Papers with Code 아카이브

Representation Learning

태스크의 하위 분류와 논문·코드·평가·기법·데이터셋을 탐색하세요.

하위 태스크
14
평가표
5
보존된 평가 기록
11

Representation Learning is a process in machine learning where algorithms extract meaningful patterns from raw data to create representations that are easier to understand and process. These representations can be designed for interpretability, reveal hidden features, or be used for transfer learning. They are valuable across many fundamental machine learning tasks like image classification and retrieval.

Deep neural networks can be considered representation learning models that typically encode information which is projected into a different subspace. These representations are then usually passed on to a linear classifier to, for instance, train a classifier.

Representation learning can be divided into:

  • Supervised representation learning: learning representations on task A using annotated data and used to solve task B
  • Unsupervised representation learning: learning representations on a task in an unsupervised way (label-free data). These are then used to address downstream tasks and reducing the need for annotated data when learning news tasks. Powerful models like GPT and BERT leverage unsupervised representation learning to tackle language tasks.

More recently, self-supervised learning (SSL) is one of the main drivers behind unsupervised representation learning in fields like computer vision and NLP.

Here are some additional readings to go deeper on the task:

<span style="color:grey; opacity: 0.6">( Image credit: Visualizing and Understanding Convolutional Networks )</span>

데이터셋

출처에 같은 태스크 이름으로 연결된 데이터셋입니다. 동명 태스크가 포함될 수 있으므로, 비교할 때는 평가 결과의 조건도 확인하세요.

GeoCoV19

GeoCoV19 is a large-scale Twitter dataset containing more than 524 million multilingual tweets. The dataset contains around 378K geotagged tweets and 5.4 million tweets with Place information. The annotations include toponyms from the user location field and tweet content and resolve them to geolocations such as country, state, or city level. In this case, 297 million tweets are annotated with geolocation using the user location field and 452 million tweets using tweet content.

IS-A

The IS-A dataset is a dataset of relations extracted from a medical ontology. The different entities in the ontology are related by the “is a” relation. For example, ‘acute leukemia’ is a ‘leukemia’. The dataset has 294,693 nodes with 356,541 edges between them. Source: https://arxiv.org/pdf/1906.05939.pdf

OLPBENCH

OLPBENCH is a large Open Link Prediction benchmark, which was derived from the state-of-the-art Open Information Extraction corpus OPIEC (Gashteovski et al., 2019). OLPBENCH contains 30M open triples, 1M distinct open relations and 2.5M distinct mentions of approximately 800K entities. Open Link Prediction is defined as follows: Given an Open Knowledge Graph and a question consisting of an entity mention and an open relation, predict mentions as answers. A predicted mention is correct if it is a mention of the correct answer entity. For example, given the question (“NBC-TV”, “has office in”, ?), correct answers include “NYC” and “New York”. Source: OLPBENCH

WikiGraphs

WikiGraphs is a dataset of Wikipedia articles each paired with a knowledge graph, to facilitate the research in conditional text generation, graph generation and graph representation learning. Existing graph-text paired datasets typically contain small graphs and short text (1 or few sentences), thus limiting the capabilities of the models that can be learned on the data. WikiGraphs is collected by pairing each Wikipedia article from the established WikiText-103 benchmark with a subgraph from the Freebase knowledge graph. This makes it easy to benchmark against other state-of-the-art text generative models that are capable of generating long paragraphs of coherent text. Both the graphs and the text data are of significantly larger scale compared to prior graph-text paired datasets.

Animals-10

It contains about 28K medium quality animal images belonging to 10 categories: dog, cat, horse, spyder, butterfly, chicken, sheep, cow, squirrel, and elephant. All the images have been collected from "google images" and have been checked by humans. There is some erroneous data to simulate real conditions (eg. images taken by users of your app). The main directory is divided into folders, one for each category. The image count for each category varies from 2K to 5 K units.

Fitness-AQA

Largest, first-of-its-kind, in-the-wild, fine-grained workout/exercise posture analysis dataset, covering three different exercises: BackSquat, Barbell Row, and Overhead Press. Seven different types of exercise errors are covered. Unlabeled data is also provided to facilitate self-supervised learning.

