text stringlengths 128 2.05k |
|---|
[MATH] Formally, a type interpretation [MATH] assigns a type variable [MATH] a triple [MATH] where [MATH] is a binary relation on closed values [MATH] , where [MATH] is of type |
[MATH] . There are two closure conditions on [MATH] The first condition on [MATH] is that it has to be closed under wrappers produced by reflexive casts : if [MATH] the value of [MATH] is related to [MATH] (for |
[MATH] ). This closure condition is needed due to polymorphic casts of the form [MATH] A polymorphic cast is a function typed at [MATH] , so it should produce values related at [MATH] when taking arguments related at |
[MATH] Since values related at [MATH] should be in [MATH] , the results of evaluating [MATH] and [MATH] should be in [MATH] for any [MATH] (if they terminate at values). Unfortunately, it could not be achieved if [MATH] were arbitrary, because, if [MATH] is instantiated with higher-order types, |
[MATH] produces wrappers (e.g., by R_Fun ) but they may not be in [MATH] Thus, instead of taking arbitrary [MATH] , we require [MATH] to contain also the wrappers. |
Actually, an alternative requirement that [MATH] relates the value of [MATH] to that of [MATH] would be enough if we are interested only in soundness of the logical relation. Our closure condition—without [MATH] —subsumes this alternative and, in fact, is a key to proving correctness of the upcast elimination and the s... |
The second closure condition is that [MATH] is closed under (semityped) CIU-equivalence so that the logical relation is complete with respect to contextual equivalence, following the prior work Ahmed_2006_ESOP CIU-equivalence Mason/Talcott_1991_JFP relates two closed terms if they behave equivalently under any evaluati... |
only after we finish proving the upcast elimination property in Section LABEL:sec:reasoning-upcast-elim . So, we have to introduce the two conditions separately. Interestingly, the closure under CIU-equivalence also enables us to show transitivity of the logical relation. We will show that CIU-equivalence, the logical ... |
4.2. Preliminaries Here, we give a few preliminary definitions, including CIU-equivalence and the closure conditions on [MATH] , to define the logical relation. |
{defi} [MATH] is the set [MATH] of all closed, well-formed types; [MATH] is the set [MATH] of all closed types; For each [MATH] [MATH] is the set [MATH] of all closed values of [MATH] ; and |
[MATH] is the set [MATH] of all closed values. In what follows, (capture-avoiding) substitutions , denoted by [MATH] , are maps from term and type variables to closed terms and types, respectively, and they can be extended to maps over terms, types, etc. straightforwardly. We write [MATH] and [MATH] for substitutions t... |
{defi} [Closing Substitutions] Substitution [MATH] is a closing substitution that respects [MATH] , written [MATH] , if and only if [MATH] for any |
[MATH] and [MATH] for any [MATH] We define CUI-equivalence below. Our CIU-equivalence rests on static evaluation contexts [MATH] , where holes do not occur under run-time term constructors such as active checks. |
[EQUATION] Since a static evaluation context is also a (single-hole) context, we use the context well-formedness judgments also for static evaluation contexts and write [MATH] Use of static evaluation contexts, instead of evaluation contexts, is important to show the equivalence-respecting property, especially, |
# Source: arxiv 1806.07055 # Title: Capacitor Based Activity Sensing for Kinetic Powered Wearable IoTs # Sections: all # Downloaded: 2026-03-03T02:18:12.158346+00:00 |
Capacitor Based Activity Sensing for Kinetic Powered Wearable IoTs Abstract. We propose a novel use of the conventional energy storage component, i.e., capacitor, in kinetic-powered wearable IoTs as a sensor to detect human activities. Since different activities accumulate energies in the capacitor at different rates, ... |
Kinetic energy harvesting, Capacitor, Activity recognition, Wearable IoTs 1. Introduction The rapid development of embedded technology has enabled wearable IoTs (Seneviratne et al 2017 that provide autonomous health and fitness monitoring services, such as step-counting (Consolvo et al 2008 and recognition of daily act... |
In general, the power consumption in sampling is directly proportional to the sampling rate, as the higher the sampling rate, the more power is consumed by the sensors as well as the microcontroller (MCU), which has to wake up more frequently to read, process, and store the samples. A large volume of past research on c... |
In this paper, we propose a new way to detect activities for kinetic energy harvesting powered wearable IoTs, which obviates the need for frequent motion signal sampling and allows very aggressive duty cycling of the MCU to reduce power consumption of wearable devices by several orders. To avoid motion signal sampling,... |
