Papers with Code · CC BY-SA 4.0

데이터셋 탐색

데이터셋의 설명·홈페이지를 찾고, 관련 태스크의 평가 기록을 살펴보세요.

관련 논문 2편

Darpa Operationally Transparent Cyber (OpTC) Dataset

Darpa OpTC
Darpa OpTC 설명 보기

Operationally Transparent Cyber (OpTC) was a technology transition pilot study funded under Boston Fusion Corp.'s Cyber APT Scenarios for Enterprise Systems (CASES) project. Its primary objective was to determine if DARPA Transparent Computing (TC) program technologies could scale without loss of detection performance to address cyber defense capability gaps identified in USTRANSCOM's Joint Deployment Distribution Enterprise (JDDE) solicitation for the government fiscal years 2019-2023. Boston Fusion along with two performers from the TC program (Five Directions providing endpoint telemetry (TA1) and BAE providing analysis over the data (TA2)) worked to scale their systems from two machines to one thousand machines. The OpTC team conducted scaling and detection tests in the fall of 2019. A third performer (Provatek), not originally associated with the TC program, acted as a red team and test coordinator. This data set represents a subset of that activity. The OpTC system architecture is based on one used in TC program evaluations. Kafka, an open-source stream-processing server, is used to pass information among system components. Each Windows 10 endpoint is equipped with an endpoint sensor that monitors host events, packs them into JSON records, and sends them to Kafka. As these records flow into Kafka, a translation server aggregates them into new data records in a format called eCAR that are then pushed back to Kafka. As the translation server pushes eCAR records to Kafka, a data analytics component integrates them into a graph data structure for analysis and visualization. OpTC took TC system components that worked well on two hosts in TC program tests and scaled them up to work with one thousand hosts. This scaled-up system was evaluated over two weeks in a highly instrumented environment, and the data in this collection contains approximately a terabyte of data in compressed JSON compatible format from that evaluation. The evaluation started with a period of benign record generation, followed by the injection of malware by a red team. Benign traffic ran continuously during red team activity. Due to constraints in collection data space during the evaluation, data from five hundred hosts were collected rather than from the full set of one-thousand hosts.

관련 논문 1편

CPTC-2018

CPTC-2018 설명 보기

Intrusion alert dataset captured through the Collegiate Penetration Testing Competition (CPTC) 2018. Contains alerts from 6 student teams. For details, see "A Cybersecurity Dataset Derived from the National Collegiate Penetration Testing Competition" by Nathan Munaiah et al.

관련 논문 1편

PowerPoint Task Completion (PPTC)

PPTC
PPTC 설명 보기

Recent evaluations of Large Language Models (LLMs) have centered around testing their zero-shot/few-shot capabilities for basic natural language tasks and their ability to translate instructions into tool APIs. However, the evaluation of LLMs utilizing complex tools to finish multi-turn, multi-modal instructions in a complex multi-modal environment has not been investigated. To address this gap, we introduce the PowerPoint Task Completion (PPTC) benchmark to assess LLMs' ability to create and edit PPT files based on user instructions. It contains 279 multi-turn sessions covering diverse topics and hundreds of instructions involving multi-modal operations.

관련 논문 1편

ShopTC-100K

ShopTC-100K 설명 보기

ShopTC-100K Dataset The ShopTC-100K dataset is collected using TermMiner, an open-source data collection and topic modeling pipeline introduced in the paper: Harmful Terms and Where to Find Them: Measuring and Modeling Unfavorable Financial Terms and Conditions in Shopping Websites at Scale If you find this dataset or the related paper useful for your research, please cite our paper: @inproceedings{tsai2025harmful, author = {Elisa Tsai and Neal Mangaokar and Boyuan Zheng and Haizhong Zheng and Atul Prakash}, title = {Harmful Terms and Where to Find Them: Measuring and Modeling Unfavorable Financial Terms and Conditions in Shopping Websites at Scale}, booktitle = {Proceedings of the ACM Web Conference 2025 (WWW ’25)}, year = {2025}, location = {Sydney, NSW, Australia}, publisher = {ACM}, address = {New York, NY, USA}, pages = {14}, month = {April 28-May 2}, doi = {10.1145/3696410.3714573} } ## Dataset Description The dataset consists of sanitized terms extracted from e-commerce websites with English terms and conditions. The websites were sourced from the Tranco list (as of April 2024).

