NY-MLRJoin Discord →

NY-MLR · NEW YORK CITY

New York Machine Learning Research Group

An informal machine-learning research community in New York. Monthly colloquia, a Discord for trading ideas, and working on research together. Open to anyone with a sincere interest with the will to do Great Work.

What this is

Activities, functions and scope of the Group

Discord
A shared space to trade knowledge, papers, and questions. Open to every member.
Monthly colloquia
Researchers from local institutions and labs present their work; members meet, network, and exchange ideas. Open to attend, limited only by room capacity.
Industry & events
Sharing sessions with industry, hackathons, group trips to conferences, and joint workshops. On the roadmap, not yet running.
Working groups
Small teams that form around one research project and carry it through to a published result. This is what the group is really for. More details below.

Culture & values

Our core guiding principles

  1. Sincere interest is enough to belong.

    Anyone genuinely curious about the field and its ideas has a place here.

  2. Collaboration comes first.

    Courtesy and respect are the baseline. Treat others as you'd want to be treated.

  3. We aim high.

    Welcoming as it is, the group means to do work that matters.

  4. The bar is there to lift, not to shut people out.

    Standards exist to push everyone toward better work, not to gatekeep for its own sake.

  5. Give back what you were given.

    Help others grow, and yield when it's called for. That's what keeps a community alive.

If you don't work on important problems, it's not likely that you'll do important work.Richard Hamming

Colloquia & events

Talks from researchers across the city. Past and upcoming in one view.

DateTalk

Details to be announced. This is placeholder text so the component renders; replace with the real event.

New York City · venue TBA

Placeholder for a second scheduled talk so the upcoming list is not empty. Replace with the real event.

New York City · venue TBA

Placeholder for a past event so the past/upcoming distinction is visible in the layout.

New York City

Another placeholder past event. Replace this array with real colloquia as they are scheduled.

New York City

The first gathering of the group. Placeholder copy — swap in the real recap and any recording link.

New York City

Party & guild system

The heart of the group is simple, and if you've ever played an RPG you already know how it works. People post on a board. Some are looking for a party to join; others have started one and are looking for members. They find each other around a concrete research project, agree on a goal and a deadline, and get to work. A loose interest becomes a committed team with something to finish, a true seasoned Team.

Every project sets its own finish line at the start: a public preprint or shared code or whatever your team decides, by an agreed date. Membership to a party is vetted, lightly, so the people who join are serious. That’s all you need to know to start.

LOOKING FOR PARTY

Interested in
mechanistic interpretability, small transformers
Bringing
training infra, PyTorch, some evals experience
Looking for
a party working on interpretability or evals
Commitment
~8 hrs/week
Note
Second-year student, one workshop paper. Want to go deeper on a real project with people who care.
[ Request to join a party → ]

PARTY FORMING — LOOKING FOR MEMBERS

Project
Evaluating long-context retrieval failures
Led by
a member with a prior accepted paper
Seeking
2 members — one for data/eval harness, one for analysis and writing
Commitment
~8 hrs/week · 12 weeks
Finish line
preprint + public repo by the deadline
Note
Serious, friendly, and we intend to publish.
[ Apply to this party → ]

PARTIES IN PROGRESS

Several parties run in parallel at any time, each at its own stage. A snapshot of what is active right now:

Long-context retrieval failuresTopicEvaluationMembers3AffiliationNYU · Cornell TechStatusDraftingProgress70%
Mechanistic interp of small transformersTopicInterpretabilityMembers4AffiliationColumbia · independentStatusExperimentsProgress45%
Efficient KV-cache compressionTopicSystemsMembers2AffiliationFlatiron InstituteStatusExperimentsProgress55%
Preference data quality for RLHFTopicAlignmentMembers5AffiliationNYU · industryStatusFormingProgress15%
Graph neural nets for materialsTopicApplicationsMembers3AffiliationCUNY · independentStatusAnalysisProgress80%
Diffusion samplers, fasterTopicGenerativeMembers2AffiliationPrincetonStatusFormingProgress10%