(1) The kinetic power of human activities are distinct. It has been widely demonstrated in the literature that the kinetic energy harvested from different activities are distinctively different (Gorlatova et al 2014 ; Yun et al 2008 . Thus, the energy generation rate of the kinetic-powered wearable device can be used a... |
(2) Capacitor provides accumulated information . In kinetic powered devices, the energy energy generation rate of the external activity. Interestingly, charging rate of the capacitor can be obtained by simply reading the capacitor voltage at the end of each activity detection period, which is typically about 5 seconds ... |
Thus, it should be possible to classify human activities by simply reading the capacitor voltage only once in every 5 seconds. Comparing with conventional motion signal based activity detection, which requires the system to wake up many times per second (Lara and Labrador, 2013a , CapSense allows very aggressive duty c... |
The contributions of this paper can be summarized as follows: (1) We propose a new method for human activity sensing, CapSense, which detects activity from the charging rate of the energy storing capacitor. To the best of our knowledge, such capacitor-based activity detection has not been explored before. |
(2) We address the first challenge of non-linear capacitor charging by jointly configuring the parameters of the capacitor and the associated energy harvesting circuits, which allows us to operate with capacitor charging cycles that are approximately linear. |
(3) We implement the idea of CapSense in shoe form factor using piezoelectric bending energy harvester. We address the single feature classification challenge by introducing two energy harvesters and capacitors, one at the rear of the sole and the other at the front. We show that the proposed dual-capacitor system sign... |
(4) Using our dual-capacitor prototype, we conducted experiments with 10 subjects performing 5 different activities. We demonstrate that CapSense is capable of detecting daily activity with up to 95% accuracy. |
(5) We conduct a detailed power profiling to quantify the power saving opportunity of CapSense. Our measurement results indicate that, compared to the state-of-the-art, CapSense reduces sampling-related power consumption by 54% and the overall IoT system power consumption by 73%. |
Partial and preliminary results of this paper have appeared in our previous work (Lan et al 2017 . In this paper we provide the following two major extensions to the conference version: (1) We redesign the previous CapSense prototype by adding a second energy harvester and capacitor to the shoe insole (at the front) th... |
The rest of the paper is organized as follows. We first introduce some background of kinetic-powered IoT in Section . Then, we present the design and implementation of CapSense in Section , followed by its performance evaluation in Section . The power measurement study is presented in Section . We review the related wo... |
2. Preliminaries in kinetic powered IoT In this section, we provide some basic background of kinetic-powered IoTs and the concept of using kinetic energy harvesting transducer for sensing. |
2.1. Kinetic Energy Harvesting Kinetic energy is the energy of an object due to its motion. Kinetic energy harvesting refers to the process of scavenging kinetic energy released from human activity or ambient vibrations. The use of kinetic energy harvesting for self-powered IoT has been widely investigated in the liter... |
There are three main energy transduction techniques that are widely used in the literature, namely, piezoelectric electromagnetic , and electrostatic . Among them, piezoelectric is the most favourable transduction mechanism for wearable IoTs, due to its simplicity and compatibility with MEMS (micro electrical mechanica... |
2.2. KEH Transducer-based Sensing Although KEH transducer is designed with the purpose of scavenging kinetic energy from motions, researchers have investigated the use of KEH transducer as a low power vibration sensor for context detection (Li et al 2013b ; Kalantarian et al 2015 ; Khalifa et al 2018 , in which, the AC... |
3. System Overview In this section, we present the concept, design, and implementation of CapSense. 3.1. CapSense Concept Figure exhibits the processing pipeline of a typical activity sensing system. It usually consists a sequence of procedures, including the acquisition of motion signal from sensor, signal processing,... |
As shown in Figure , unlike accelerometer or KEH transducer-based systems that require a time-series of instantaneous motion signal sampled from the sensor at a high frequency, CapSense utilizes a single sample of the capacitor voltage for activity recognition. The feasibility of CapSense relies on two fundamental fact... |