관련 논문 0편

5Char CAPTCHA Dataset

5Char CAPTCHA Dataset 설명 보기

Description: <a href="https://gts.ai/dataset-download/underpass-image-dataset/" target="_blank">👉 Download the dataset here</a> The 5Char CAPTCHA Dataset is specifically curated to facilitate the training and testing of machine learning models for CAPTCHA recognition. CAPTCHA (Completely Automated Public Turing test to tell Computers and Humans Apart) images are a widely used method to prevent bots from interacting with web services by presenting characters or digits that humans can easily recognize but are difficult for automated systems to decode. This dataset includes a collection of CAPTCHA images that consist of five alphanumeric characters, making it ideal for developing models aimed at breaking or solving CAPTCHA challenges. Dataset Overview: Total Images: 1000 CAPTCHA images. Image Format: PNG. Image Resolution: Each image has a resolution of 180×50 pixels. Character Count: Every CAPTCHA image contains exactly 5 characters, which could be any combination of uppercase letters (A-Z) and digits (0-9). Download Dataset Data Structure: The dataset is structured into a folder named “captcha_dataset,” which contains 1000 CAPTCHA images. Each image is uniquely named according to the characters it contains. For example, an image with the filename “AB123.png” corresponds to a CAPTCHA where the string “AB123” is displayed. This naming convention makes it easy to extract labels directly from the filenames without needing a separate annotation file. Potential Applications: Security Testing: Develop CAPTCHA-solving AI to assess the robustness of CAPTCHA-based security systems. OCR Enhancement: Use the dataset to improve the accuracy of OCR technologies in recognizing distorted or noisy text in real-world applications. Web Automation: Implement automated CAPTCHA-solving bots for various web-based tasks like form submissions, web scraping, or accessing restricted content. CAPTCHA Generation: Train AI to generate new CAPTCHA variations to further enhance security or develop new CAPTCHA systems for websites. Conclusion: The 5Char CAPTCHA Dataset is an excellent resource for those working on CAPTCHA recognition, security solutions, and OCR technologies. It presents a range of challenges that can help push the boundaries of existing models, providing an opportunity to improve machine learning techniques in image processing, particularly for security-based applications. This dataset is sourced from Kaggle.

관련 논문 0편

RECAPTCHA Dataset for MBBank

RECAPTCHA Dataset for MBBank 설명 보기

Description: <a href="https://gts.ai/dataset-download/recaptcha-dataset-for-mbbank/" target="_blank">👉 Download the dataset here</a> In the ever-evolving landscape of digital banking, security is paramount. MB Bank employs reCAPTCHA technology to protect its users from automated bots, ensuring the integrity of online banking systems. This dataset is a comprehensive collection of CAPTCHA images specifically used by MB Bank, designed to bolster research into CAPTCHA-solving techniques. Dataset Overview The dataset consists of various CAPTCHA formats used in real-world scenarios. It includes traditional text-based CAPTCHAs, image-based puzzles, and more complex variants that challenge machine learning models while remaining user-friendly. Download Dataset Importance for Developers and Researchers For developers, the dataset provides an opportunity to train models on CAPTCHA recognition, improving system robustness against bot-driven attacks. It helps in building more accurate, adaptive algorithms that can distinguish between human users and bots. Researchers can also use this dataset to explore new CAPTCHA-solving techniques, increasing the success rate of CAPTCHA bypassing in real-time applications. Applications in Security The dataset can be used to analyze trends in CAPTCHA security breaches and to develop new methods to safeguard digital platforms. By training on this dataset, developers can create more secure, user-centric CAPTCHA systems that deter fraud while ensuring smooth user experiences. CAPTCHA Variability and Adaptability This dataset is ideal for testing algorithms against a wide range of CAPTCHA formats. The variability ensures that AI models are versatile enough to handle different challenges. Moreover, the dataset’s rich annotations provide the necessary information to enhance accuracy in CAPTCHA recognition. Future Potential The MB Bank reCAPTCHA dataset is not just a valuable resource for improving current security systems but also paves the way for future innovations in online security. As bots become more sophisticated, the need for adaptive and robust CAPTCHA systems grows, making this dataset a critical tool for continuous improvement. This dataset is sourced from Kaggle.

관련 논문 0편

Vulnerable Captchas

Vulnerable Captchas 설명 보기

Description: <a href="https://gts.ai/dataset-download/vulnerable-captchas/" target="_blank">👉 Download the dataset here</a> This dataset focuses on an interesting example of weak CAPTCHA implementations, highlighting potential security vulnerabilities in systems that rely on simple alphanumeric captchas. CAPTCHAs (Completely Automated Public Turing Test to Tell Computers and Humans Apart) are widely used to protect websites from bots and automated scripts. However, not all CAPTCHA implementations are equally secure, and some are prone to exploitation through automated processes. Download Dataset Context The inspiration for this dataset came from a personal experience while accessing a website I frequently use, which I will refer to as “System” for privacy reasons. I wanted to automate a repetitive task on the site using a Python script, but I was initially blocked by a CAPTCHA that was required to complete the login process. CAPTCHAs are generally effective in stopping bots, especially those like Google’s reCAPTCHA, which are difficult to bypass with machine learning models due to their sophisticated design. However, in this case, the CAPTCHA images were simple enough for human eyes to decipher, consisting only of clearly readable alphanumeric characters. The challenge intrigued me, and as I was simultaneously reading “Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow” by Aurélien Géron, I decided to use this scenario as an opportunity to apply my newly acquired knowledge in machine learning. Problem and Approach The dataset captures images of these vulnerable CAPTCHA challenges and provides annotations for each. During the process of automating the CAPTCHA resolution, I learned that the system did not just rely on the image itself. Upon inspection of the HTML, I found that the CAPTCHA content was hashed and stored inside a hidden form field. Which could easily be manipulate to bypass the verification entirely. Key Learnings CAPTCHA Design Matters: Not all CAPTCHAs are created equal. Simpler alphanumeric CAPTCHAs can often be defeated by image recognition models or form manipulation. Image Classification: This dataset offers a collection of label CAPTCHA images. That could be use to train image classification models aimed at recognizing and solving CAPTCHAs automatically. Security Implications: The project sheds light on the importance of implementing proper security mechanisms beyond just CAPTCHA images. Such as encryption, hashing, and verification strategies that prevent easy manipulation. Practical Approach: Sometimes, simpler solutions such as analyzing the webpage structure and finding security loopholes can be more efficient than complex machine learning models. This dataset is sourced from Kaggle.