References

Core papers

• Rendle, S. (2010). Factorization Machines. ICDM. Key formula: y = w0 + sum_i w_i x_i + sum_{i<j} <v_i, v_j> x_i x_j, with the O(nk) pairwise reformulation used by all fast implementations.
• Juan, Y., Zhuang, Y., Chin, W.-S., Lin, C.-J. (2016). Field-aware Factorization Machines for CTR Prediction. RecSys. Key formula: interaction(i, j) = <v_{i, field_j}, v_{j, field_i}> x_i x_j.

Roadmap-model papers (v0.3+)

• AFM — Xiao et al., Attentional Factorization Machines (arXiv:1708.04617). Learns per-interaction importance via attention.
• FEFM — Pande, Field-Embedded Factorization Machines (arXiv:2009.09931). Field-pair matrix embeddings, lower complexity than FFM.
• FmFM — Sun et al., FM^2: Field-matrixed Factorization Machines (arXiv:2102.12994).
• FwFM — Pan et al., Field-weighted Factorization Machines (WWW 2018).

Existing libraries

• libffm — canonical C++ FFM implementation (logistic loss, OpenMP/SSE): https://www.csie.ntu.edu.tw/~cjlin/libffm/ Note: libffm omits w0 and the linear term; we keep both (see math_spec.md).
• xLearn — C++ LR/FM/FFM library: https://github.com/aksnzhy/xlearn
• fastFM — existing FM library (name collision avoided: we are modern-fm)
• DeepCTR-Torch — PyTorch CTR models including FM-family

Design inspirations

• scikit-learn estimator API: https://scikit-learn.org/stable/developers/develop.html
• cuML sklearn-like GPU API: https://docs.rapids.ai/api/cuml/stable/
• PyO3 / maturin for Rust-Python binding: https://pyo3.rs/
• cibuildwheel for wheel distribution: https://cibuildwheel.pypa.io/