The Cortex, An AI Research OS
date
May 8, 2023
slug
AI-Research-OS
status
Published
tags
Research
summary
“The Cortex, An AI Research OS”, a groundbreaking mesh of Standard Operating Procedures and playbooks to conducting AI research in a way that’s accessible, efficient, and inspiring.
type
Post
I’m delighted to introduce you to “The Cortex, An AI Research OS”, a groundbreaking mesh of Standard Operating Procedures and playbooks to conducting AI research in a way that’s accessible, efficient, and inspiring.
As someone who has always been passionate about the pursuit of knowledge and the wonders of science, I believe that democratizing the tools and techniques for AI research is essential for accelerating innovation and unlocking our full potential.
The Cortex is an easily reproducible system designed to streamline AI research, allowing curious minds to replicate cutting-edge studies and conduct their own experiments with ease.
By removing barriers and simplifying the process, we’re empowering individuals to create state-of-the-art neural networks in the comfort of their own workspaces, regardless of their background or experience level.
In my own quest for understanding, I’ve always emphasized the importance of curiosity, exploration, and continuous learning.
The Cortex embodies these principles, fostering an environment where researchers can ask questions, investigate new ideas, and seek a deeper understanding of the world around them.
The true power of The Cortex lies in its ability to accelerate the rate of iteration.
As I often say, “the number of iterations x time and progress between iterations = innovation.”
By enabling researchers to rapidly iterate and refine their ideas, we can compress billions of years of evolution into mere months, weeks, or even days, thereby propelling the field of artificial intelligence to unprecedented heights.
In this article, we’ll explore the inner workings of The Cortex, An AI Research OS, and discover how it’s revolutionizing the way we approach AI research.
Join me on this fascinating journey, and let’s embrace the spirit of innovation and intellectual humility that has always driven human progress.
Now, The Cortex will be updated indefinitely but will follow the structure of the traditional scientific method, observation → hypothesis → experiment procedure → experiment → analysis → report discoveries.
[Epoch1] Idea Generation:
The first step is discovering and generating innovative ideas that have the potential to advance the field of AI research.
To generate valuable ideas, focus on understanding the existing state of the art, identifying gaps in the knowledge, and formulating questions that drive further research.
Look for interdisciplinary connections and seek inspiration from other fields to broaden the scope of ideas. Be open to challenging existing assumptions and paradigms.
The Most FruitFul Areas now are:
- Multi-Modality [Models that can understand, perceive and reason in context in multiple modalities at a post-Human rate]
- Large Language Models + Reinforcement Learning: Aligning Language Models with Rewards
- And, Embodied Agents [Agents with the ability to perceive the environment, take actions, and receieve rewards]
3 action steps:
- Write 10 experiments you wish to conduct, for example: Integrating Flash Attention with a base transformer and evaluating it on GLUE.
- Regularly review AI research papers and repositories from sources like arXiv, NeurIPS, ICLR, and GitHub, AK
- Attend AI conferences, webinars, and workshops to stay updated on the latest developments.
- Collaborate with researchers, engineers, and domain experts to brainstorm and generate new ideas.
- Open source your ideas as much as possible
Resources: arXiv, NeurIPS, ICLR, GitHub, AI conferences, webinars, workshops, research groups, interdisciplinary collaboration platforms, https://paperswithcode.com
[Epoch2] Assess Research Implications:
Process: Evaluating the potential impact, feasibility, and relevance of generated ideas in the context of the broader AI research community.
Once you have generated ideas, assess their potential by considering the implications for AI research, the feasibility of implementation, and how they fit within the broader landscape.
This evaluation should take into account the societal and technological implications of each idea.
Be critical and objective in this assessment, as it will help you identify the most promising ideas to pursue.
And, remember to ask yourself and the community if your idea is even worth experimenting on, as a scientist do not cling to your ideas as painful as it may sound because at the end of the day the purpose of this research is not to experiment for the sake of experimenting but rather to advance Humanity and absolve some of the most ferocious problems that plague us on a daily basis.