Fact 1. The kinetic power of human activities are distinct . It has been widely demonstrated in the literature that the kinetic power harvested from different activities are different (Gorlatova et al 2014 ; Yun et al 2008 . Thus, the energy generation rate of the kinetic-powered wearable device can be used as a featur... |
Fact 2. Capacitor provides accumulated information . As shown previously in Figure , in kinetic powered devices, the energy energy generation rate of the external activity. Interestingly, as it will be shown in Section 3.5 , charging rate of the capacitor over the last [MATH] second(s) can be estimated by simply readin... |
Those two facts imply that by leveraging the capacitor voltage change over a time period of [MATH] , we can estimated the corresponding energy generation rate and leverage it to recognize the activity performed by the user in the last [MATH] period of time. As we will demonstrate later in Section , CapSense can detect ... |
3.2. CapSense Architecture There are a various of design options in kinetic energy harvesting powered wearable devices, such as backpack (Xie and Cai, 2014 , fabric (Yang and Yun, 2012 , wristband (SEIKO, 2016 , and footwear (Kymissis et al 1998 ; Shenck and Paradiso, 2001 . We designed our system in the form-factor of... |
Figure exhibits the system architecture of CapSense and visualizes how it works. Considering the scenario in which a subject is wearing the KEH-powered shoes and doing some activities, e.g., walking or running, her foot will hit the ground floor and the pressure induced by both the heel and forefoot strikes will bend t... |
As shown in Figure , CapSense consists of two parts: Load and Energy Harvesting Load represents any system components responsible for data sensing, processing, and communication, or could be a rechargeable battery that can be used to power a wearable system. The Energy Harvesting corresponds to the functional component... |
3.3. CapSense Prototype Design In the following, we present the design and implementation of CapSense. Figure 5(a) gives the pictures of our prototype which we implemented in the form of shoe. As discussed previously, our prototype consists of two parts, the Energy Harvesting and Load . For the Energy Harvesting part, ... |
The output pins of the PEH transducer are connected to an energy harvesting circuit, namely the LTC3588-1 from the Linear Technology . The LTC3588-1 integrates a low power-loss bridge rectifier that can be used to rectify the AC voltage output from the PEH transducer, and a high efficiency buck converter that is able t... |
[MATH] , and equals to [MATH] in our setting), the buck converter will be enabled to discharge the energy stored in the capacitor. On the other hand, when the capacitor voltage has been discharged below the lockout falling threshold (i.e., denoted by [MATH] |
[MATH] , and equals to 3.08 [MATH] in our setting), the buck converter will be turned off, and the capacitor starts to accumulate any harvested energy. |
The simplified circuit diagram of the prototype is shown in Figure . The front and rear capacitors are used to store the energy harvested from the front and rear PEH, respectively. For analysis purpose, we use an Arduino Uno board to sample the voltage of the two capacitors through the onboard 10-bit ADC at 100 Hz and ... |
3.4. Ensuring Linearity in Capacitor Voltage Before presenting the details of capacitor-based sensing, we first analyze some properties of the capacitor when the system is powered by an energy harvester, and discuss the feasibility and design requirement of leveraging the capacitor voltage for activity sensing. |
The voltage of the capacitor, [MATH] , at time [MATH] during the charging is given by: [EQUATION] in which, [MATH] is the maximum voltage to which the capacitor can be charged, and it is bounded by [MATH] , where [MATH] is the voltage applied to the capacitor, i.e., the rectified DC voltage from the rectifier in our ca... |
The relation between the capacitor voltage, [MATH] , and time [MATH] is visualized in Figure . The theoretical curve indicates the voltage of the capacitor when it is charged by the supply power [MATH] over time (in our case, [MATH] is the rectified DC voltage from the rectifier). The first observation is that, within ... |
Fortunately, as we can observe in Figure , with time [MATH] , the theoretical curve of [MATH] (defined in Equation ) can be approximated by a linear curve (Macdonald and Brachman, 1955 . According to the RC circuit theory, a capacitor can be charged to 39.3% of [MATH] with a charging time of [MATH] , this means that to... |
[EQUATION] which yields: [EQUATION] in which, [MATH] is the undervoltage lockout rising threshold of the buck converter, at which the capacitor starts to be discharged. This means that, to ensure the linearity in the capacitor voltage, the selection and configuration of the hardware components (i.e., the PEH transducer... |
