This work introduces Fractional Adaptive Linear Units (FALUs), a flexible generalization of adaptive activation functions. Leveraging principles from fractional calculus, FALUs define a diverse family of activation functions (AFs) that encompass many traditional and state-of-the-art activation functions. This family includes the Sigmoid, Gaussian, ReLU, GELU, and Swish functions, as well as a large variety of smooth interpolations between these functions. Our technique requires only a small number of additional trainable parameters, and needs no further specialized optimization or initialization procedures. For this reason, FALUs present a seamless and rich automated solution to the problem of activation function optimization. Through experiments on a variety of conventional tasks and network architectures, we demonstrate the effectiveness of FALUs when compared to traditional and state-of-the-art AFs. To facilitate practical use of this work, we plan to make our code publicly available

The combination of collaborative robots and end-to-end AI, promises flexible automation of human tasks in factories and warehouses. However, such promise seems a few breakthroughs away. In the meantime, humans and cobots will collaborate helping each other. For these collaborations to be effective and safe, robots need to model, predict and exploit human's intents for responsive decision making processes. Approximate Bayesian Computation (ABC) is approach to perform probabilistic predictions upon uncertain quantities. ABC includes priors conveniently, leverages sampling algorithms for inference and is flexible to benefit from complex models, e.g. via simulators. However, ABC is known to be computationally too intensive to run at interactive frame rates required for effective human-robot collaboration tasks. In this paper, we formulate human intent prediction as an ABC problem and describe two key performance innovations which allow computations at interactive rates. Our real-world experiments with a collaborative robot set-up, demonstrate the viability of our proposed approach. Experimental evaluations convey the advantages and value of human intent prediction for packing cooperative tasks. Qualitative results show how anticipating human's intent improves human-robot collaboration without compromising safety. Quantitative task fluency metrics confirm the qualitative claims.

Contextually aware intelligent agents are often required to understand the users and their surroundings in real-time. Our goal is to build Artificial Intelligence (AI) systems that can assist children in their learning process. Within such complex frameworks, Spoken Dialogue Systems (SDS) are crucial building blocks to handle efficient task-oriented communication with children in game-based learning settings. We are working towards a multimodal dialogue system for younger kids learning basic math concepts. Our focus is on improving the Natural Language Understanding (NLU) module of the task-oriented SDS pipeline with limited datasets. This work explores the potential benefits of data augmentation with paraphrase generation for the NLU models trained on small task-specific datasets. We also investigate the effects of extracting entities for conceivably further data expansion. We have shown that paraphrasing with model-in-the-loop (MITL) strategies using small seed data is a promising approach yielding improved performance results for the Intent Recognition task.

Deviant eating behavior such as skipping meals and consuming unhealthy meals has a significant association with mental well-being in college students. However, there is more to what an individual eats. While eating patterns form a critical component of their mental well-being, insights and assessments related to the interplay of eating patterns and mental well-being remain under-explored in theory and practice. To bridge this gap, we use an existing real-time eating detection system that captures context during meals to examine how college students’ eating context associates with their mental well-being, particularly their affect, anxiety, depression, and stress. Our findings suggest that students’ irregularity or skipping meals negatively correlates with their mental well-being, whereas eating with family and friends positively correlates with improved mental well-being. We discuss the implications of our study in designing dietary intervention technologies and guiding student-centric well-being technologies.

In this paper, we introduce ACAT-G, an interactive dialogue learning framework that incorporates constant human feedback into fine-tuning language models in order to assist conditioned dialog generation. The system takes in a limited amount of input from a human and generates personalized response corresponding to the context of the conversation within natural dialog time-frame. By combining inspirations from online learning, reinforcement learning, and large scale language models, we expect this project to provide a foundation for human-in-the-loop conditional dialog generation tasks.

Algorithmic transparency entails exposing system properties to various stakeholders for purposes that include understanding, improving, and contesting predictions. Until now, most research into algorithmic transparency has predominantly focused on explainability. Explainability attempts to provide reasons for a machine learning model's behavior to stakeholders. However, understanding a model's specific behavior alone might not be enough for stakeholders to gauge whether the model is wrong or lacks sufficient knowledge to solve the task at hand. In this paper, we argue for considering a complementary form of transparency by estimating and communicating the uncertainty associated with model predictions. First, we discuss methods for assessing uncertainty. Then, we characterize how uncertainty can be used to mitigate model unfairness, augment decision-making, and build trustworthy systems. Finally, we outline methods for displaying uncertainty to stakeholders and recommend how to collect information required for incorporating uncertainty into existing ML pipelines. This work constitutes an interdisciplinary review drawn from literature spanning machine learning, visualization/HCI, design, decision-making, and fairness. We aim to encourage researchers and practitioners to measure, communicate, and use uncertainty as a form of transparency.