3 action steps:
- For each paper or repository, identify 3 key implications that could advance AI research.
- Consider the potential impact on different AI subfields, such as natural language processing, computer vision, reinforcement learning, etc.
- Evaluate the novelty, feasibility, and relevance of the proposed ideas to the broader AI community.
List of resources:
Agora:
Join the Agora Discord Server!Agora advances Humanity with Multi-Modality AI Research | 142 membersdiscord.gg
The Agora discord is a safe space where all forms of AI researchers can come together to indulge in revolutionary research
Some others include AI research papers, repositories, domain experts, AI subfield knowledge, ethical guidelines, societal impact assessments, Yannic’s Discord, AK’s discord
- Miro: A collaborative online whiteboard platform for brainstorming and discussing research implications.
- Zotero: A free, open-source reference management software for organizing and citing research.
- Notion: An all-in-one workspace for note-taking, project management, and collaboration.
[Epoch3] Establish a Collaborative Workflow:
Set up an efficient and collaborative environment to facilitate teamwork, idea exchange, and the research process.
Establish a collaborative workflow by using tools and platforms that promote seamless communication, code sharing, and project management.
Clearly define roles and responsibilities for yourself or for each team member and create a shared repository to store research materials, code, and data.
3 action steps:
- Set up a collaboration platform (e.g., Jupyter Notebook, Google Colab) to facilitate teamwork and idea exchange.
- Define clear roles and responsibilities for yourself or team members to streamline the research process.
- Create a shared repository to store code, data, and other research-related materials.
List of resources:
- Jupyter Notebook: An open-source web application for creating and sharing live code, equations, visualizations, and narrative text.
- Google Colab: A cloud-based Jupyter notebook environment that allows for collaborative coding, experimentation, and model training.
- GitHub: A web-based platform for version control and collaboration, allowing researchers to share code, track changes, and collaborate on projects.
[Epoch4] Conduct a Literature Review:
Performing a comprehensive analysis of existing literature to identify related work and understand the state-of-the-art techniques.
Conduct a thorough literature review is essential to understand the current state of the art and identify gaps in knowledge.
Analyze the strengths and weaknesses of existing methods, and document the findings in a systematic manner with proper citations and references.
3 action steps:
- Perform a comprehensive literature review to identify related work and understand the state-of-the-art (SOTA) techniques.
- Analyze the strengths and weaknesses of existing methods to determine areas for improvement.
- Document the findings in a systematic manner, with proper citations and references and then open source them on github or a blog.
List of resources: Research databases, academic journals, conference proceedings, AI research papers, citation management tools, systematic review guidelines.
[Epoch5] Develop a Research Plan:
Formulating research questions, hypotheses, and designing experiments to test them based on the literature review and research implications.
Based on the findings from the literature review, formulate a list of clear research questions and hypotheses.
Design experiments to test these hypotheses, including data collection, preprocessing, model development, and evaluation.
Establish a timeline and milestones for the research project: [1 week or less]
3 action steps:
- Formulate list of research questions and hypotheses based on the literature review and research implications.
- Design experiments to test the hypotheses, including data collection, preprocessing, model development, and evaluation.
- Establish a timeline and milestones for the research project [less than 1 week]
List of resources:
- arXiv: A repository of electronic preprints for research papers in various fields, including AI and machine learning.
- Semantic Scholar: An AI-powered research tool for finding relevant academic papers and literature.
- Google Scholar: A search engine for scholarly literature across various disciplines and sources.
- Literature review findings, research questions and hypotheses, experimental design guidelines, project management tools, timeline templates, Linear App
- Miro: A collaborative online whiteboard platform for brainstorming and discussing research implications.
- Zotero: A free, open-source reference management software for organizing and citing research.
- Notion: An all-in-one workspace for note-taking, project management, and collaboration.
- Google Colab: A cloud-based Jupyter notebook environment that allows for collaborative coding, experimentation, and model training.
- PyTorch: An open-source machine learning library for Python, providing a flexible deep learning development platform.
