Papers with Code: Complete Guide & Tutorial

Introduction

The field of machine learning and artificial intelligence evolves at a breathtaking pace. Every day, hundreds of new research papers are published, proposing novel architectures, training techniques, and state-of-the-art results. For researchers, students, and practitioners, keeping up with this flood of information while also finding reproducible code can be a daunting task. This is where Papers with Code becomes an indispensable tool.

Papers with Code (accessible at https://paperswithcode.com/) is a free, open platform that bridges the gap between academic research and practical implementation. Its core mission is simple but powerful: connect every machine learning paper with its corresponding code repository and performance benchmarks. Instead of reading a paper and then spending hours searching GitHub for an implementation, Papers with Code does the heavy lifting for you. It aggregates papers from arXiv, links them to their official code repositories (primarily GitHub), and tracks which methods achieve the best results on hundreds of benchmark datasets.

Whether you are a PhD student looking for the latest breakthrough in natural language processing, a data scientist trying to reproduce a model for a project, or a hobbyist exploring deep learning, Papers with Code provides a structured, community-driven way to discover, evaluate, and implement cutting-edge research. This tutorial will guide you through every aspect of the platform, from your first visit to advanced usage tips, ensuring you can leverage its full potential.

Getting Started

Getting started with Papers with Code requires no registration for basic browsing, though creating a free account unlocks additional features like upvoting and personalized recommendations. Here is how to begin:

1. Navigating to the Platform

Open your web browser and go to https://paperswithcode.com/. You will be greeted by a clean, modern interface. The homepage displays the current trending papers, a search bar at the top, and navigation links to different sections of the site.

2. Creating an Account (Optional but Recommended)

While you can browse papers and code without an account, signing up allows you to participate in the community. Click the “Sign Up” button in the top-right corner. You can register using your email address and a password, or you can use your Google or GitHub account for a faster login. Once registered, you can upvote papers, bookmark them, and receive personalized suggestions based on your interests.

3. Understanding the Homepage Layout

The homepage is divided into several key areas:

• Search Bar: Located at the top, this is your primary tool for finding specific papers, authors, or topics.
• Trending Papers Feed: The main content area shows a list of papers that are currently popular. By default, it shows the daily trending list, but you can switch to weekly or monthly views.
• Navigation Menu: Links to “Papers,” “Datasets,” “Methods,” “SOTA” (State-of-the-Art), and “Blog” are available in the header.
• Upvote Buttons: Each paper card has a small upward arrow. Clicking it (when logged in) upvotes the paper, helping the community surface the most valuable research.

Key Features

Papers with Code is packed with features that go far beyond a simple paper repository. Understanding these features is crucial to using the tool effectively.

Trending Papers Feed (Daily, Weekly, Monthly)

The trending feed is the heartbeat of the platform. It shows which papers are generating the most buzz in the community. You can filter the feed by three time windows:

• Daily: Shows papers that have gained the most upvotes and attention in the last 24 hours. This is perfect for catching breaking news and fresh preprints.
• Weekly: A broader view of the past seven days. This filter helps you identify papers that have sustained interest beyond the initial hype.
• Monthly: The most curated list, showing papers that have proven to be significant over a longer period. This is useful for discovering foundational work or papers with lasting impact.

Each paper in the feed displays its title, authors, a short abstract snippet, the number of upvotes, and badges indicating whether code is available and if it is linked to a benchmark result.

Links to Code Repositories and arXiv Papers

This is the core value proposition of Papers with Code. Every paper entry includes direct links to two critical resources:

• arXiv Link: A direct link to the full PDF of the paper on arXiv.org. This ensures you can always read the original research without paywalls.
• Code Repository Link: A prominent button labeled “Code” that takes you directly to the GitHub repository (or other code hosting sites) where the official implementation is hosted. The platform also indicates the programming language (e.g., Python) and the framework used (e.g., PyTorch, TensorFlow).

If a paper has multiple code implementations, Papers with Code lists them, often marking the official one with a badge. This saves you from the frustration of finding abandoned or unofficial forks.

Upvote System for Community Curation

The platform relies on a simple but effective upvote system. Registered users can upvote papers they find valuable. This crowd-sourced curation mechanism ensures that high-quality, reproducible, or groundbreaking research rises to the top of the trending feeds. The number of upvotes is prominently displayed next to each paper. You can also see which of your peers or followed users have upvoted a paper, adding a social discovery layer.

State-of-the-Art Benchmark Tracking

One of the most powerful features for researchers is the State-of-the-Art (SOTA) tracking. Papers with Code maintains a comprehensive database of benchmarks across hundreds of tasks, from image classification (e.g., ImageNet) to question answering (e.g., SQuAD). For each benchmark, the platform lists every method that has been evaluated, sorted by performance metric (e.g., accuracy, F1 score). You can see which paper currently holds the top spot, which methods are most recent, and how different approaches compare. This turns the platform into a live leaderboard for machine learning research.

Email Subscription for Trending Papers

To stay updated without constantly refreshing the website, Papers with Code offers a free email subscription service. You can subscribe to receive a daily or weekly digest of the most trending papers. The email includes a curated list of top papers with direct links to the paper and code. This is an excellent way to integrate research discovery into your morning routine. To subscribe, scroll to the bottom of any page and enter your email address in the “Subscribe to our newsletter” field.

How to Use

Now that you understand the features, let us walk through practical, step-by-step instructions for common tasks.

Finding a Specific Paper and Its Code

Suppose you heard about a paper called “Attention Is All You Need” and want to find its code.

