Skip to content
OUTTA

OUTTA AI research community

Think out of the box

Learn the foundations, build them yourself, and grow your own question into research with your peers.

How to join See the year ahead
OUTTA

OUTTA means out of. The name carries our belief in new experiences outside the usual frame.

Lining up everyone with a single yardstick.

But grades don’t decide who has world-changing ideas.

So OUTTA creates chances to learn by asking your own questions.

Then we look for answers with our peers, learning beyond the desk.

Across 6 cohorts, 147 people in total

From Cohort 1 to Cohort 5, people who learned, built and asked together.

  1. Cohort 1: 35 in total, 35 new
  2. Cohort 1.5: 49 in total, 14 new
  3. Cohort 2: 57 in total, 8 new
  4. Cohort 3: 104 in total, 47 new
  5. Cohort 4: 137 in total, 33 new
  6. Cohort 5: 147 in total, 10 new

One year, from learning to research

  1. Meet your peers and plan the year

    Late NovMeet your peers and plan the year

    At the orientation, you learn how the year works and meet the people you’ll learn and research with.

  2. Build what you learn

    Dec – JanBuild what you learn

    You build your foundations with OUTTA’s own curriculum and complete at least three mini-projects.

    OutputMini-projects5-min presentation

  3. One problem, start to finish

    FebOne problem, start to finish

    Pick one problem and explain it yourself, from implementation to analysis and limits.

    OutputGitHub10-min presentation

  4. Find the team that fits you

    Late FebFind the team that fits you

    Read the team introductions, choose your preferences, and get placed after a project review.

  5. Ask, experiment and write together

    Mar – NovAsk, experiment and write together

    Narrow your question through prior work, then implement, experiment and analyze, aiming to submit a paper.

DRAFT → SUBMIT ↗ORIENTATIONMINI 01MINI 02MINI 03DATA · AI · ML / CV / NLPProblemBuildTestAnalyzeTEAM ATEAM BTEAM CTEAM DAttention Is All You NeedVaswani, Shazeer, Parmar, Uszkoreit, Jones, Gomez, Kaiser, PolosukhinNIPS 2017

Abstract

Sequence transduction models have long relied on recurrent or convolutional networks arranged as an encoder and a decoder. This work introduces the Transformer, an architecture built entirely on attention, with no recurrence or convolution. It parallelizes far better and trains in a fraction of the time, while reaching strong translation quality on the WMT 2014 English–German and English–French benchmarks.

1   Introduction

Recurrent networks process a sequence one step at a time, which limits parallel computation over long inputs. Attention instead lets every position relate to every other position directly. We stack multi-head self-attention with simple position-wise feed-forward layers, and add positional encodings so the model still knows the order of tokens.

Figure 1. The Transformer: encoder and decoder stacks of multi-head attention and position-wise feed-forward layers.

On shared foundations, toward deeper questions

Everyone learns data analysis and applied AI together, then digs into at least one of ML · CV · NLP.

  • ML

    Machine learning · Deep learning

    From core theory to practice. Learn how to find the boundaries in data.

  • CV

    Computer vision

    From CNNs to segmentation, object detection and GANs, built by hand.

  • NLP

    Natural language processing

    From RNNs and seq2seq to Transformers, BERT and GPT: the ideas behind LLMs, in practice.

  • 01Data science fundamentals
  • 02Current AI research
  • 03Applied AI

New to research? You can still start with us

You don’t need publications first. We look at what you’ve learned and tried, how you work with others, and whether you can make time for the year.

Recruiting Cohort 6

October 26 – November 13, 2026

Applications close on November 13, 2026 at 23:59 (KST).

Details Ask on KakaoTalk

What OUTTA is made of