As discussed in Section 3.3 , in our prototype, [MATH] of the buck converter has been configured to 4V . This means that, given Equation (3), we have: [MATH] . The rectified DC voltage from the rectifier, [MATH] , depends on the energy harvester that is used in the system. Given different materials and configurations o... |
3.5. Activity Sensing using Capacitor Voltage In the following, we discuss how to leverage the capacitor voltage for activity sensing. Figure exhibits an actual voltage trace showing the charging and discharging cycles of the capacitor when it is powered by the energy harvester. The charging/discharging behavior of the... |
During the charging/discharging of capacitor, CapSense duty-cycles the system MCU to periodically sample the capacitor voltage. As an example shown in Figure , MCU wakes up to sample the capacitor voltage once every [MATH] seconds. Using any two adjacent voltage samples, it is straightforward to estimate the capacitor ... |
[EQUATION] in which, [MATH] and [MATH] is the capacitor voltage at time [MATH] and [MATH] , respectively. Therefore, by periodically sampling the capacitor voltage at a frequency of [MATH] Hz, we can estimate the energy generation rate of the PEH transducer (either the front or the rear PEH transducer in our prototype)... |
3.5.1. Impact from the capacitor discharging However, the estimation of [MATH] may affect by the discharging time of the capacitor. As if it takes a long time for the buck converter to discharge the capacitor from 4V to 3V, it is possible that the MCU wakes up and samples an incorrect voltage value during the discharge... |
3.5.2. Impact from the capacitor self-discharge Another factor that may affect the estimation of [MATH] is the self-discharge of the capacitor. That is, the voltage leakage of the capacitor when it is not charged by the energy harvester. A high voltage leakage may result in an underestimation in [MATH] , as part of the... |
3.5.3. Putting all together As an example, Figure 10 plots the voltage traces of the capacitor when a subject is doing different activities. We can observe that, for all five activities, our hardware design ensures an approximately linear increase in the voltage when the capacitor is powered by the energy harvesters. A... |
We can also notice from Figure 10 that the difference in capacitor voltages among different activities increases with the accumulation time. For instance, with [MATH] =1s, the voltage increment rates are similar among different activities. Since CapSense merely uses the voltage increment rate for activity classificatio... |
4. System Evaluation 4.1. Experimental Setup and Data Collection The subjects were asked to wear our energy-harvesting embedded shoe during the data collection. We prepared shoes with different sizes to meet the requirement of our subjects. The prototype system is attached to the subject’s ankle (as shown in Figure 5(a... |
Recall our discussions in Section 3.5 that CapSense leverages the increasing rate [MATH] over the last accumulation time of [MATH] for activity recognition. Following Equation , we calculate the [MATH] of different activities with different [MATH] for all the 10 subjects using our dataset. The estimated set of [MATH] a... |
4.2. Activity Recognition Performance The evaluation is carried out in WEKA using 10-folds cross validation with 10 repetitions for each test. Four typical machine learning algorithms are used: the C4.5 decision tree algorithm (C4.5) (Quinlan, 2014 , IBk K-Nearest Neighbor classifier (Aha et al 1991 , Naive Bayes with ... |
4.2.1. Recognition Accuracy vs. Subject Intuitively, given the diversity in subject’s gender, weight, and height, the foot strike pressures applied on the energy harvesters differ in the way subjects perform the activity. In the following, we consider RandomForest as the classifier and fix the accumulation window [MATH... |
Figure 12 exhibits the achieved accuracy for all the 10 subjects. As expected, the accuracy varies with subject. For example, Subject 4 achieves 75.50% of accuracy using the voltage samples from the Front PEH as signal, whereas, Subject 6 achieves a much higher accuracy of 96.75% using the same PEH. The corresponding c... |
As shown in Figure 13(a) , the confusion matrix indicates that the major error happens in the classification between ‘WALK & DS’ , and ‘WALK & DS’. This results from the high similarity in the voltage increment rate [MATH] when Subject 4 is conducting those activities. As shown in Figure 14(a) , we can observe a high o... |
4.2.2. Recognition Accuracy vs. Different PEH In the following, we examine the accuracy achieved by using the signal from different PEHs, and investigate the possibility of using signal fusing from two PEHs to increase the accuracy. |
The results given in Figure 12 indicate that by using the capacitor voltage from the Rear PEH we can achieve a higher classification accuracy for most of the subjects (except Subject 6). Taking Subject 4 as an example again. We can notice from Figure 14(b) that the voltage increment rate are more separated among differ... |