Causal Triplet

Causal Triplet is a causal representation learning benchmark featuring not only visually more complex scenes, but also two crucial desiderata commonly overlooked in previous works: 1) An actionable counterfactual setting, where only certain object-level variables allow forcounterfactual observations whereas others do not. 2) An interventional downstream task with an emphasis on out-of-distribution robustness from the independent causal mechanisms principle. Source: Causal Triplet: An Open Challenge for Intervention-centric Causal Representation Learning Image Source: https://arxiv.org/pdf/2301.05169v1.pdf

KGRC-RDF-star

KGRC-RDF-star is an RDF-star dataset converted from KGRC-RDF, which is a Knowledge graph dataset of novel stories. RDF-star (also known as RDF*) introduces quoted triple (QT), which is a triple used as the subject or object of another triple. KGRC-RDF-star is a complex RDF-star graph dataset that contains nested structures of statements and scenes, e.g., "Person A said "Person B saw "Person C was in D" " ." The same s, p, and o combinations may occur in different scenes when the KGRC-RDF is converted to the KGRC-RDF-star. It is necessary to distinguish these QTs and assign different metadata to them. Therefore, we solved this problem by assigning a unique ID to each QT and nested these triples as a QT as follows: << << s p o >> id val >> p' o'.

xView3-SAR

Unsustainable fishing practices worldwide pose a major threat to marine resources and ecosystems. Identifying vessels that do not show up in conventional monitoring systems---known as ``dark vessels''---is key to managing and securing the health of marine environments. With the rise of satellite-based synthetic aperture radar (SAR) imaging and modern machine learning (ML), it is now possible to automate detection of dark vessels day or night, under all-weather conditions. SAR images, however, require a domain-specific treatment and are not widely accessible to the ML community. Maritime objects (vessels and offshore infrastructure) are relatively small and sparse, challenging traditional computer vision approaches. We present the largest labeled dataset for training ML models to detect and characterize vessels and ocean structures in SAR imagery. xView3-SAR consists of nearly 1,000 analysis-ready SAR images from the Sentinel-1 mission that are, on average, 29,400-by-24,400 pixels each. The images are annotated using a combination of automated and manual analysis. Co-located bathymetry and wind state rasters accompany every SAR image. We also provide an overview of the xView3 Computer Vision Challenge, an international competition using xView3-SAR for ship detection and characterization at large scale. We release the data (https://iuu.xview.us/) and code (https://github.com/DIUx-xView) to support ongoing development and evaluation of ML approaches for this important application.

BIRDeep

The BIRDeep Audio Annotations dataset is a collection of bird vocalizations from Doñana National Park, Spain. It was created as part of the BIRDeep project, which aims to optimize the detection and classification of bird species in audio recordings using deep learning techniques. The dataset is intended for use in training and evaluating models for bird vocalization detection and identification.

Claim Matching Robustness

  • An evaluation test bed for assessing the robustness of sentence embedding models against user-informed misinformation edits. - A dataset containing perturbed and unperturbed claim pairs used for improving embedding model robustness through knowledge distillation.

Coil100-Augmented

This dataset derives from Coil100. There are more than 1,1M images of 100 objects. Each object was turned on a turnable through 360 degrees to vary object pose with respect to a fixed color camera. Images of the objects were taken at pose intervals of 5 degrees. This corresponds to 72 poses per object. Then planar rotation (9 angles) and 18 scaling factors has been applied. Objects have a wide variety of complex geometric and reflectance characteristics. This augmented version of Coil100 has been design for Disentangled Representation Learning for real images, the Factors of Variations are: | Factors | # values | |----------|----------| | Object | 100 | | 3D Pose | 72 | | Rotation | 9 | | Scale | 18 | The binarized version is also available.

