Paper Club · Bi-weekly · Open · Hybrid
N° 01Vol. I · 2026

We read what we want to write.

A reading group on topological data analysis, interpretable ML, and applied mathematics for African systems. One paper, two discussants, an hour of careful argument. Hybrid: in-person at the Cotonou office or by Zoom. Open to all — first-timers welcome.

How a session works

One hour. Five movements.

00 · 10 min

Arrivals & framing

Soft start; one of the discussants frames why the paper was chosen and what question we are testing it against.

01 · 15 min

First discussant

Steel-man the paper. Walk the room through the contribution as the authors meant it to be read.

02 · 15 min

Second discussant

Critique. What does the paper not earn? What hidden assumption is doing the work? What would replicate, what would not?

03 · 15 min

Open discussion

The room argues. Newcomers and senior researchers in the same conversation; the chair keeps the thread tight.

04 · 5 min

Close

What the room agreed on, what stayed contested, what we want to read next. Notes go on the mailing list within 48 hours.

Format & logistics

The practical details.

Cadence

Bi-weekly, Thursdays, 16:00 WAT. The specific Thursday and the paper go out on the mailing list one week ahead.

Where

Hybrid. In-person at the AIRINA Labs office in Cotonou, Bénin. Remote attendees join by Zoom — link sent with the mailing-list announcement.

Languages

Bilingual EN / FR. Discussion switches between the two as the room requires; the chair will translate on request.

Preparation

Read it if you can. Come anyway if you cannot — the discussants frame the paper at the top of the session, and the discussion is more useful for newcomers when they listen first.

Notes & archive

Short discussion notes are sent on the mailing list within 48 hours of each session. The archive of past sessions and notes is shared with subscribers.

Cost

Free. The Paper Club is an open institutional activity; no registration fee in person or online.

Next session

Where we are in the cycle.

Cycle 1 · In progress Next paper · date announced via mailing list

Get on the list.

We're working through Cycle 1's reading list now. Each session's paper and date go out on the mailing list a week ahead, with the Zoom link.

Reading list

Ten papers, two cycles, one trajectory.

Cycle 1 Foundations to deployment
  1. 01

    Topology and Data

    Carlsson, G. (2009) · Bulletin of the AMS 46(2), 255–308 · DOI

    The canonical introduction to TDA. Every subsequent paper assumes you have read this one — we open with it for exactly that reason.

  2. 02

    Topological persistence and simplification

    Edelsbrunner, H., Letscher, D., & Zomorodian, A. (2002) · Discrete & Computational Geometry 28(4), 511–533

    The founding algorithm of persistent homology. Where the field was named and proven.

  3. 03

    On the local behavior of spaces of natural images

    Carlsson, G., Ishkhanov, T., de Silva, V., & Zomorodian, A. (2008) · International Journal of Computer Vision 76(1), 1–12

    The most beautiful applied-TDA paper: a Klein bottle hiding in the patches of every photograph.

  4. 04

    Topological methods for the analysis of high-dimensional data sets and 3D object recognition

    Singh, G., Mémoli, F., & Carlsson, G. (2007) · Eurographics Symposium on Point-Based Graphics

    Mapper, the second pillar of TDA. Required before anything Mapper-shaped shows up in our own work.

  5. 05

    Persistence images: a stable vector representation of persistent homology

    Adams, H., Emerson, T., Kirby, M., et al. (2017) · Journal of Machine Learning Research 18(8), 1–35

    The bridge from persistence diagrams to standard ML pipelines. Required before any downstream classifier reads a TDA feature.

  6. 06

    Behavior revealed in mobile phone usage predicts credit repayment

    Björkegren, D., & Grissen, D. (2020) · The World Bank Economic Review 34(3), 618–634

    The literature directly under our Phase 1 products. We agree, we disagree, we read it carefully.

Cycle 2 On deck
  1. 07

    Stability of persistence diagrams

    Cohen-Steiner, D., Edelsbrunner, H., & Harer, J. (2007) · Discrete & Computational Geometry 37(1), 103–120

    The stability theorem. Why persistent homology is a real signal and not noise.

  2. 08

    Statistical topological data analysis using persistence landscapes

    Bubenik, P. (2015) · Journal of Machine Learning Research 16, 77–102

    The other major vectorization of persistence diagrams. Bubenik vs. Adams, head to head.

  3. 09

    Stop explaining black-box machine-learning models for high-stakes decisions and use interpretable models instead

    Rudin, C. (2019) · Nature Machine Intelligence 1(5), 206–215

    The case for interpretable models, not post-hoc interpretations of black boxes. Why AIRINA products must be the former.

  4. 10

    Topology-based data analysis identifies a subgroup of breast cancers with a unique mutational profile and excellent survival

    Nicolau, M., Levine, A. J., & Carlsson, G. (2011) · PNAS 108(17), 7265–7270

    The high-stakes-medicine TDA paper. Precedent for our Phase 3 health work.

Two ways in

Subscribe, or propose a paper.

Subscribe to the mailing list

Get the next session's paper, date, and Zoom link a week ahead. Plus the discussion notes within 48 hours of each session. The list is low-volume: typically one announcement and one notes email per session.

Subscribe →

Propose a paper for a future cycle

Suggest a paper that fits the institute's lines: TDA, interpretable ML, applied mathematics for African systems. One paragraph on why it matters and what you would want the room to argue about. Cycle 3 is open for proposals.

Propose a paper →
Past sessions

Archive in progress.

The Paper Club is in its first volume. Once Cycle 1 closes, the discussion notes, contested points, and follow-up references for each past session will be linked here — freely readable. Subscribers receive the notes by mail within 48 hours of each session in the meantime.