Publishing a research paper at a top AI conference is one of the most impactful things you can do as a student interested in machine learning. It signals to graduate admissions committees, internship reviewers, and future collaborators that you can identify a real problem, design experiments, and communicate results at a professional level.
NeurIPS (the Conference on Neural Information Processing Systems) is among the most prestigious venues in artificial intelligence and machine learning. The main conference accepted roughly 25.8% of over 15,600 submissions in 2024 -- and those numbers get more competitive each year. But here is the part most students miss: you do not need to publish at the main conference to get a NeurIPS publication. Workshops are the realistic and highly respected entry point, and they are where the vast majority of student-led research gets presented.
This guide walks you through the entire process -- from choosing a research topic to standing at your poster in the convention center -- based on patterns we have seen work repeatedly for students at every level, from high schoolers to PhD candidates.
Understanding NeurIPS: Main Conference vs. Workshops
Before you start writing, you need to understand how NeurIPS is structured, because the path you take matters.
The Main Conference Track
The main conference accepts full papers (up to nine content pages) through an extremely competitive review process. Submissions go through double-blind review by multiple reviewers, area chairs, and senior area chairs. The acceptance rate hovers around 25-26%, but the caliber of competition is intense -- you are up against papers from DeepMind, OpenAI, Meta FAIR, and top university research labs. For most students, especially those without years of research experience, the main track is not the right first target.
The Workshop Track
NeurIPS hosts 50-80+ workshops each year, each focused on a specific subfield or emerging topic. Workshops accept shorter papers -- typically four-page extended abstracts plus unlimited references -- and the review process, while still rigorous, is more accessible. Workshop acceptance rates vary by workshop but are generally higher than the main track, and the format is designed to encourage new ideas, preliminary results, and work from emerging researchers.
Critically, a NeurIPS workshop publication is still a NeurIPS publication. It appears on your CV, it gets indexed, and it tells anyone reading your application that your work was reviewed and accepted by experts in the field.
Some well-known recurring NeurIPS workshops include:
- Machine Learning and the Physical Sciences (ML4PS) -- ML applied to physics, chemistry, and related domains
- AI for Accelerated Materials Discovery (AI4Mat) -- AI-driven materials science
- Tackling Climate Change with Machine Learning -- ML approaches to climate and sustainability challenges
- Socially Responsible Language Modelling Research (SoLaR) -- responsible development of language models
- Workshop on Multimodal Algorithmic Reasoning -- reasoning across text, vision, and other modalities
- Efficient Natural Language and Speech Processing (ENLSP) -- efficiency in NLP models and inference
The list changes each year as new workshops are proposed and accepted, so check the official NeurIPS workshops page for the current lineup.
NeurIPS Workshop Submission Timeline
Understanding the timeline is essential because missing a deadline means waiting an entire year. Here is the typical annual cadence, based on the NeurIPS 2025 schedule:
| Milestone | Typical Timing |
|---|---|
| Workshop proposals open | Mid-April |
| Workshop proposal deadline | Late May |
| Workshops announced | Early July |
| Workshop paper submission deadline | Mid-August to early September (varies by workshop) |
| Accept/reject notifications | By late September |
| Camera-ready deadline | Mid to late October |
| Conference and workshops | Early December |
The critical date for you as a paper author is the workshop paper submission deadline, which each workshop sets independently. Most fall between mid-August and early September. That means if the conference is in December, you need a submission-ready paper by roughly August -- which means your research should be substantially complete by July at the latest.
Plan backwards from August. If you are reading this in January, you have seven months. If you are reading this in May, you need to move fast.
Step 1: Choose a Research Topic
The topic you pick will determine almost everything about your experience -- how feasible the project is, how competitive the landscape looks, and which workshops will be a good fit.
What Makes a Good Student Research Topic
A strong topic for a student paper sits at the intersection of three things:
A genuine gap or question. You need to identify something that existing work has not fully addressed. This does not mean you need to invent a new field. It can be as focused as "method X has not been evaluated on domain Y" or "these two techniques have never been combined."
Feasible scope. You likely have 3-6 months and limited compute. Avoid topics that require training models from scratch on massive datasets unless you have access to significant GPU resources. Benchmarking, novel applications of existing methods, dataset creation, fairness/bias analysis, and efficiency improvements are all viable directions that do not require a supercomputer.
Alignment with active workshops. Browse last year's NeurIPS workshop list and the papers they accepted. If your topic fits naturally into one or two workshops, you have a viable submission target.