In this work, we develop a dataset for incremental temporal summarization in a multiparty dialogue.We use crowd-sourcing paradigm with a model-in-loop approach for collecting the summaries and compare them with the expert-generated summaries.We leverage the question generation paradigm to automatically generate questions from the dialogue, which can be used to validate the user participation and potentially also draw attention of the user towards the contents that need to be summarized.We then develop several models for abstractive summary generation in the Incremental temporal scenario.We perform a detailed analysis of the results and show that including the past context into the summary generation yields better summaries as measured by ROUGE scores.

Automating parts of the user interface (UI) design process has been a longstanding challenge. We present an automated technique for optimizing the layouts of mobile UIs. Our method uses gradient descent on a neural network model of task performance with respect to the model's inputs to make layout modifications that result in improved predicted error rates and task completion times. We start by extending prior work on neural network based performance prediction to 2-dimensional mobile UIs with an expanded interaction space. We then apply our method to two UIs, including one that the model had not been trained on, to discover layout alternatives with significantly improved predicted performance. Finally, we confirm these predictions experimentally, showing improvements up to 9.2 percent in the optimized layouts. This demonstrates the algorithm's efficacy in improving the task performance of a layout, and its ability to generalize and improve layouts of new interfaces.

Although cost-effective at-home blood pressure monitors are available, a complementary mobile solution can ease the burden of measuring BP at critical points throughout the day. In this work, we developed and evaluated a smartphone-based BP monitoring application called textitSeismo. The technique relies on measuring the time between the opening of the aortic valve and the pulse later reaching a periphery arterial site. It uses the smartphone's accelerometer to measure the vibration caused by the heart valve movements and the smartphone's camera to measure the pulse at the fingertip. The system was evaluated in a nine participant longitudinal BP perturbation study. Each participant participated in four sessions that involved stationary biking at multiple intensities. The Pearson correlation coefficient of the blood pressure estimation across participants is 0.20-0.77 ($mu$=0.55, $sigma$=0.19), with an RMSE of 3.3-9.2 mmHg ($mu$=5.2, $sigma$=2.0).

Natural emotional experiences happen "in the wild" as people are mobile, living their daily lives. To capture these experiences, emotion researchers often give participants smartphone applications with various graphical user interfaces (GUIs) to record how they are feeling, however, there exist few empirical tests that assess the comparative benefits and drawbacks of different GUI designs. This paper presents two empirical evaluations of three types of GUI designs for capturing emotion using both a 10 participant in-lab trial and a 100 participant AMT trial. We define GUI scoring metrics and report on participants' ability to rate real world scenarios and evocative images, respectively, in ways that are consistent with population norms and with respect their own emotion word choices. We additionally report on users preferences for different designs, their perceived ease of use and the average time taken to complete an assessment for the different designs.

Determining the mode of transport of an individual is an important element of contextual information. In particular, we focus on differentiating between different forms of motorized transport such as car, bus, subway etc. Our approach uses location information and features derived from transit route information (schedule information, not real-time) published by transit agencies. This enables no up-front training or learning of routes and can be deployed instantly to a new place since most transit agencies publish this information. Combined with motion detection using phone accelerometers, we obtain a classification accuracy of around 90% on 50+ hours of car and transit data.

The proliferation of powerful smart devices is revolutionizing mobile computing systems. A particular set of applications that is gaining wide interest is recommender systems. Recommender systems provide their users with recommendations on variety of personal and relevant items or activities. They can play a significant role in today's life whether in E-commerce or for daily decisions that we need to make. We introduce a hybrid approach for solving the problem of finding the ratings of unrated items in a user-item ranking matrix through a weighted combination of user-based and item-based collaborative filtering. The proposed technique provides improvements in addressing two major challenges of recommender systems: accuracy of recommender systems and sparsity of data by simultaneously incorporating users' correlations and items ones. The evaluation of the system shows superiority of the solution compared to stand-alone user-based collaborative filtering or item-based collaborative filtering.

On school field trips, chaperoning teachers' foremost concern is the safety of the children, particularly ensuring that none of them go missing. However, they have limited attention resources and face many challenges in keeping track of their charges. We present RubberBand, an assistive application that helps alleviate the teacher's burden. Our approach adapts to diverse field trip environmental and child behavioral dynamicity, utilizing observations of the relative dispersion of children and their tendency to form sub-groups.

Field trips in kindergarten imply excellent chances to attain a wide spectrum of educational clues for the children. However, in-depth observation on their exploratory behaviors is uniquely challenging. Teachers mostly take all possible precautions against any incidents, sparing little time and attention for observation. We collaborated with kindergarten teachers to develop a system for delegated observation of the children's exploratory behaviors by using smartphones and sensor technologies.

