Alan Ferrari

Reducing your local footprint with anyrun computing

Computational offloading is the standard approach to running computationally intensive tasks on resource-limited smart devices, while reducing the local footprint, i.e., the local resource consumption. The natural candidate for computational offloading is the cloud, but recent results point out the hidden costs of cloud reliance in terms of latency and energy. Strategies that rely on local computing power have been proposed that enable fine-grained energy-aware code offloading from a mobile device to a nearby piece of infrastructure. Even state-of-the-art cloud-free solutions are centralized and suffer from a lack of flexibility, because computational offloading is tied to the presence of a specific piece of computing infrastructure. We propose AnyRun Computing?(ARC), a system to dynamically select the most adequate piece of local computing infrastructure. With ARC, code can run anywhere and be offloaded not only to nearby dedicated devices, as in existing approaches, but also to peer devices. We present a detailed system description and a thorough evaluation of ARC?under a wide variety of conditions. We show that ARC?matches the performance of the state-of-the-art solution (MAUI), in reducing the local footprint with stationary network topology conditions and outperforms it by up to one order of magnitude under more realistic topological conditions.

Detecting energy leaks in Android app with POEM

This paper presents the design and implementation of a Portable Open Source Energy Monitor (POEM) to enable developers to automatically test and measure the energy consumption of every single application component down to the control flow level. Based on existing portable power meter designs, POEM extends the state of the art of application analysxis with the energy annotation of the control flow down to the basic blocks, the call graph, and the Android API calls, allowing developers to locate energy leaks in their applications with high accuracy. Because the power consumption is tied to the system status, energy annotation is also coupled with system activities.

Using barometric pressure data to recognize vertical displacement activities on smartphones

A method for detecting major vertical displacements in human activities is proposed.Prediction is based on barometers available on smartphones and inference models.Decision trees are a good choice for their high accuracy and low energy consumption.Barometers offer high accuracy, energy efficiency and independence from position. We introduce a novel, efficient methodology for the automatic recognition of major vertical displacements in human activities. It is based exclusively on barometric pressure measured by sensors commonly available on smartphones and tablets. We evaluate various algorithms to distinguish dynamic activities, identifying four different categories: standing/walking on the same floor, climbing stairs, riding an elevator and riding a cable-car. Activities are classified using standard deviation and slope of barometric pressure. We leverage three different inference models to predict the action performed by a user, namely: Bayesian networks, decision trees, and recurrent neural networks. We find that the best results are achieved with a recurrent neural network (reaching an overall error rate of less than 1%). We also show that decision tree classifiers can achieve good accuracy and offer a better trade-off between computational overhead and energy consumption; therefore, they are good candidates for smartphone implementations. As a proof of concept, we integrate the decision tree classifier in an App that infers user activity and measures elevation differences. Test results with various users show an average recognition accuracy rate of about 95%. We further show the power consumption of running barometric pressure measurements and analyse the correlation of pressure with environmental factors. Finally, we compare our approach to other standard methodologies for activity detection based on accelerometer and/or on GPS data. Our results show that our technique achieves similar accuracy while offering superior energy efficiency, independence from the sensor location, and immunity to environmental factors (e.g., weather conditions, air handlers).

Characterization of the impact of resource availability on opportunistic computing

With opportunistic computing, devices are no longer restricted to using their own services and resources, but can access services and resources made available by other devices. The performance of opportunistic computing is greatly affected by the resource topology in the network: what resources/services are available, as well as when and where they can be tapped. This paper presents a preliminary investigation of the impact of the resource availability on the performance of opportunistic computing. Specifically, we propose a metric called Expected Resource Availability, ERA, that attempts to capture the impact of the topology of services and resources. The ERA offers a proxy for the applicability of opportunistic computing schemes to a given network: if the ERA is low, any opportunistic scheme can be expected to fail due to a sheer lack of resources and/or connectivity among them. On the other hand, if the ERA is high, success can be expected. To gain perspective on the properties of the ERA, we tackle the problem of service allocation in opportunistic computing, which suffers to combinatorial explosion when looking for the optimal solution. We also present some preliminary simulation results that confirm the validity of the ERA as a metric to gauge whether opportunistic computing can be achieved in a given network.

Gesture-based soft authentication

This paper presents a novel authentication strategy for Bluetooth-equipped smartwatches. We use the built-in smartwatch sensors to detect whether two users have shaken hands. If this is the case each devices give to the other a soft authentication privilege, which is suitable for applications with relaxed security needs (for instance, an application to exchange business cards). We evaluate the system using different machine learning techniques and we investigate their performance in the dimensions of accuracy and energy consumption.

Managing your privacy in mobile applications with MockingBird

Mocking is a standard technique in software testing; its main goal is to mimic the real object behavior in a controllable way. Recently, mocking techniques have been used in mobile environments to increase the user privacy and their goal is to allow users to select the kind of information they want to pass to the application (if real or randomly generated). This work presents MockingBird, a novel solution to mocking that uses recorded context-traces instead of randomly generated data, which is easily detected by applications. We also propose a flexible methodology to mock an Android application that does not require any changes at the operating system level and at the virtual machine level. MockingBird is a very promising solution; we are currently testing its performance and increasing its functionality.

Code offloading on opportunistic computing

Although mobile smart devices are becoming more and more resourceful, they cannot compete with higher-end devices in terms of computational capabilities. Therefore, it is generally advantageous to offload computationally intensive tasks. While cloud-based offloading is popular, it has a non-negligible impact on the energy consumption of mobile devices. Our solution is a novel approach based on locally opportunistic code offloading that leverages the local availability of higher-end devices. We aim to demonstrate the benefits of our approach with respect to a widely adopted benchmark: face recognition.

Code mobility for on-demand computational offloading

Computational offloading is a traditional approach to augment the capabilities of smart devices. It has been shown that offloading is most efficient when carried out within a Local Area Network. Several frameworks have been proposed to tackle this problem but fail to consider that some of the application code may be part of the offloading request. This is a key element in cloudlets where it is virtually impossible to load all the necessary application code a priori on all nodes. It is even more crucial in opportunistic computing where it cannot be known a priori which devices will become the offloading targets.

Poster: Can Smart Devices Protect Us from Violent Crime?

To explore the applicability of mobile smart devices to personal protection against violent crime, we propose a system that can detect the onset of hazardous situations involving violent crime by leveraging standard activity recognition strategies on smartphones and sensory inputs from wearable devices, as well as send help requests to alert the authorities.

On the usage of smart devices to augment the user interaction with multimedia applications

Wearable devices have recently gained a foothold in the market with the uptake of smartwatches. The strong tie between a smartwatch and its owner, the highly predictable position of a smartwatch on the body, and its internal sensors are enabling a wide array of applications that leverage the user context. In this paper we focus on a gesture recognition system to augment the user interaction with multimedia applications. We define a set of seven gestures that are relevant across several applications and we collect an extensive dataset with two smartwatches (the Motorola Moto360 and Apples Watch). We use Long Short Term Memory neural networks for gesture recognition based on sensor data from both smartwatches. We provide an extensive evaluation of the classification accuracy of the system and provide a sensitivity analysis to find the Long Short Term Memory configuration that maximizes the classification accuracy. We also show the extent to which Long Short Term Memory neural networks outperform traditional machine learning approaches. We also illustrate an application we built for the Android and iOS platforms that allows developers to easily integrate the gesture recognition in their own systems. We conclude the paper with a description of use cases to underscore the potential impact of our contribution.