recently gained large interest for enabling various proximity-based application services. Given the ubiquitously deployed nature of Bluetooth devices including mobile smartphones, using BLE and iBeacon technologies seemed to be a promising future to come. This work started off with the belief that this was true: iBeacons could provide us with the accuracy in proximity and distance estimation to enable and simplify the development of many previously difficult applications. However, our empirical studies with three different iBeacon devices from various vendors and two types of smartphone platforms prove that this is not the case. Signal strength readings vary significantly over different iBeacon vendors, mobile platforms, environmental or deployment factors, and usage scenarios. This variability in signal strength naturally complicates the process of extracting an accurate location/proximity estimation in real environments. Our lessons on the limitations of iBeacon technique lead us to design a simple class attendance checking application by performing a simple form of geometric adjustments to compensate for the natural variations in beacon signal strength readings. We believe that the negative observations made in this work can provide future researchers with a reference on how well of a performance to expect from iBeacon devices as they enter their system design phases.

This article presents LiteZKP a framework for supporting multiple anonymous payments using a smart contract-based zero-knowledge proof (ZKP) protocol on resource-limited devices. Specifically, to address challenges related to minimizing the computational overhead and offer a fully anonymous system, LiteZKP includes novel schemes such a new merkle tree mechanism to reduce the burden of ZKP operations, and integrates smart contract-based ZKP with an off-chain payment channel to reduce the amount of ZKP operations when performing continuous data exchange. We present evaluation results of LiteZKP from both PC-scale and resource-limited client devices and our results suggest that LiteZKP reduces the latency and energy consumption by more than 55% on Internet of Things (IoT)/mobile edge computing platforms, while requiring only 8% of block processing fee compared to a naive ZKP-based scheme.