We demonstrate E-Gesture, a collaborative architecture for energy-efficient gesture recognition on a hand-worn sensor device and an off-the-shelf smartphone that greatly reduces energy consumption while achieving high accuracy recognition under dynamic mobile situations. E-gesture employs a novel gesture segmentation and classification architecture carefully crafted by studying sporadic occurrence patterns of gestures in continuous sensor data streams and analyzing energy consumption characteristics in both sensor and smartphone.

Gesture is a promising mobile User Interface modality that enables eyes-free interaction without stopping or impeding movement. In this paper, we present the design, implementation, and evaluation of E-Gesture, an energy-efficient gesture recognition system using a hand-worn sensor device and a smartphone. E-gesture employs a novel gesture recognition architecture carefully crafted by studying sporadic occurrence patterns of gestures in continuous sensor data streams and analyzing the energy consumption characteristics of both sensors and smartphones. We developed a closed-loop collaborative segmentation architecture, that can (1) be implemented in resource-scarce sensor devices, (2) adaptively turn off power-hungry motion sensors without compromising recognition accuracy, and (3) reduce false segmentations generated from dynamic changes of body movement. We also developed a mobile gesture classification architecture for smartphones that enables HMM-based classification models to better fit multiple mobility situations.

Wireless Positioning Systems (WPS) are popular alternative localization methods, especially in dense urban areas where GPS has known limitations. Map-matching (MM) has been used as an approach to improve the accuracy of the estimated locations of WiFi Access Points (APs), and thus the accuracy of a wireless positioning system. Large-scale wireless positioning differs from satellite based positioning in at least two aspects: First, wireless positioning systems typically derive the location estimates based on war-driving access point (AP) data. Second, the locations of the AP beacons are not generally known at the same precision as that of the satellite locations. This results in lower accuracy and a lower confidence factor in the use of wireless positioning. This paper presents a fast selective look-ahead map-matching technique, called SLAMM. Existing MM algorithms developed for real-time location tracking of a moving vehicle are ill-suited for matching large collections of war-driving data due to the time complexity. Another unique feature of SLAMM is the map-matching of critical location samples in an AP trace to the road network before matching non-critical samples. Our experiments over a real dataset of 70 million AP samples show that SLAMM is accurate and significantly faster than the traditional MM approaches.

A kindergarten is an interesting community of young children. The children continuously share their interactions and experiences, and grow along similar developmental stages. In this setting, studying relative differences among them can be an interesting approach to investigating how to help their individual and social development. In this study, we present our intuition on inter-child behavioral relativity and apply it to a real kindergarten environment. We conduct a close user study necessitating the monitoring of the children's behavior. Then, utilizing wearable sensor technologies, we perform a field study to explore various interesting aspects of behavioral relativity in an automatic and quantitative fashion. We consulted the kindergarten teachers with our results obtained from our field study in order to validate the practical benefits in the kindergarten environment. We further discuss the potential, limitations, and opportunities of our approach.

This paper presents a machine learning approach for accurate estimation of energy expenditure using a fusion of accelerometer and heart rate sensing. To address short comings in existing off-the-shelf solutions, we designed Jog Falls, an end to end system for weight management in collaboration with physicians in India. This system is meant to enable people to accurately monitor their energy expenditure and intake and make educated tradeoffs to reach their weight goals. In this paper we describe the sensing components of Jog Falls and focus on the energy expenditure estimation algorithm. We present results from controlled experiments in the lab, as well results from a 15 participant user study over a period of 63 days. We show how our algorithm mitigates many of the issues in existing solutions and yields more accurate results.

Proper frequency planning is essential to enablemultiple IEEE 802.15.4 networks to coexist in the samespace, otherwise the network performance can beadversely affected due to interference. While this maybe feasible for static networks, it is not really an optionfor mobile networks such as body area networks due tothe dynamic nature of these interactions. This workdemonstrates one approach to detecting and mitigatinginterference in IEEE 802.15.4 networks from other802.15.4 networks. Interference is detected byobserving packets from other networks while themitigation strategy leverages collaboration betweeninterfering networks to determine which networkshould switch to a different channel. The approach canalso work with legacy networks that do not implementthis protocol. The system is implemented anddemonstrated on Intel Mote2 devices running TinyOS.

