AI research tools
1.0.0
- Recommend some scientific research tools that I like (some are limited to the AI field)
- The tools marked [To be tried] are tools that I have not used yet but feel are useful. I will try them as soon as possible and decide whether to keep them based on the experience.
- Please recommend to me through issues etc. scientific research tools in the AI field that you find useful but are not on this list. Thank you.
Generally speaking, when I need to find PDF files based on the title of a paper or other information, I will first use dblp (the search results are clear and no scientific Internet is required). For some special cases, I would use Google Scholar again.
dblp: computer science bibliography: A paper query website specially designed for CS, which includes top-notch papers that can be retrieved. You can search the author's papers according to conferences, journals, etc., which is useful when you want to search all articles in a computer conference.
Google scholar: In addition to searching for papers, you can view paper statistics and citation references here. You can also get new paper update reminders by following authors or papers, and use automated recommendations to provide a basic library.
Semantic scholar: Can be combined with external materials to conduct semantic analysis of the paper. Functions include: displaying citations and references, measuring paper influence, displaying paper charts, automatically generating keywords (based on titles), analyzing authors, finding additional resources on the Internet (for example, relevant YouTube videos), and recommending papers.
arXiv: A collection website for preprints of papers.
arXiv-sanity: It has great improvements over arXiv in terms of functionality, including displaying summaries, comments and very basic social, recommendation, and library functions during browsing. Search also works better.
Semantic Sanity: A Personalized Adaptive Feed: Create your own personalized arXiv reading feed. When creating each feed, you will be asked to select a few papers first, and then start to recommend based on these papers. You can like or dislike the recommended results to help adjust the recommended results.
Paper Digest – AI for tracking and summarizing papers: Provides a list of papers published the day before the email subscription, with a one-sentence summary of each paper. What is more important to me is that this website will also organize the papers of each top conference when they are released.
Papers With Code: Automatically connect papers to the GitHub repository and data sets that implement the code, and sort them according to GitHub collections. Show the SOTA on each task for comparison.
labmlai/annotated_deep_learning_paper_implementations: PyTorch implementations of many algorithms, with online annotations. The disadvantage is that the part shown is not implemented from scratch, and part of the code is placed in their own package.
Mendeley: I currently use this. Supports multiple platforms such as web, PC, Mac and mobile phones, can directly annotate and highlight PDFs, and has free cloud storage with a limit.
Zotero: It can directly capture the literature references in the literature database web page. It is often used in conjunction with Nut Cloud to achieve multi-end synchronization of literature management (supports MacOS, Windows, iPad, Linux, iOS).
Kimi Chat: ChatPGT-like product, everyone should be familiar with it.
Cool Papers: A website for writing papers developed by Mr. Su Jianlin. You can explore and use it yourself according to the README.
CopyTranslator: The biggest advantage is that it has functions such as pinning, clicking to copy, and monitoring the clipboard. It can be seamlessly switched when reading documents and is used together, which is very convenient.
Saladict Saladict: Browser translation plug-in, with very rich settings to match the user's usage habits. In addition to the official documents, you can also check out Shalaqi + Alfred to create the best document translation experience! Come and find out. I'm also exploring how to implement out-of-browser translation on Windows systems.
I used to use Evernote but have stopped it and am considering migrating.
Yuque: Very lightweight, I usually use the web version. Occasionally, some lists can be synchronized to the mobile phone, which is also very useful. New students can fill in my invitation code QPFTUN to receive a 30-day membership: kissing_heart:
Notion: If you need multi-person collaboration and project management, Notion may be more suitable.
the-incredible-pytorch: Various tutorials, projects, videos and other resources about PyTorch.
computervision-recipes: Produced by Microsoft, tutorials on various CV tasks based on PyTorch.
Pytorch-Project-Template: An extensible PyTorch project template, including examples of image segmentation, object classification, GANs, and reinforcement learning.
pytorch-template: Another PyTorch project template.
torchinfo: Prints PyTorch model information, including the number of parameters of each layer of the model, output tensor size, etc.
flops-counter.pytorch: Calculate the total FLOPs of the model (floating point operations, understood as the amount of calculation, which can be used to measure the complexity of the algorithm/model) and the proportion of each layer. The disadvantage is that it does not seem to support RNN related layers, and the printed information is not very convenient to read.
The latest version of PyTorch already comes with tensorboard. Official tutorial.
visdom: [To be tried] Flexible tool for creating, organizing and sharing real-time rich data visualizations.
Convolution Visualizer: If the convolution layer configuration is relatively complex and it is inconvenient to calculate the output size, you can use this visualization tool to assist.
Google Dataset Search
Data Search | Bifrost: Visual dataset search.
optuna: a framework for automatic hyperparameter optimization.
microsoft/nni: [To be tried] An open source automatic machine learning (AutoML) toolkit for neural model search and hyperparameter tuning, supporting most mainstream frameworks and operating environments.
Hyperopt: [To be tried] Distributed asynchronous hyperparameter optimization. I saw someone recommending it on Zhihu, but according to the documentation, there are only two optimization algorithms currently supported, and they do not include Bayesian optimization.
