Datasets. This Data Use Agreement (“Agreement”) is made and entered into by and between the SOUTH ALAMO REGIONAL ALLIANCE FOR THE HOMELESS (“Covered Entity”), a 501(c)(3) non-profit organization in the State of Texas, and Insert Name of Entity (“Data Recipient”), a [insert description of legal status (public / private / profit / non-profit / company incorporated in what state, LLP or LLC or other entity incorporated or registered in what state)] on (Today’s Date) for the purpose of collecting and analyzing Homeless Management Information System (HMIS) data from XXXXX.
Datasets. In order to ensure a fair comparison, we used the same data as used by [2]. The data is part of CoMon project which is a monitoring infrastructure for PlanetLab [2]. It contains CPU utilization of more than thousands of virtual machines measured from servers located at different places all around the world. The data of ten random days are chosen from the workload traces that are collected in March and April 2011. The data files contain CPU utilization values of virtual machines measured every 5 min for 24 h. Note that each line in a file represents a single request, and data for each day are combined in a single folder. In total, all the folders contain 11,746 files, which contain over 3.3 million user requests. The properties of the data are summarized in Tab. 2.
Datasets. Section 4.11(o) of the Disclosure Schedule lists and describes all datasets that: (1) the Company considers proprietary and material to the Business; or (2) have been used within or to make available the Company Products and related services (the “Company Datasets”). The Company or its Subsidiaries have a valid legal right in and to use the Company Datasets. The Company and its Subsidiaries have obtained all necessary permissions and consents from customers to use data from the equipment and premises of such customers that are collected by, and provided to, the Company or its Subsidiaries, including through its products and services. Except as set forth in Schedule 4.11(o) of the Disclosure Schedule, the Company Products do not include or use artificial intelligence models.
Datasets. We evaluate the proposed system on two datasets of different complexity. Exemplary screenshots are shown in Fig. 3, with statistics on these two datasets in Table I. The first one is a popular 2D laser dataset recorded with a stationary SICK LMS-291 laser scanner at the Main Station in Freiburg, Germany. The second, new dataset is significantly more complex in terms of crowd dynamics and number of tracks within sensor range at a given time. This dataset has been synthetically generated using a combination of the pedestrian simulator PedSim, and the robot simulator Gazebo. Using our ROS wrapper of PedSim4, we have modelled a scenario similar to the one our service robot will encounter in an airport terminal (Fig. 1), with large flows of people moving around corners towards specific goals, static groups and single persons spread all over the environment, and other people queuing up. Person interac- tions are modelled in PedSim using the social force model after Xxxxxxx [20], and pedestrian positions are then fed to Gazebo to continuously reposition 3D meshes in a scene in order to generate 2D laser scans via raycasting from a simulated mobile platform. While this approach obviously cannot correctly depict reality in all its detail, especially in terms of background complexity, it allows us to quickly study different scenarios, robot behaviors, and obtain exact groundtruth without need for manual annotations. 4xxxxx://xxxxxx.xxx/srl-freiburg/pedsim_ros As discussed extensively in [22], XXXX scores can vary between implementations and are dependent on meta- parameters such as the matching distance threshold θd and the way in which track hypotheses are assigned to groundtruth objects. Therefore, it is important to use the same metrics implementation when comparing different tracking approaches. While in our previous work [2], the track- ing metrics were tightly integrated into the tracker, our framework provides a standalone Python implementation of CLEAR-MOT metrics as a separate ROS node that can be used to evaluate any kind of tracking algorithm compati- ble with our ROS message definitions. In our version, we compute groundtruth correspondences using a variant of the Hungarian method based upon Euclidean distances between object centroids in 2D world coordinates, with θd = 1m. Although XXXX can give a good impression of overall tracking performance, it is questionable if it alone can be a sufficient measure of tracking performance, as it weights all types o...
Datasets. For our experiments, we have recorded two entirely new, multi-modal datasets (cf. Fig. 3). The Motion Capture Se- quence has been recorded in a narrow lab environment in front of our robot platform, shown in Fig. 2c, which remains stationary throughout this sequence. The recorded sensor data includes a 190-degree frontal 2D laser scan from a
Datasets. To validate the proposed approach, we have consid- ered five different benchmarks, namely: Handwritten digits dataset (MNIST), XXXXX00, XxxxxXxxx, Animals with At- tributes 2 (AwA2), Street View House Numbers (SVHN). Figure 3 shows some image samples from these data sources. × MNIST [28] dataset consists of 70000 images with dimen- sion 28 28. The dataset has been split in 60000 and 10000 images for training and testing respectively. The dataset is a collection of greyscale images of handwritten numbers clas- sified among 10 classes. SVHN contains 73000 and 26000 real-life digit images for training and testing. × CIFAR10 [25] is a well known standard dataset for im- age recognition experimentation, it consists of 60000 im- ages from 10 classes of objects from different contexts. We maintain the dataset split in training and test suggested by the dataset authors: 50000 images in training set and 10000 images in test set. The images have dimension 32 32 and they are defined over three colour channels (RGB colour space). Model MNIST SVHN XXXXX00 XxxxxXXXX XxX0 Baseline CapsNet AA-Caps (Ours) 99.67% (100E) 99.34% (100E) 93.23% (100E) 92.13% (100E) 68.70% 71.60% 89.56% (50E) 89.72% (50E) 12.1% (100E) 23.97% (100E) Table 1. Summary of evaluation results. The model is validated over different bechmark to prove the contribution provided respect to the original CapsNet. We present results obtained with MNIST, SVHN, XXXXX00, XxxxxXXXX, and AwA2 datasets. Model Description Parameters Test Acc. CapsNet (Baseline) Conv - Primary Capsules - Final Capsules 8.2M 99.67% AA-Caps (Ours) Conv - Primary Capsules - Self-Attention - Conv 6.6M 99.34% Table 2. Comparison of CapsNet model with AA-Caps . The table presents a brief description layers that compose the structure of baseline CapsNet compared to the structure of AA-Caps, the number of trainable parameters, and the accuracy achieved by the model after 100 epochs on MNIST dataset. × SmallNORB [29] consists of 24300 image 96 96 stereo grey-scale images defined over 2 colour channels. We re- sized the images to 48x48 and during training processed random 32x32 crops, and central 32x32 patch during test. AwA2 [51] consists of 37322 images of 50 animals classes. The images are collected from public sources, that makes the dataset challenging due to the uncontrolled images.