Trending Research Areas
Based on current activity in the ML community, these areas are producing a high volume of workshop-quality research:
- AI agents and multi-agent systems -- tool use, planning, coordination among LLM-based agents
- Multimodal reasoning -- integrating language, vision, and structured data
- Efficient inference and model compression -- making large models practical on smaller hardware
- AI for scientific discovery -- applications in biology, chemistry, materials science, drug discovery
- Responsible AI and fairness -- bias detection, dialect equity, evaluation of harms in language models
- Robustness and safety -- adversarial attacks, alignment, safe deployment of AI systems
- Retrieval-augmented generation (RAG) -- improving how LLMs access and use external knowledge
How to Find Your Specific Question
Start by reading 10-15 recent papers in your area of interest. Pay attention to the "Limitations" and "Future Work" sections -- researchers often spell out exactly what still needs to be done. Look for recurring themes across multiple papers. If three different groups mention the same open problem, that is a signal that the community cares about it and a solution would be well received.
Step 2: Find a Mentor or Research Program
This is where many students either succeed or stall. Working alone on your first research paper is technically possible but practically very difficult. You need someone who can help you scope the project correctly, catch methodological mistakes early, and navigate the submission process.
Options for Finding Mentorship
- University professors. If you are an undergraduate, approach professors whose work aligns with your interests. Come with a specific idea, not just "I want to do research." Show them you have read their papers.
- Graduate student mentors. PhD students are often more accessible than professors and can provide hands-on guidance with experiments and writing.
- Structured research programs. Programs like Algoverse pair students with principal investigators from leading labs -- including researchers at Meta FAIR, OpenAI, Google DeepMind, Stanford, and CMU -- and provide structured timelines aligned with conference deadlines. Algoverse, for instance, has seen 230 students accepted to NeurIPS 2025 workshops across 60 research papers, with a 68-73% conference acceptance rate. The structure matters: having a mentor who has published at these venues before dramatically reduces the odds of making avoidable mistakes.
- Online research communities. Groups on Twitter/X, Discord, and Reddit focused on ML research sometimes facilitate collaborations, though quality varies.
The key question to ask any potential mentor: Have you published at NeurIPS or a comparable venue before? Experience with the specific norms, expectations, and review culture of NeurIPS is invaluable.
Step 3: Conduct Your Research
With a topic defined and mentorship in place, the research itself follows a pattern that is consistent across virtually every ML paper.
Literature Review
Before running any experiments, you need to know what has already been done. Use Google Scholar, Semantic Scholar, and arXiv to find all relevant prior work. Read the foundational papers carefully, and skim the rest for key results and methods. Keep a structured document tracking each paper's contribution, method, and limitations.
A thorough literature review serves two purposes: it prevents you from accidentally duplicating existing work, and it provides the "Related Work" section of your paper almost for free.
Methodology
Define your approach clearly before you start coding. Write out:
- What your method does and why it should work
- What baselines you will compare against
- What datasets you will use
- What metrics you will report
Get feedback on this plan from your mentor before investing weeks in implementation. A common student mistake is jumping straight to code without a clear experimental plan, then realizing three months later that the experiment does not actually test the hypothesis.
Experimentation
Run your experiments systematically. Use version control for your code. Log all hyperparameters and results. Track random seeds. Save model checkpoints. The goal is reproducibility -- another researcher (or reviewer) should be able to understand exactly what you did.
Build tables and figures as you go, not at the end. If your results are not looking promising, discuss with your mentor early. Pivoting on a specific experiment in month three is fine; pivoting on your entire topic in month five is painful.
Step 4: Write Your Paper
The paper is the product. A strong paper with moderate results will often be accepted over a weak paper with strong results, because reviewers evaluate the clarity of thinking, not just the numbers.
Standard Paper Structure
Workshop papers are typically four pages (plus references), so every sentence needs to earn its place. The standard structure is:
Abstract (150-250 words). State the problem, your approach, and your key result. A reviewer who reads only the abstract should understand what you did and why it matters.
Introduction (0.5-0.75 pages). Motivate the problem. Why should anyone care? What is the gap in existing work? State your contribution clearly -- typically as a numbered list of 2-3 specific claims.
Related Work (0.5 pages). Position your work relative to prior research. Do not just list papers; explain how your approach differs from or builds on each one.
Method (1-1.5 pages). Describe your approach in enough detail that someone could reproduce it. Include mathematical notation where it adds clarity, but do not add equations for the sake of looking rigorous.
Experiments (1-1.5 pages). Present your results with clear tables and figures. Compare against relevant baselines. Include ablation studies if space allows -- reviewers love seeing which components of your method actually matter.
Conclusion (0.25 pages). Summarize your findings and state limitations honestly. Suggesting future work is fine but keep it brief.
