Lewati ke isi

[GS07] MapCoder: Multi-Agent Code Generation for Competitive Problem Solving

Kunci Stabil: GS07 | Sitasi BibTeX: 1

Ringkasan Metadata

  • Penulis & Tahun: Md Ashraful Islam and others (BUET & Rochester), 2024
  • Venue & Tier: ACL 2024 Long
  • DOI / URL: 10.18653/v1/2024.acl-long.269
  • Skor Kualitas (QA Score): - (QA korpus baru OPEN, belum dinilai)
  • Peran Studi: QGS (penguji recall P1+P2)
  • Tingkat Ekstraksi: metadata-calibration
  • Cakupan (Scope): Competitive problem solving (HumanEval, MBPP, APPS, CodeContests), function-level, bukan repo.

1. Masalah Utama (Core Problem)

Generasi kode problem-solving kompetitif via multi-agen.

2. Arsitektur yang Diajukan (Architecture)

4 agen (Retrieval, Planning, Coding, Debugging); siklus plan-code-debug retrieval-augmented.

3. Mekanisme Koordinasi & Kontrol (Coordination)

Siklik retrieval-planning; otonom penuh tanpa hierarki formal dan tanpa gate.

4. Foundation Model & Infrastruktur (FM)

LLM umum era 2024.

5. Tolok Ukur & Dataset (Benchmark)

HumanEval, MBPP, APPS, CodeContests.

6. Artefak & Kode Sumber (Artifact)

ideis/MapCoder.

7. Metrik Primer Eksak (Primary exact)

UNDOCUMENTED di SSOT jalur baru (QA/DEF korpus baru OPEN).

8. Metrik Sekunder (Secondary)

-

9. Pola Kegagalan (Failure Modes)

Generalisasi ke repo skala besar belum terbukti.

10. Ancaman Validitas (Threats)

UNDOCUMENTED (QA korpus baru OPEN).

11. Jejak Bukti & Provenansi (Provenance)

Bukti Verifikasi: EVIDENCED-FACT (GS_quasi_gold.md GS07; Scopus-769 HIT DOI eksak; arXiv 2405.11403v1 HIT).

  1. Md Ashraful Islam and 2024 others (BUET & Rochester). MapCoder: Multi-Agent Code Generation for Competitive Problem Solving. 2024. Venue: ACL 2024 Long, DOI: 10.18653/v1/2024.acl-long.269, Role: QGS (penguji recall P1+P2). ↩