- Trello: A visual tool for organizing and prioritizing projects and tasks using boards, lists, and cards.
- Asana: A web and mobile application for work management and tracking, allowing teams to organize, plan, and collaborate on projects.
- Basecamp: A project management and team collaboration tool that helps manage projects, tasks, and communication.
[Epoch6] Implement Experiments:
Developing and sharing code for data processing, model implementation, and evaluation using the collaborative workflow platform.
Implement experiments using the collaborative workflow platform to ensure seamless communication and sharing of code, data, and results among team members.
Regularly conduct code reviews and debugging sessions to maintain the quality of the research work. Document the experimental setup, results, and findings in detail.
3 action steps:
- Use a collaborative workflow platform like Google’s Collab or Github to develop and share code for data processing, model implementation, and evaluation.
- Regularly conduct code reviews and debugging sessions to ensure the quality of the research work.
- Document the experimental setup, results, and findings in detail.
List of resources:
- Collaborative workflow platform (e.g., Jupyter Notebook, Google Colab),
- code sharing tools (e.g., GitHub),
- code review guidelines,
- debugging tools,
- documentation templates
[Epoch7] Train Models on AWS EC2 Instances:
Now it’s time to train your model.
You can configure AWS EC2 instances or use Collab or paperspace notebooks for model training and implementing efficient training pipelines.
Set up AWS EC2 instances with the necessary hardware (e.g., GPUs) and software (e.g., TensorFlow, PyTorch) for model training. Implement efficient training pipelines to minimize computational costs and training time. Monitor the training progress and adjust hyperparameters or model architectures as needed.
3 action steps:
- Select training platform [AWS EC3, Google Collab, Papeerspace]
- Configure AWS EC2 instances with the necessary hardware (e.g., GPUs) and software (e.g., TensorFlow, PyTorch) for model training.
- Implement efficient training pipelines to minimize computational costs and training time.
- Monitor the training progress, and adjust hyperparameters or model architectures as needed weights and biases and tensorboard
List of resources: AWS EC2, GPU hardware, TensorFlow, PyTorch, training pipeline optimization techniques, hyperparameter tuning tools like optuna
[Epoch8] Evaluate and Iterate:
Evaluating model performance and refining the research process as needed.
Evaluate the performance of developed models using appropriate metrics and benchmark datasets: Like GLUE, common crawl, VQA
Compare the results with state-of-the-art methods to assess the significance of the research contribution.
Iterate the research process, refining the hypotheses, experiments, and models as needed.
3 action steps:
- Evaluate the performance of the developed models using appropriate metrics and benchmark datasets.
- Compare the results with SOTA methods to assess the significance of the research contribution.
- Iterate the research process, refining the hypotheses, experiments, and models as needed.
List of resources: Paperswithcode Evaluation metrics, benchmark datasets, performance analysis tools, iterative research process guidelines, model refinement techniques.
[Epoch9] Document and Share Results:
Preparing a research paper and sharing findings with the AI community.
Document the research process, methods, results, and conclusions in a research paper.
Share the research findings in conferences, journals, or preprint servers to receive feedback from the AI community.
Contribute the developed code and models to open-source repositories to facilitate further research and collaboration.
3 action steps:
- Prepare a research paper, detailing the research process, methods, results, and conclusions.
- Share the research findings in conferences, journals, or preprint servers to receive feedback from the AI community.
- Contribute the developed code and models to open-source repositories to facilitate further research and collaboration.
List of resources:
- Research paper templates, writing guidelines, citation management tools, conferences, journals, preprint servers (e.g., arXiv, bioRxiv), open-source repositories (e.g., GitHub, GitLab), collaboration platforms
- Sphinx: A tool for creating intelligent and beautiful documentation for Python projects.
- MkDocs: A static site generator geared towards building project documentation using Markdown.
- Read the Docs: A platform for building, hosting, and sharing documentation for open-source projects.
Join Agora:
The Cortex is brought to you by Agora, a community of brave Humans devoted to advancing Humanity.