BoTorch: [To be tried] Bayesian optimization library based on PyTorch.
automl/Auto-PyTorch: [To be tried] Automatic structure search and hyperparameter search based on PyTorch.
Various LaTeX templates including papers, reports, posters, etc.
Templates from Overleaf
LaTeX Templates
For multi-person collaboration, I recommend Overleaf, which is also my choice most of the time. Single-person projects can use VSCode for offline LaTeX writing and cooperate with Github private libraries for version management.
You can take a look at the recommendations of this article: With these websites, writing English papers is no longer difficult (Introduction and usage tips of 15 English paper writing assistance websites) - Zhihu
Lingle: Search for the most frequently occurring English word collocations. Use when you are not sure whether your expression is correct.
Corpus of Contemporary American English (COCA): You can check the American English corpus of word collocations, and you can check the specific sentences using this word. British National Corpus (BYU-BNC): A corpus of British English, with less corpus than that of American English.
Thesaurus: Convert low-end words into synonymous high-end words.
ESODA: A phrase collocation query tool suitable for Chinese English writing produced by Tsinghua HCI Lab. A corpus of papers that can switch specific research directions, display relevant alternative usages, and support mixed searches in Chinese and English.
Words and phrases: frequency, genres, collocates, concordances, synonyms, and WordNet: Use different colors to distinguish high, medium and low-frequency words, display vocabulary that represents the type of article, and classify related replaceable words. Although it is said to be an English writing wording aid, I feel that its greatest use is to learn vocabulary and collocations commonly used in papers in related fields.
Some options besides the various ChatGPT-like ones. Please be aware of the leakage risks of online detection tools and handle key text with caution.
Grammarly: Check and modify grammar, sentence patterns, punctuation, and word selection, with browser plug-ins.
Nounplus.net: Free online English grammar checker.
Mathpix: Take screenshots of complex mathematical equations and convert them into LaTeX code. It can handle prints in PDFs and handwritten formulas in photos, etc.
MyScript Webdemo: The Math module can convert handwritten formulas into LaTeX code; at the same time, the Diagram module can convert hand-drawn block diagrams into neat block diagrams.
Detexify LaTeX handwritten symbol recognition: If you forget how to represent certain characters in LaTeX, you can query them by handwriting on this website.
OmniGraffle is available for Mac OS systems.
PPT is usually my first choice: it’s quick to get started, has many types, has a high degree of freedom, and supports exporting vector graphics.
Paper-Picture-Writing-Code: Drawing code based on LaTex, including line charts, histograms, scatter plots, attention visualization and structure charts.
academic-drawing: Matlab/Python drawing, mainly used to draw time series data.
awesome-latex-drawing: LaTeX drawing, mainly used to draw Bayesian networks, tensor decomposition, etc.
PlotNeuralNet: Python gets graphs that can be used in LaTeX, mainly drawing CNN.
There are many websites that provide conversion services for images in various formats. Here is just a brief list, you can also find other similar websites through search engines.
Convert PNG/JPEG (Raster) to EPS/PDF (Vector) Format: Convert image files in jpg, png format to eps files.
EPS to PDF converter: You can also convert EPS files to images in other formats.
Crop PDF files online - PDF Tools: Crop the white edges of pdf files.
TexLive comes with some command line tools:
epstopdf <file.eps> ;pdfcrop <file.pdf> . ccf-deadlines: You can filter conferences with deadlines this year based on research direction and CCF level. It is relatively friendly to domestic students.
AI Conference Deadlines: Conferences can be filtered based on research directions. But it seems that you need to surf the Internet scientifically to see all the information.
Conference List: Sorted according to deadline, expired conferences will not appear on the home page. There is a page where you can see which conferences are held in each research direction, but there is no way to filter out the conferences that have not expired based on the research direction.
Conference Partner: List of the latest international computer conferences and journals. You can register to follow a conference or journal. It is relatively complete, but the information is not updated in a timely manner.
For blind review purposes, sometimes links to files (such as source code) need to be anonymous. Some people choose to create an anonymous account on Github, but creating an account for every paper in every conference may be too cumbersome. I found some tools that support anonymous file sharing as follows.
Dropbox: Probably the most commonly used.
Open Science Framework
Figshare
Arxiv paper submission process - Just read this article to understand the process of submitting a paper to arXiv.
arxiv-latex-cleaner: Clean the LaTeX code of the paper to meet the requirements for submission on arXiv. One highlight is the ability to automatically clean up all commented out content in the paper.
overleaf -> arxiv Smooth submission process: If you are using Overleaf (rather than compiling LaTeX code into a paper locally), you can refer to this article to download the appropriate source code package first, and then consider using arxiv-latex-cleaner.
Providing clear, reproducible code for published papers can effectively advance the field. Here are some helpful tools for open source code.
ReproducibilityChecklist-v2.0: A machine learning reproducibility checklist that lists what documents you should provide to enhance the reproducibility of your paper.
pigar: A tool for automatically generating requirements files for Python projects.
Rainyscope rain simulator: rain sound.
LofiGirl's music study room: Lo-Fi music live broadcast room.