Moreover, CapSense achieves better performance by fusing the signal from the two PEHs. That is, a two-dimensional vector, [MATH] , is used as the input for classification, in which [MATH] and [MATH] refers to the capacitor voltage increment rate from the Rear and Front PEH, respectively. As an example, the scatter plot... |
However, as shown in Figure 15(c) , after fusing the signal of Front and Rear PEHs, the confusions between those classes are significantly resolved. Intuitively, as exhibited in Figure 16 and 15(c) , after fusing the signal from both Front and Rear PEHs and applying the two-dimensional voltage vector for classification... |
4.2.3. Recognition Accuracy vs. Accumulation Time Now, we investigate the impact of accumulation time [MATH] on the recognition accuracy. Figure 17 exhibits the achieved accuracy given different [MATH] . The classifier used in this experiment is RandomForest. For a particular [MATH] , the reported results are the avera... |
4.2.4. Recognition Accuracy v.s. Classifier Lastly, we analyze the performance of CapSense with different classifiers. Table exhibits the accuracy of CapSense with different classifiers given [MATH] . The results are averaged across all the 10 subjects. We can observe that, all the four examined classifiers can achieve... |
5. Energy Consumption Analysis High energy consumption is the major roadblock for the pervasive use of wearable technology (Seneviratne et al 2017 . In this section, we will conduct an extensive power consumption profiling of off-the-shelf wearable activity recognition systems to investigate the superiority of CapSense... |
5.1. Setup for Energy Consumption Analysis We use an off-the-shelf Texas Instrument SensorTag as the target device, which is embedded with the ultra-low power ARM Cortex-M3 MCU that is specifically designed for today’s energy-efficient wearable devices . The SensorTag is running the Contiki 3.0 operating system which d... |
5.2. Power Consumption in Sampling First, we investigate the power consumption in data sampling. In the measurement, both capacitor voltage and KEH transducer signal are simpled through the on board ADC of SensorTag. The sampling frequency of MCU is configured as 25Hz to meet the requirement of KEH transducer-based sen... |
In general, for the duty-cycled activity sensing system, the average power consumption in data sampling, [MATH] , can be obtained by the following equation: |
[EQUATION] where, [MATH] is the average power consumption of the system during the sampling event, and [MATH] is the average power consumption when the MCU is in deep-sleep mode (with all the other system components power-off). [MATH] is the sampling frequency, and [MATH] is the duration of time (in milli-second) requi... |
The results are compared in Table . As shown, CapSense is able to save 54% of the overall power consumed by the transducer-based system in data sampling . We can also notice that, for CapSense, the main energy expenditure is the MCU Sleep (i.e., the unavoidable power consumption of the system when MCU is in the deep-sl... |
5.3. Power Consumption for Data Transmission In the following, we investigate the power consumption in wireless data transmission using the BLE beacons. We programmed the Contiki OS to wake-up the CC2650 wireless MCU periodically and transmits a BLE beacon packet, which broadcasts three times on three separate channels... |
For transducer-based system with a sampling frequency of 25Hz, it has [MATH] voltage samples (2 Bytes for each 12-Bits ADC reading, and 250 Bytes in total) to be transmitted per channel once every five seconds. Given the maximum additional data can be added to each beacon packet is 28 Bytes, this requires [MATH] packet... |
Combining the power consumption in data sampling and transmission together, the overall system power consumption for KEH transducer-based system is 28.15 [MATH] , whereas, the overall system power consumption for CapSense is only 7.53 [MATH] . This means that CapSense is able to save 73% of the overall system power con... |
6. Related Work In this section, we review existing works in developing energy-efficient mobile sensing system. We first review the efforts in building insole-based self-powered wearable system. Then, we introduce some recent efforts in utilizing KEH-transducer as the motion sensor for energy-efficient sensing. Lastly,... |
6.1. Insole-based Energy Harvesting System With recent advances in energy harvesting hardware, researchers are now turning to kinetic energy harvesting as a viable source of power to extend battery life or even replace the batteries altogether in wearable devices (Paradiso and Starner, 2005 ; Mitcheson et al 2008 . Som... |