US water utilities are faced with mounting operational and maintenance costs as a result of aging pipeline infrastructures. Leaks and ruptures in water supply pipelines and blockages and overflow events in sewer collectors cost millions of dollars a year, and monitoring and repairing this underground infrastructure presents a severe challenge. In this paper, we discuss how wireless sensor networks (WSNs) can increase the spatial and temporal resolution of operational data from pipeline infrastructures and thus address the challenge of near real-time monitoring and eventually control. We focus on the use of WSNs for monitoring large diameter bulk-water transmission pipelines. We outline a system, PipeNet, we have been developing for collecting hydraulic and acoustic/vibration data at high sampling rates as well as algorithms for analyzing this data to detect and locate leaks. Challenges include sampling at high data rates, maintaining aggressive duty cycles, and ensuring tightly time-synchronized data collection, all under a strict power budget. We have carried out an extensive field trial with Boston Water and Sewer Commission in order to evaluate some of the critical components of PipeNet. Along with the results of this preliminary trial, we describe the results of extensive laboratory experiments which are used to evaluate our analysis and data processing solutions. Our prototype deployment has led to the development of a reusable, field-reprogrammable software infrastructure for distributed high-rate signal processing in wireless sensor networks, which we also describe.

The Intel Mote is a new sensor node platform with improved radio bandwidth and reliability due to the usage of Bluetooth radio. The connection-oriented nature of Bluetooth raises the issues of effective multi-hop network (scatternet) formation and maintenance that network and routing layer must address on top of the TinyOS abstractions. The hop distance and wireless link quality pose major challenges to multi-hop network performance, especially the connection-oriented networks such as Bluetooth scatternet. In this paper, we present a metric-based scatternet formation algorithm for the Intel Mote, which can optimize the Bluetooth network formation from the hop distance and link quality perspectives. In addition, a smart repair mechanism is proposed to deal with link/node failure and recover the network connectivity promptly with low overhead. The experiments with the Intel Mote platform demonstrate the effectiveness of the optimizations, which make the platform more powerful

The Intel mote is a new sensor node platform motivated by several design goals: increased CPU performance, improved radio bandwidth and reliability, and the usage of commercial off-the-shelf components in order to maintain cost-effectiveness. This new platform is built around an integrated wireless microcontroller consisting of an ARM*7 core, a Bluetooth radio, SRAM and FLASH memory, as well as various I/O options. The Intel Mote software architecture is based on an ARM port of TinyOS. Networking and routing layers have been created on top of the TinyOS base to provide Bluetooth-based multi-hop functionality. The network is self-organizing on startup and has mechanisms to repair failed links and circumvent failed nodes. A reliable high bandwidth streaming transport layer has also been created. The Intel Mote was deployed in an equipment monitoring application using industrial vibration sensors. This application was chosen since it benefits from the increased platform capabilities and network bandwidth of the Intel Mote platform. The paper presents a detailed analysis of the observed network operation, packet transfer rates, and power consumption.

No abstract available.

Sensing technology is a cornerstone for many industrial applications. Manufacturing plants and engineering facilities, such as shipboard engine rooms, require sensors to ensure product quality and efficient and safe operation. We focus on one representative application, preventative equipment maintenance, in which vibration signatures are gathered to predict equipment failure. Based on application requirements and site surveys, we develop a general architecture for this class of industrial applications. This architecture meets the application's data fidelity needs through careful state preservation and over-sampling. We describe the impact of implementing the architecture on two sensing platforms with differing processor and communication capabilities. We present a systematic performance comparison between these platforms in the context of the application. We also describe our experience and lessons learned in two settings: in a semiconductor fabrication plant and onboard an oil tanker in the North Sea. Finally, we establish design guidelines for an ideal platform and architecture for industrial applications. This paper includes several unique contributions: a study of the impact of platform on architecture, a comparison of two deployments in the same application class, and a demonstration of application return on investment.

The Intel Mote is a new sensor node platform motivated by several design goals: increased CPU performance, improved radio bandwidth and reliability and the usage of commercial off-the-shelf components in order to maintain cost-effectiveness. This new platform is built around an integrated wireless microcontroller consisting of an ARM*7 core, a Bluetooth* radio, RAM and FLASH memory as well as various I/O options. Due to the connection-oriented nature of Bluetooth, a new network formation and maintenance algorithms that are optimized for this protocol have been created. In particular, the "scatternet" mode of Bluetooth has been successfully adapted to form networks comprised of multiple piconet.

Random pattern testing methods are known to result in poor fault coverage for most sequential circuits unless costly circuit modifications are made. In this paper, we propose a novel approach to improve the random pattern testability of sequential circuits. We introduce the concept of holding signals at primary inputs and scan flipflops of a partially scanned sequential circuit for a certain length of time, instead of applying a new random vector at each clock cycle. When a random vector is held at the primary inputs of the circuit under test or at the scan flip-flops, the system clock is applied and the primary outputs of the circuit are observed. Information obtained from a testability analysis or test generator is used to determine the number of clock cycles for which each random vector is to be held constant. The method is low cost and the results of our experiment on the benchmark circuits show that it is very effective in providing fault coverage close to the maximum obtainable fault coverage using random patterns with full scan.