1. Go to paperswithcode.com.
2. Type “Attention Is All You Need” into the search bar at the top and press Enter.
3. The search results will show the paper as the top result. Click on its title.
4. On the paper’s dedicated page, you will see the abstract, authors, and publication date. Below this information, look for the “Code” button. It will be labeled with the language (e.g., “Python”) and framework (e.g., “PyTorch”).
5. Click the “Code” button. You will be taken directly to the official GitHub repository where you can clone, fork, or download the implementation.
6. Next to the code button, you will also see an “arXiv” link. Click it to read the full paper PDF.

Discovering Trending Research in Your Field

To see what is hot in machine learning right now:

1. On the homepage, look at the trending feed. By default, it shows “Daily” trending papers.
2. To change the time window, click on the tabs labeled “Daily”, “Weekly”, or “Monthly” just above the list of papers.
3. Browse through the list. Each entry shows the paper title, a short description, and the number of upvotes.
4. If you find a paper interesting, click its title to go to its dedicated page. From there, you can read the abstract, access the code, and see its performance on benchmarks.
5. If you are logged in, click the upvote arrow (▲) next to a paper to recommend it to the community.

Comparing Methods on a Benchmark

Let us say you want to know which model performs best on the ImageNet dataset.

1. Click on “SOTA” (State-of-the-Art) in the top navigation menu.
2. You will see a list of tasks and datasets. Use the search bar on this page to type “ImageNet”.
3. Click on the “ImageNet” result (likely under “Image Classification”).
4. You will now see a leaderboard. It lists all methods that have been evaluated on ImageNet, sorted by top-1 accuracy (or another metric). The current best method is at the top.
5. For each method, you can see the paper title, the year it was published, the reported accuracy, and a link to the paper and code.
6. Click on any method to see more details, including the specific training setup and hyperparameters used.

Subscribing to Email Updates

1. Scroll down to the very bottom of any page on Papers with Code.
2. In the footer, look for the section labeled “Subscribe to our newsletter”.
3. Enter your email address in the text field.
4. Click the “Subscribe” button. You will receive a confirmation email. Click the link in that email to activate your subscription.
5. You can choose between a daily or weekly digest by adjusting your preferences (usually found in your account settings after logging in).

Using the Dataset Browser

Papers with Code also catalogs datasets. This is useful when you are looking for a specific benchmark to test your model.

1. Click on “Datasets” in the top navigation menu.
2. Browse the list or use the search bar to find a dataset, such as “COCO” (for object detection).
3. Click on the dataset name. You will see a page with a description, the number of papers that use it, and a leaderboard of the best results achieved on that dataset.
4. This page also provides links to download the dataset and a list of all papers that have reported results on it.

Tips

To get the most out of Papers with Code, keep these practical tips in mind:

Tip 1: Always Verify the Code License

While Papers with Code links to code repositories, the platform does not enforce licensing. Before using any code in your own projects, always check the GitHub repository for a LICENSE file. Some code is released under permissive licenses like MIT or Apache 2.0, while others may have restrictions for commercial use. Ignoring this can lead to legal issues.

Tip 2: Use the “Reproducibility” Badges

Some papers on the platform have a small badge indicating “Reproduced” or “Official.” Prioritize papers with the “Official” badge, as this means the code was released by the original authors. Papers marked as “Reproduced” are third-party implementations, which may have subtle differences from the original paper.

Tip 3: Combine Trending Feeds with SOTA Tracking

Do not rely solely on the trending feed. A paper might be trending because it is controversial or has a compelling title, but it may not be state-of-the-art. Use the SOTA section to verify whether a paper actually advances the field on a specific benchmark. The combination of trending discovery and SOTA verification gives you a complete picture.

Tip 4: Check the “Results” Tab on Paper Pages

When you open a specific paper, look for the “Results” tab. This tab shows exactly which datasets and benchmarks the paper evaluated on, along with the reported numbers. This is more reliable than trusting the abstract alone, as it provides granular performance data.

Tip 5: Follow Authors and Institutions

If you create an account, you can follow specific researchers or institutions. When you follow someone, their new papers and upvoted papers will appear more prominently in your feed. This turns Papers with Code into a personalized research assistant that learns your interests.

Tip 6: Use the API for Advanced Users

Papers with Code offers a free API (Application Programming Interface) for developers. If you want to programmatically search for papers, retrieve benchmark results, or build your own tools on top of the platform, visit the “API” link in the footer. The API is RESTful and well-documented, making it easy to integrate into your workflows.

Tip 7: Be Critical of Upvote Counts

The upvote system is useful, but it is not a perfect measure of quality. Papers from famous authors or labs often receive many upvotes simply due to name recognition. Conversely, excellent but niche papers might have fewer upvotes. Use upvotes as a signal, but always read the abstract and check the code before deciding if a paper is relevant to you.

Tip 8: Bookmark Papers for Later Reading

When logged in, you can click the bookmark icon (usually a small ribbon or flag) on any paper page. This saves the paper to your personal library. Over time, you can build a curated collection of papers to read during weekends or free time. You can access your bookmarks from your profile page.

Tip 9: Explore the “Methods” Section

The “Methods” tab in the navigation menu lists popular architectures and techniques (e.g., “Transformer,” “ResNet,” “GAN”). Clicking on a method shows you all papers that use or extend that method, along with the code repositories. This is a fantastic way to understand how a particular technique has evolved over time.

Tip 10: Use the Platform as a Literature Review Tool

If you are starting a new research project, use Papers with Code to perform a rapid literature review. Search for your topic, sort by “Most Cited” or “Highest Accuracy”, and then look at the code for the top 5-10 papers. This gives you a practical understanding of the current state of the art, including which frameworks and architectures are most common, saving you weeks of manual searching.

Papers with Code is more than just a website; it is a community-driven ecosystem that accelerates reproducible research. By mastering its features—from the trending feed and SOTA leaderboards to the email subscriptions and API—you can stay at the cutting edge of machine learning with minimal effort. Start exploring today, and let the code guide your research.