In their prototype consists a pair of shoes, an accelerometer and Bluetooth wireless communication unit are powered separately by the energy harvested from each of the two feet, and coordinated by ambient backscatter. Such that, the accelerometer can sense the activity of the user, while the Bluetooth can transmit the ... |
6.2. KEH-transducer based Context Sensing Thanks to the existing efforts in kinetic-powered wearable systems, some studies in the literature start to apply KEH transducer as a low power vibration sensor. The motivation behind this idea is to further reduce the energy consumption in powering conventional motion sensor, ... |
6.3. Reducing Sampling Frequency For both KEH-transducer based and conventional motion sensor based sensing systems, the energy consumption in sensing is proportional to the sampling frequency. Thus, a large volume of works in the literature focused on reducing the sampling rates to save the energy (Bulling et al 2014 ... |
7. Conclusion In this paper, we present CapSense, a novel activity sensing scheme for KEH-powered wearable devices. By simply using the voltage readings of the energy harvesting capacitor at 0.2Hz, CapSense is able to daily activities with 95% accuracy, and reduce system power consumption by 73%. The current work is a ... |
# Source: arxiv 1806.07106 # Title: A study of the stability properties of Sagdeev solutions in the ion-acoustic regime using kinetic simulations # Sections: all # Downloaded: 2026-03-02T08:56:04.172975+00:00 |
A study of the stability properties of Sagdeev solutions in the ion-acoustic regime using kinetic simulations Abstract The Sagdeev pseudo-potential approach has been employed extensively in theoretical studies to determine large-amplitude (fully) nonlinear solutions in a variety of multi-species plasmas. Although these... |
Introduction I.1 Theoretical study of ion-acoustic solitary waves The study of nonlinear waves plays a crucial role in science as we know it today Scott ( 2007 , and the field of nonlinear physics has opened numerous research frontiers and technological applications. In the heart of this research field stands the cruci... |
in “the happiest day of his life” Russell ( 1865 these waves have defied the narrative of physics since their discovery and pushed physicists to a new realm, i.e. nonlinear physics. |
The study of solitary waves in plasmas started with the work of Washimi and Taniuti Washimi and Taniuti ( 1966 who successfully derived the Korteweg–de Vries (KdV) equation |
De Jager ( 2006 for ion-acoustic waves in the nonlinear regime utilizing the well-known “reductive perturbation method” Taniuti and Wei ( 1968 ); Kakutani et al. 1968 Although effective in deriving evolution equations for weakly nonlinear systems, this method cannot be applied beyond the small amplitude limit. |
The Sagdeev pseudo-potential method Sagdeev ( 1966b (also known as mechanical-motion analog) offers an alternative approach to the problem of finding solitary waves, and it overcomes the limitation of small amplitude approximations. However, it does not provide information about the temporal evolution. Since it does no... |
The method basically alters the Poisson’s equation in order to rearrange it as a general energy equation of the form, [EQUATION] |
which can be compared to a general energy conservation equation. In Eq. , the first term plays the role of kinetic energy and [MATH] is the so called Sagdeev pseudo-potential If we consider [MATH] as a position and [MATH] as time, this represents a classical particle of unit mass moving in a one-dimensional potential [... |
and magnetospheric plasmas Berthomier, Pottelette, and Malingre ( 1998 In this context, three concepts should be recognized and defined properly to avoid misunderstanding: |
Sagdeev solutions solitary waves solitons. Sagdeev solutions are localized solutions (pulses) to Eq. (i.e. solutions such that [MATH] and [MATH] when [MATH] ), which might or might not be solitary waves or solitons. To prove that they are true solitary waves, one needs to demonstrate the stability against small perturb... |
It should be noted that both the reductive perturbation theory and the Sagdeev pseudo-potential approach have mostly been developed within the fluid framework, i.e. ignoring kinetic effects. They relay on the number density (the first moment of the distribution function) as the starting step of the dynamical model. The... |
I.2 Simulation study of electron holes Alongside the theoretical attempts to predict the shape solitary waves and their stability, there exist a line of research based on simulation focusing on the property and stability of phase space holes. Since solitary waves are always accompanied by an hole in the phase space (in... |
1. chain formation process (disintegration) Zabusky and Kruskal ( 1965 ); Kakad, Omura, and Kakad ( 2013 ); Kakad, Kakad, and Omura ( 2014 ); Sharma, Sengupta, and Sen ( 2015 ); Qi et al. 2015 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.