COSTRA 1.0

COSTRA 1.0 is a dataset of complex sentence transformations. The dataset is intended for the study of sentence-level embeddings beyond simple word alternations or standard paraphrasing. The first version of the dataset is limited to sentences in Czech but the construction method is universal and the authors plan to use it also for other languages. The dataset consist of 4,262 unique sentences with average length of 10 words, illustrating 15 types of modifications such as simplification, generalization, or formal and informal language variation. Source: COSTRA 1.0: A Dataset of Complex Sentence Transformations

DAPFAM

Dataset DAPFAM See the accompanying paper: Ayaou et al., 2025 — “DAPFAM: A Domain‑Aware Patent Retrieval Dataset Aggregated at the Family Level” (arXiv:2506.22141). ## Summary DAPFAM provides 1 247 balanced query patent families and 45 336 target families with forward/backward‑citation relevance labels (≈ 50 K pairs). Each relevant link is explicitly marked in‑domain or out‑of‑domain according to IPC 3‑char overlap, enabling rigorous cross‑domain evaluation. * Full text (title · abstract · claims · description) plus rich metadata for every family. * Multi‑jurisdictional, English‑only text (families may originate in US, JP, EP, CN, …). * Parquet qrel file: qrels_all.parquet. ## Dataset Structure corpus.parquet # 45 336 rows, targets – every original column from the paper queries.parquet # 1 247 rows, queries – same columns + abstract_keywords qrels_all.parquet # (all | in | out) four‑column tables → query_id · relevant_id · relevance_score · domain_rel ## Citation @misc{ayaou2025dapfam, title={DAPFAM: A Domain-Aware Patent Retrieval Dataset Aggregated at the Family Level}, author={Iliass Ayaou and Denis Cavallucci and Hicham Chibane}, year={2025}, eprint={2506.22141}, archivePrefix={arXiv}, primaryClass={cs.CL} } ## Quick Stats * Queries: 1,247 * Corpus (targets): 45,336 * Qrels (all): 49,869 * Qrels (in): 19,736 * Qrels (out): 5,193

French CASS dataset

Composed of judgments from the French Court of cassation and their corresponding summaries. Source: STRASS: A Light and Effective Method for Extractive Summarization Based on Sentence Embeddings

GloREPlus

A distant supervision dataset by linking the entire English ClueWeb09 corpus to Freebase. Source: Global Textual Relation Embedding for Relational Understanding

Myket Android Application Install

This dataset contains information on application install interactions of users in the Myket android application market. The dataset was created for the purpose of evaluating interaction prediction models, requiring user and item identifiers along with timestamps of the interactions. Hence, the dataset can be used for interaction prediction and building a recommendation system. Furthermore, the data forms a dynamic network of interactions, and we can also perform network representation learning on the nodes in the network, which are users and applications. ## Data Creation The dataset was initially generated by the Myket data team, and later cleaned and subsampled by Erfan Loghmani a master student at Sharif University of Technology at the time. The data team focused on a two-week period and randomly sampled 1/3 of the users with interactions during that period. They then selected install and update interactions for three months before and after the two-week period, resulting in interactions spanning about 6 months and two weeks. We further subsampled and cleaned the data to focus on application download interactions. We identified the top 8000 most installed applications and selected interactions related to them. We retained users with more than 32 interactions, resulting in 280,391 users. From this group, we randomly selected 10,000 users, and the data was filtered to include only interactions for these users. The detailed procedure can be found in here. ## Data Structure The dataset has two main files. - myket.csv: This file contains the interaction information and follows the same format as the datasets used in the "JODIE: Predicting Dynamic Embedding Trajectory in Temporal Interaction Networks" (ACM SIGKDD 2019) project. However, this data does not contain state labels and interaction features, resulting in associated columns being all zero. - app_info_sample.csv: This file comprises features associated with applications present in the sample. For each individual application, information such as the approximate number of installs, average rating, count of ratings, and category are included. These features provide insights into the applications present in the dataset. ## Dataset Details - Total Instances: 694,121 install interaction instances - Instances Format: Triplets of user_id, app_name, timestamp - 10,000 users and 7,988 android applications - Item features for 7,606 applications For a detailed summary of the data's statistics, including information on users, applications, and interactions, please refer to the Python notebook available at summary-stats.ipynb. The notebook provides an overview of the dataset's characteristics and can be helpful for understanding the data's structure before using it for research or analysis. ### Top 20 Most Installed Applications | Package Name | Count of Interactions | | ---------------------------------- | --------------------- | | com.instagram.android | 15292 | | ir.resaneh1.iptv | 12143 | | com.tencent.ig | 7919 | | com.ForgeGames.SpecialForcesGroup2 | 7797 | | ir.nomogame.ClutchGame | 6193 | | com.dts.freefireth | 6041 | | com.whatsapp | 5876 | | com.supercell.clashofclans | 5817 | | com.mojang.minecraftpe | 5649 | | com.lenovo.anyshare.gps | 5076 | | ir.medu.shad | 4673 | | com.firsttouchgames.dls3 | 4641 | | com.activision.callofduty.shooter | 4357 | | com.tencent.iglite | 4126 | | com.aparat | 3598 | | com.kiloo.subwaysurf | 3135 | | com.supercell.clashroyale | 2793 | | co.palang.QuizOfKings | 2589 | | com.nazdika.app | 2436 | | com.digikala | 2413 | ## Comparison with SNAP Datasets The Myket dataset introduced in this repository exhibits distinct characteristics compared to the real-world datasets used by the project. The table below provides a comparative overview of the key dataset characteristics: | Dataset | #Users | #Items | #Interactions | Average Interactions per User | Average Unique Items per User | | --------------- | ---------------- | --------------- | ------------- | ----------------------------- | ----------------------------- | | Myket | 10,000 | 7,988 | 694,121 | 69.4 | 54.6 | | LastFM | 980 | 1,000 | 1,293,103 | 1,319.5 | 158.2 | | Reddit | 10,000 | 984 | 672,447 | 67.2 | 7.9 | | Wikipedia | 8,227 | 1,000 | 157,474 | 19.1 | 2.2 | | MOOC | 7,047 | 97 | 411,749 | 58.4 | 25.3 | The Myket dataset stands out by having an ample number of both users and items, highlighting its relevance for real-world, large-scale applications. Unlike LastFM, Reddit, and Wikipedia datasets, where users exhibit repetitive item interactions, the Myket dataset contains a comparatively lower amount of repetitive interactions. This unique characteristic reflects the diverse nature of user behaviors in the Android application market environment. ## Citation If you use this dataset in your research, please cite the following preprint: @misc{loghmani2023effect, title={Effect of Choosing Loss Function when Using T-batching for Representation Learning on Dynamic Networks}, author={Erfan Loghmani and MohammadAmin Fazli}, year={2023}, eprint={2308.06862}, archivePrefix={arXiv}, primaryClass={cs.LG} }

