Model-released photos taken in similar environments. Professional lighting, cameras, and makeup. Diversity: demographics, ages, facial expressions, and poses.

Dataset Description: This dataset contains images of 192 different scene categories, with both AI-generated and real-world images for each class. It is designed for research and benchmarking in computer vision, deep learning, and AI-generated image detection.

Key Features: 📸 192 Scene Classes: Includes diverse environments like forests, cities, beaches, deserts, and more. 🤖 AI-Generated vs. Real Images: Each class contains images generated by AI models as well as real-world photographs. 🖼️ High-Quality Images: The dataset ensures a variety of resolutions and sources to improve model generalization. 🏆 Perfect for Research: Ideal for training models in AI-generated image detection, scene classification, and image authenticity verification. Potential Use Cases: 🔍 AI-generated vs. real image classification 🏙️ Scene recognition and segmentation 🖥️ Training deep learning models for synthetic image detection 📊 Analyzing AI image generation trends

and secure

Broad classes of statistical classification algorithms have beendeveloped and applied successfully to a wide range of real worlddomains. In general, ensuring that the particular classificationalgorithm matches the properties of the data is crucial inproviding results that meet the needs of the particular applicationdomain. One way in which the impact of this algorithm/applicationmatch can be alleviated is by using ensembles of classifiers, wherea variety of classifiers (either different types of classifiers ordifferent instantiations of the same classifier) are pooled before afinal classification decision is made. Intuitively, classifierensembles allow the different needs of a difficult problem to behandled by classifiers suited to those particular needs.Mathematically, classifier ensembles provide an extra degree offreedom in the classical bias/variance tradeoff, allowing solutionsthat would be difficult (if not impossible) to reach with only asingle classifier. Because of these advantages, classifier ensembles have been applied to many difficult real world problems. In this paper, we surveyselect applications of ensemble methods to problems that havehistorically been most representative of the difficulties inclassification. In particular, we survey applications of ensemblemethods to remote sensing, person recognition, one vs. allrecognition, and medicine.

SYNTHETIC-1

This is a subset of the task data used to construct SYNTHETIC-1. You can find the full collection here

This dataset is about books. It has 1 row and is filtered where the book is A real life elsewhere. It features 7 columns including author, publication date, language, and book publisher.

The anonymised data is related to the policies in force on December 31, 2009 in a portfolio of life-risk insurance policies corresponding to the year 2009 of a Spanish insurance company.The data sheet consists of 76,102 rows and 15 columns, with each row corresponding to a policy and each column to a variable

This dataset is originally created for the Knowledge Graph Reasoning Challenge for Social Issues (KGRC4SI) Video data that simulates daily life actions in a virtual space from Scenario Data. Knowledge graphs, and transcriptions of the Video Data content ("who" did what "action" with what "object," when and where, and the resulting "state" or "position" of the object). Knowledge Graph Embedding Data are created for reasoning based on machine learning

Humans can vividly simulate hypothetical experiences. This ability draws on our memories (e.g., of familiar people and locations) to construct imaginings that resemble real-life events (e.g., of meeting a person at a location). Here, we examine the hypothesis that we also learn from such simulated episodes much like from actual experiences. Specifically, we show that the mere simulation of meeting a familiar person (unconditioned stimulus; US) at a known location (conditioned stimulus; CS) changes how people value the location. We provide key evidence that this simulation-based learning strengthens pre-existing CS-US associations and that it leads to a transfer of valence from the US to the CS. The data thus highlight a mechanism by which we learn from simulated experiences.

Experiment design

The Daily Life Activities (DLA) dataset consists of trials of daily life object manipulation tasks performed by a human. The dataset consists of ten tasks: cutting, painting, pouring with cup, putting cup away, quarter turn, scooping and pouring, scooping food, shaking, sinusoidal motion, and table wiping. In this dataset, a high variation in the context was purposefully introduced. That is, the tasks were performed with respect to three different viewpoints (V1, V2, V3) and with four different execution styles (normal, with larger spatial scale, with different velocity profile, and with longer time duration). This resulted in a total of (3x4=12) twelve different contexts in which the tasks were performed. Each task was performed ten times in every context, resulting in a total of (10x3x4x10=1200) trials.

Experimental setup

The trials were recorded using a Krypton K600 camera from NIKON Metrology by tracking up to nine LED markers attached to the manipulated object. The 3D position of each LED marker was recorded with a sampling rate of 50 Hz and expected accuracy of 0.4mm with respect to the measurement frame of the camera system.

Data format

Every trial_xxx.mat file is a Matlab structure array. The trailing number xxx refers to the order in which the trials were performed. For every task:

trial_001.mat up to trial_040.mat were recorded in sensor viewpoint 1.

trial_001.mat up to trial_010.mat were executed with execution style: normal.

trial_011.mat up to trial_020.mat were executed with execution style: longer time duration.

trial_021.mat up to trial_030.mat were executed with execution style: larger spatial scale.

trial_031.mat up to trial_040.mat were executed with execution style: different velocity profile.

trial_041.mat up to trial_080.mat were recorded in sensor viewpoint 2.

trial_041.mat up to trial_050.mat were executed with execution style: normal.

trial_051.mat up to trial_060.mat were executed with execution style: longer time duration.

trial_061.mat up to trial_070.mat were executed with execution style: larger spatial scale.

trial_071.mat up to trial_080.mat were executed with execution style: different velocity profile.

trial_081.mat up to trial_120.mat were recorded in sensor viewpoint 3.

trial_081.mat up to trial_090.mat were executed with execution style: normal.

trial_091.mat up to trial_100.mat were executed with execution style: longer time duration.

trial_101.mat up to trial_110.mat were executed with execution style: larger spatial scale.

trial_111.mat up to trial_120.mat were executed with execution style: different velocity profile.