Datasets. The selected algorithms were examined on 30 datasets, which were selected manually from the UCI Machine Learning Repository and are widely used by the pattern recognition community for evaluating learning algorithms. The datasets vary across such dimensions as the number of target classes, of instances, of input features and their type (nominal, numeric).
Datasets. Pilot 1a aims to predict the maintenance status of wind turbine electrical drivetrain components, such as generators and power converters. It examines onshore and offshore wind turbines powered by a doubly-fed induction generator and provides five datasets: • La Haute-Lys dataset- Wind turbine SCADA data (SCADA-Pilot1a): This dataset is collected from various wind turbines situated in multiple wind farms. There is a unique data structure and tag name for every turbine brand. A wind turbine's Supervisory Control and Data Acquisition system contains sensor data at the most important subcomponents of the wind turbine; the collected data are analysed at 10- minute intervals. Turbines during a period where the electrical subcomponents had faults are also included. • High-frequency Data (VUB-Pilot1a): This data is derived from a dedicated measurement campaign on onshore wind turbines and consists of a limited set of electric measurements and operational parameters (e.g., wind speed). • Open wind speed dataset (Flemish-banks-data-Pilot1a): includes environmental measurements (e.g., wind speeds, wind directions) collected along the Belgian North Sea. As a basis for defining semantic labels describing wind conditions, the dataset in LLUC 1a-01 shows the typical range of wind measurements occurring in the field. • Offshore measurement campaign data (High-frequency-accelerations-Pilot1a): This dataset includes acceleration measurements, that were taken of the drivetrain of an offshore wind turbine. • Dedicated current measurement campaign data (ENGIE-VUB-Pilot1a): This dataset consists of current signals that are acquired on an onshore wind turbine. These data are similar to the La Haute Lys dataset. As such they will be merged in further discussions on data handling and analytics with the La Haute Lys data as the same processing methodology applies.
Datasets. Pilot 2a focuses on integrating and deploying different PLATOON analytical services with the Institute Xxxxxxx Xxxxx (IMP) proprietary VIEW4 Supervisory control and data acquisition (SCADA) system deploys the energy value chain in Serbia. Energy resources related to Renewable Energy Sources (RES) in this pilot include: wind power plants and PV power Plants. Electricity production from solar and wind plants is subject to forecast errors that drive demand for balancing. These data sources are described as follows: PUPIN-RES-PROD: Historical Wind Power Production Measurements; it contains measurements of the production from the wind power plant, as well as topology data. PUPIN-RES-PV (Predictive Maintenance): Data is collected by the Phasor Measurement Unit installed at Institute Xxxxxxx Xxxxx. PUPIN-WeatherBit: Meteorological Data for RES Production (Generation) Forecasting Modelling Data. Meteorological dataset is utilized for RES production forecasting models training process as input data. Data is historical observational data. PUPIN-RES-Effects: Power System calculated based on the input by Phasor Measurement Unit installed at PUPIN. PUPIN-ENTSO-E: Transparency Platform-Energy Identification Codes (EICs); it maintains data about 39 electricity transmission system operators (TSOs) from 35 countries in Europe. These four data sources composed the catalog EBPM (Electricity Balance and Predictive Maintenance); they provide data in English (ENG) and Serbian (RS). More details in D2.4.
Datasets. Two use cases are demonstrated in ParcBit's technological park in Palma de Mallorca, Spain. ParcBit's grid consists of a 5 km long mid-voltage network and 5 km long low-voltage network. Pilot 2B uses three datasets: • Power grid ZIV Power Meters (Power-grid-ZIV-Pilot2b): dataset consists of hourly measurements of active and reactive power conveyed to users (measured by Smart Meters), gathered by concentrator and recognized by power meter. • Transformer Sensors data (TTEMP-Pilot2b): The data is collected from eight temperature sensors installed in various parts of the transformers, two sensors for ambient temperature, humidity, and pressure, and one sensor for oil temperature. • Medium-voltage Network Analyzer (MVNA-Pilot2b): contains an Electrical Network analyzer used for current transformers.