Natural Sprites

This csv consists of (x-position, y-position, area) tuples of three views (left, middle, right) of downscaled binary masks with aspect ratio kept (64 x 128) from the 2019 YouTube-VIS challenge, which can be found at https://competitions.codalab.org/competitions/20127#participate-get-data. Extracting pairs from this csv results in 234,652 transitions in the given statistics. These statistics can be used to augment ground truth factor distributions with natural transitions, which we demonstrate with spriteworld. For details, we refer to our paper, which can be found at https://openreview.net/forum?id=EbIDjBynYJ8.

PART-OF

The PART-OF dataset is a dataset of relations extracted from a medical ontology. The different entities in the ontology are parts of the human body. The dataset has 16,894 nodes with 19,436 edges between them. Source: https://arxiv.org/pdf/1906.05939.pdf

TYC Dataset

We introduce the trapped yeast cell (TYC) dataset, a novel dataset for understanding instance-level semantics and motions of cells in microstructures. We release $105$ dense annotated high-resolution brightfield microscopy images, including about $19$k instance masks. We also release $261$ curated video clips composed of $1293$ high-resolution microscopy images to facilitate unsupervised understanding of cell motions and morphology.

XYSquares

Synthetic dataset intended for benchmarking disentanglement frameworks. XYSquares is adversarial in nature; the distance between any two observations in the dataset is constant when measured using a pixel-wise distance function. It is usually impossible for VAE frameworks that use pixel-wise losses to disentangle this dataset. The dataset is constructed from 3 non-overlapping red, green and blue squares that are each $8\times8$ pixels in size. Each square can move along the $x$ and $y$ coordinates of an $8\times8$ grid. The resulting images are $64\times64$ pixels in size. With this construction the dataset has a total of 8 ground-truth factors for a total of $8^6 = 262144$ observations.

출처: Papers with Code · CC BY-SA 4.0. 아카이브 자료와 새로 수집한 논문은 별도로 관리합니다.