LaTeX and Formatting
NeurIPS requires submissions in LaTeX using the official NeurIPS style file. If you have not used LaTeX before, start learning now -- Overleaf is the easiest way to get started and supports real-time collaboration.
Download the NeurIPS LaTeX template from the official conference website. Do not modify the margins, font sizes, or spacing. Papers that violate formatting guidelines get desk-rejected without review.
Common Writing Mistakes to Avoid
- Overselling results. Do not claim your method "solves" a problem unless it actually does. Use precise language: "improves over baseline X by Y% on metric Z."
- Ignoring limitations. Reviewers will find them anyway. Acknowledging limitations honestly builds credibility.
- Poor figure quality. Blurry screenshots, unlabeled axes, tiny text in plots -- all of these signal carelessness. Use vector graphics (PDF/SVG) and make every figure readable at print size.
- Wall-of-text syndrome. Use bullet points, numbered lists, and whitespace. A four-page paper that is hard to scan will frustrate busy reviewers.
Step 5: Submit to a Workshop
Identifying the Right Workshop
Start monitoring the NeurIPS workshops page as soon as accepted workshops are announced (typically early July). For each workshop, read the call for papers carefully. You are looking for:
- Topic alignment. Does your paper fit the workshop's stated scope?
- Submission format. Most workshops require four-page extended abstracts, but some accept longer papers or have different requirements.
- Deadlines. Workshop deadlines vary. Some close in mid-August; others extend into September.
You can submit to multiple workshops, but check each workshop's policy on dual submissions. Some workshops prohibit papers that are simultaneously under review elsewhere.
Submission Mechanics
Most NeurIPS workshops use OpenReview for submissions. You will need an OpenReview account with a verified institutional or academic email. Create your account well before the deadline -- verification can take time.
When submitting:
- Follow the anonymization requirements. Remove author names and affiliations from the PDF. Do not include acknowledgments that reveal your identity.
- Ensure your PDF is within the page limit. References do not count against the limit; appendices typically do not either, but reviewers are not required to read them.
- Upload supplementary materials (code, data) if the workshop allows it.
Submit Early
Do not wait until the last hour. OpenReview occasionally has server issues near deadlines. Aim to submit at least 24 hours early. You can update your submission up until the deadline.
Step 6: Respond to Reviews
After submission, you will typically receive 2-3 reviews within 3-5 weeks. Workshop reviews are shorter than main conference reviews but follow a similar format: a summary of the paper, strengths, weaknesses, and a recommendation.
Understanding Review Feedback
Read all reviews carefully before responding. Categorize the feedback:
- Factual corrections. If a reviewer misunderstood your method, clarify it. This is often a sign that your writing was unclear, not that the reviewer was wrong.
- Missing experiments. If a reviewer asks for an additional baseline or ablation, try to run it during the rebuttal period if one is offered.
- Subjective concerns. Some feedback reflects taste or emphasis. Acknowledge these points respectfully.
Writing a Rebuttal
Not all workshops offer a rebuttal period, but if one is available, use it strategically. Be concise, factual, and polite. Address each reviewer's major concerns point by point. If you ran additional experiments in response to feedback, present the results.
The golden rule: never be adversarial. Even if a review feels unfair, respond as if you are having a conversation with a respected colleague.
If You Are Rejected
Rejection is common and is not a statement about your ability. Use the feedback constructively:
- Revise the paper based on reviewer comments.
- Submit to another workshop at a different conference (ICML, ICLR, AAAI, and EMNLP all have workshops with similar formats).
- Consider whether the core idea needs more experimental support or a different framing.
Many papers that become influential were rejected on their first submission.
Step 7: Present Your Work
If your paper is accepted, congratulations -- you now need to present it. Most workshop papers are presented as posters, though some workshops select a handful of papers for short oral presentations.
Poster Presentations
Design your poster to be readable from four feet away. Use large fonts (24pt minimum for body text), clear figures, and a logical flow from top-left to bottom-right. Your poster should tell the story of your paper in three minutes or less.
Practice your "elevator pitch" -- a 60-second summary of what you did and why it matters. Conference attendees will walk up to your poster, and you need to hook them quickly.
Oral Presentations
If selected for a talk, you will typically have 5-10 minutes plus questions. Structure your slides around the key insight and main result -- do not try to compress your entire paper into slides. Rehearse the timing multiple times.
Networking at the Conference
NeurIPS is not just about your paper. It is one of the best opportunities you will ever have to meet researchers, potential advisors, and collaborators. Some practical tips:
- Attend other workshops and poster sessions. Ask questions. Show genuine interest in other people's work.
- Introduce yourself to researchers whose papers you have read. "I read your paper on X and had a question about Y" is one of the best opening lines in academia.
