1Where AI/ML Papers Live
AI/ML research is distributed across preprint servers, conference proceedings, and journals. Knowing which repository to search for which type of paper saves hours per week. For cutting-edge ML, arXiv is always the first stop — most papers appear there weeks before the conference proceedings are published.
arXiv (cs.LG, cs.AI)
Free preprint server. Most ML papers appear here 1-6 months before conference publication. No peer review — quality varies. The standard for sharing work in the field.
Semantic Scholar
AI-powered academic search by Allen Institute. Parses citation graphs, extracts key claims, and provides free paper feeds. Best for citation discovery and influence mapping.
OpenReview
Hosts reviews and papers for NeurIPS, ICML, ICLR, and others. Invaluable: you can read the actual reviewer critiques of any accepted or rejected paper.
Google Scholar
Broadest coverage across all fields and all eras. Best for finding foundational papers from the 1990s-2000s that predate arXiv's popularity in ML.
Papers With Code
Links papers to their GitHub implementations and benchmark results. Essential for reproducibility — see exactly what state-of-the-art looks like on every dataset.
ACM DL / IEEE Xplore
Formal proceedings for systems, HCI, and classical CS. Much older literature lives only here. Institutional access often required — check your university library first.
2Search Strategies That Work
Keyword search is a starting point, not an ending point. The best researchers use a combination of keyword search, author tracking, venue-based search, and citation chaining. Keyword search finds what you know to ask for; citation chaining finds what you didn't know existed.
The snowball method: start with one seminal paper in your area. Read its references to find older foundational work (backward citation). Find papers that cite your seminal paper on Semantic Scholar or Google Scholar (forward citation). This bidirectional expansion covers most of the relevant literature in a field.
Effective search tactics
- ✓ Search author names once you find a relevant researcher
- ✓ Browse NeurIPS/ICML/ICLR proceedings for the current year
- ✓ Use "cited by" on a landmark paper for forward citations
- ✓ Add synonyms: "representation learning" vs "self-supervised"
Common pitfalls
- ✗ Trusting citation count as proxy for quality (popular ≠ correct)
- ✗ Ignoring rejected papers on OpenReview — they often contain useful ideas
- ✗ Not checking the arXiv version vs conference version for updates
- ✗ Missing workshop papers — many ideas debut in workshops 1-2 years early
3Staying Current
The field publishes 500+ papers per week on arXiv cs.LG alone. No human reads all of them. The solution is a curated signal system: set up alerts for specific topics, follow key researchers, and use aggregators that surface the most-discussed papers daily.
arXiv email digest
Subscribe at arxiv.org to receive daily new submissions in cs.LG, cs.AI, and stat.ML. A 10-min morning scan of titles and abstracts covers 90% of what matters.
Hugging Face Daily Papers
Community-curated: researchers upvote the day's most interesting arXiv papers. Algorithmic signal cuts noise dramatically. Free at huggingface.co/papers.
Twitter/X + newsletters
Follow top ML researchers for paper recommendations and hot takes. Newsletters like The Batch (deeplearning.ai) and Import AI curate weekly highlights.
4Citation Networks: Going Deeper
Some of the most useful papers in a field are not the most cited — they are the ones that influenced the papers that influenced the most cited papers. Citation network tools visualise this graph and surface papers you would never find with keyword search.
Connected Papers
Enter one paper; get a force-directed graph of related papers weighted by citation overlap. Excellent for mapping an unfamiliar sub-field in under 5 minutes.
ResearchRabbit
Often called "Spotify for papers". Create a collection; it learns your taste and recommends similar papers. Particularly strong for interdisciplinary work.
Semantic Scholar Feeds
Follow authors and topics. Get notified when a tracked author publishes. The "influential citations" filter surfaces papers that made a methodological impact, not just a citation count.
Forward citations (who cited this paper?) are more useful than backward citations for finding recent work. Backward citations map history; forward citations map the present and future of a line of research. Always check both directions on any landmark paper.