The structure array trial_xxx.mat has the following fields:

'number_of_timesamples': the total number of timesamples (N) for the recorded task,

'K6C_12250_3_x': a 4xN matrix containing the 3D position coordinates of the LED marker expressed in millimeters. The trailing number x in 'K6C_12250_3_x' refers to the LED number, which can range from 1 to 9.

In case the LED marker was visible, the first, second and third row contain the x-, y-, and z-coordinates of the marker, respectively. The fourth row contains the zero value in this case.

In case the LED marker was not visible, the first, second and third row contain zero values. The fourth row contains a non-zero value in this case.

Financial overview and grant giving statistics of Real Life Catholic

A real-world radio frequency (RF) fingerprinting dataset for commercial off-the-shelf (COTS) Bluetooth and WiFi emitters under challenging testbed setups is presented in this dataset. The chipsets within the devices (2 laptops and 8 commercial chips) are WiFi-Bluetooth combo transceivers. The emissions are captured with a National Instruments Ettus USRP X300 radio outfitted with a UBX160 daughterboard and a VERT2450 antenna. The receiver is tuned to record a 66.67 MHz bandwidth of the spectrum centered at the 2.414 GHz frequency.

Real World Data from a simulator for 1,000 people belonging to any of four behavioural groups: athletic, normal, unfit and feeble, simulated over 2 years and 3 months.

The dataset is organised into 4 CSV files:

• aggregations.csv contains the daily physiological data, including their short- & long-term averages and their trends. Every row corresponds to a day for a person.
• answers.csv contains the answers to questions. Each row corresponds to a question answered.
• activity_summaries.csv contains summaries of the detected activities. Each row corresponds to an activity someone performed.
• participants_features.csv contains the demographic info of each simulated person. This is static info, every row corresponding to one person. An ML engineer can combine information from any of these files to setup model learning datasets. This is done in the “Processed Synthetic RWD for binary modelling” and the “Processed Synthetic RWD for tristate modelling” datasets that are also shared here.

RLKWiC Add-on: Benchmarking Dataset for Entity Recommendation

This benchmark, built on top of the Real-Life Knowledge Work in Context (RLKWiC) dataset, is designed to evaluate context-based entity recommendation by simulating a scenario where participants receive entities extracted from their activities across their defined contexts.

In total, 1850 entity recommendations were generated across 56 contexts. After deduplication, these entities were presented to participants for explicit relevance assessment on a 3-point scale:

• 0 [Irrelevant]: Signifying a lack of relevance between the recommended entity and the context.
• 1 [Relevant] Denoting a connection between the entity and the context, although it may not fully represent it.
• 2 [Representative]: The entity closely aligns with the context, indicating a high level of relevance where the context can be inferred to be about this entity.

Participants could also suggest additional relevant entities. The resulting dataset comprises 1067 entities with explicit relevance scores, offering a resource for benchmarking entity recommendation in real-life knowledge work.

Paper: Context-based Entity Recommendation for Knowledge Workers: Establishing a Benchmark on Real-life Data

According to an online survey conducted in the United Kingdom in November 2021, when asked to think about a recently enjoyable time, over seven in 10 reported having enjoyed playing in non-digital contexts. Furthermore, girls agreed more strongly than boys that they had a great time playing in real life. By comparison, 45 percent of children strongly agreed that they recently had a great time being playful in a digital setting. Overall, 50 percent of girls reported having had a great time on TikTok, whilst boys reporting to feel the same were over four in 10.

The real-world misleading visualization QA dataset accompanies the paper "'Protecting multimodal large language models againts misleading visualizations". The dataset contains 42 multiple-choice QA pairs, and the URLs to the 42 corresponding chart images used to answer the questions. The dataset is made available under a CC-BY-SA-4.0 license. Please cite our paper if you find this dataset useful to your work.

This dataset is about book subjects. It has 4 rows and is filtered where the books is Real-world machine learning. It features 10 columns including number of authors, number of books, earliest publication date, and latest publication date.

The best results corresponding to each metric for individual networks are highlighted.

This event log consists of the interpreted location data of patients within a hospital facility. This real-life event log is gathered by using Real-time Location Systems (RTLS).

Event-logs and Business Process Simulation Models used in the experimentation of the paper "Enhancing Business Process Simulation Models with Extraneous Activity Delays", where the 'inputs' folder contains all the files used as input, and the 'output' folder the results of the evaluation.

Inputs: event-logs, BPS models, and simulation parameters used as input in the experimentation.

Real-life: real-life event logs, corresponding to two disjoint subsets of traces from an Academic Credentials' process, and the BPIC 2012 and BPIC 2017 event logs (filtered as explained in the paper), and the BPS model (plus simulation parameters) used as input for each dataset in the presented approach.

Synthetic: simulated event-logs and corresponding BPS models (plus simulation parameters) for four different processes with 0, 1, 3 and 5 timer events.

Outputs: results of the experimentation.

Real-life: results corresponding to the evaluation with real-life event logs. Each of the folders is composed by the original and the enhanced BPS models, 10 event logs simulated with each of them, two folders with the best iteration of the two hyperparameter optimization processes, and the values for the injected timers in each case. In addition, a CSV file with the EMD metrics (cycle time and absolute hour event distribution) for each dataset is provided.

Synthetic: results corresponding to the simulated event-logs.

Before-After: BPS models and discovered timer events for the four synthetic processes, with five timers placed before and after different activity instances.

Complete: BPS models and quality measures (precision, recall, and SMAPE of the discovered timers) for the four synthetic processes with zero, one, three, and five timer events.

Individual: event logs enhanced with the discovered extraneous delay for each activity instance, for the four synthetic processes with zero, one, three, and five timer events; and SMAPE of the estimations.