- Exchange contact information. Follow up with people you meet within a week of the conference.
- Attend social events. Many of the most productive conversations happen at receptions and dinners, not during formal sessions.
Tips from Students Who Have Done It
The advice above is general. Here is what we have seen specifically from students who successfully published at NeurIPS workshops.
Start earlier than you think you need to. Students who begin their research in January or February for a December conference consistently produce stronger work than those who start in May. The extra time is not about working more hours -- it is about having room to iterate on ideas and recover from dead ends.
Let your mentor shape the scope. The most common failure mode is a project that is too ambitious. A focused contribution with clean experiments beats a sprawling project with incomplete results every time. Trust your mentor's judgment on what is feasible.
Read accepted papers from last year's workshops. This is the single best way to calibrate your expectations. Download five papers from the workshop you are targeting and study their structure, depth, and contribution size.
Do not underestimate writing quality. Among students who worked through programs like Algoverse, those whose papers were accepted consistently cited writing feedback as one of the most valuable parts of the process. Clarity is a skill, and it is learnable.
Some notable recent outcomes show what is possible. Two Algoverse students -- Abhay Gupta and Philip Meng -- were named 2025 Davidson Fellows for their AI fairness research on dialect equity in large language models (the Davidson Fellows Scholarship selects only 20 recipients annually from over 1,200 applicants). Another group of Algoverse students had their paper Semantic Self-Consistency featured among 20 state-of-the-art papers in OpenAI's PaperBench project, originally accepted at the NeurIPS MATH-AI workshop. These are not outliers -- they are examples of what happens when motivated students get proper mentorship and commit to the process.
Frequently Asked Questions
Can high school and college students publish at NeurIPS?
Yes. NeurIPS does not have age or enrollment requirements for authorship. What matters is the quality of the work. Algoverse has had 230 students accepted to NeurIPS 2025, including high school students, college students, and graduate students. With experienced mentorship from PIs at Meta FAIR, OpenAI, Google DeepMind, Stanford, and CMU, students consistently produce work that passes peer review. For a full roadmap, see our guide on how to get into AI research as a high school student.
How long does it take to go from zero research experience to a NeurIPS paper?
Algoverse's program runs 12 weeks and aims for publication in approximately 3 months. This includes ramping up on the relevant literature, conducting research and running experiments, and writing and revising the paper. The timeline is achievable because students work with experienced PIs who know what NeurIPS reviewers expect, and Algoverse covers all GPU and compute costs so there are no resource bottlenecks.
Do I need access to expensive GPUs or compute resources?
If you work with Algoverse, no -- Algoverse covers all GPU and compute costs for students. Many research directions also involve clever experimental design rather than raw compute: fine-tuning smaller models, evaluating existing models on new benchmarks, creating curated datasets, or conducting analysis and fairness audits. The compute barrier should never prevent a motivated student from publishing.
Can I submit the same paper to multiple NeurIPS workshops?
Policies vary by workshop. Some explicitly prohibit dual submissions; others allow it. Read each workshop's call for papers carefully. As a general practice, it is better to target the single best-fit workshop and put all your effort into tailoring the paper for that audience.
What if my paper gets rejected?
Rejection is a normal part of academic research. A good program will help you understand reviewer feedback, revise your work, and resubmit to another venue -- NeurIPS is one of many top conferences, and ICML, ICLR, ACL, EMNLP, and AAAI all have similar workshop tracks. Algoverse maintains a 68-73% acceptance rate across these venues, and supports students through resubmission if needed.
Getting Started
Publishing at NeurIPS as a student is not easy, but it is a well-defined process. The students who succeed are not necessarily the most brilliant -- they are the ones who start early, find good mentorship, scope their projects realistically, and commit to iterating on both their research and their writing.
If you are looking for structured support, programs like Algoverse provide mentorship from PIs at leading AI labs, project guidance aligned with conference timelines, and a peer community of students working toward the same goal. But regardless of how you get there, the path is the same: find a question worth asking, do the work to answer it rigorously, and communicate your findings clearly.
The next NeurIPS deadline is closer than you think. Start today.
Related reading:
Related Articles
50 AI Research Topics for High School Students [2026]
A curated list of 50 specific, feasible AI research topics for high school students, organized by subfield -- from NLP and computer vision to AI safety and healthcare AI.
How to Write Your First AI Research Paper [Student Guide]
Learn how to write an AI research paper as a student. Covers paper structure, literature review, common mistakes, LaTeX tools, and mentor collaboration tips.
How to Get Into AI Research as a High School Student [2026 Guide]
Learn how to get into AI research as a high school student. Step-by-step guide covering prerequisites, finding mentors, and publishing at top conferences.
