{"paper":{"title":"Phylogenetic Tree Inference with Tropical Axial Attention","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Tropical axial attention replaces standard attention with max-plus operators to learn phylogenetic distances aligned with tree geometry.","cross_cats":["cs.LG"],"primary_cat":"q-bio.PE","authors_text":"Baran Hashemi, Chris Teska, Kurt Pasque, Ruriko Yoshida","submitted_at":"2026-05-12T10:54:37Z","abstract_excerpt":"In this work, we introduce a Tropical Axial Attention neural reasoning architecture that replaces vanilla softmax dot-product attention with max-plus operators, inducing a piecewise-linear structure aligned with dynamic programming formulations. From multi-species sequence alignments, our model learns all possible pairwise distances and is trained using a combination of $\\ell_1$ and tropical symmetric distance metric losses with an ultrametric violation penalty. We leverage the well known isomorphic relationship between the space of all phylogenetic trees with $n$ species and tropical Grassman"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"tropical attention provides a natural geometric framework for phylogenetic inference. On empirical DS1-DS11 alignments, the tropical model produces distance matrices that are substantially closer to their BME-induced tree metrics than the baseline models.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The piecewise-linear structure induced by max-plus operators and the ultrametric violation penalty will align with actual phylogenetic tree geometry even when true trees are unknown and evaluation relies on BME-induced metrics.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Tropical axial attention learns pairwise distances from alignments and produces matrices closer to BME tree metrics than standard attention baselines by replacing dot-product attention with max-plus operators and adding ultrametric penalties.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Tropical axial attention replaces standard attention with max-plus operators to learn phylogenetic distances aligned with tree geometry.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"603ed01cfd8e617d4c6c135e2d655dab4924b371ceea8c7fb917153187c0c8b3"},"source":{"id":"2605.13894","kind":"arxiv","version":1},"verdict":{"id":"b62262e8-a769-4384-9363-67964eec899d","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T06:07:07.596298Z","strongest_claim":"tropical attention provides a natural geometric framework for phylogenetic inference. On empirical DS1-DS11 alignments, the tropical model produces distance matrices that are substantially closer to their BME-induced tree metrics than the baseline models.","one_line_summary":"Tropical axial attention learns pairwise distances from alignments and produces matrices closer to BME tree metrics than standard attention baselines by replacing dot-product attention with max-plus operators and adding ultrametric penalties.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The piecewise-linear structure induced by max-plus operators and the ultrametric violation penalty will align with actual phylogenetic tree geometry even when true trees are unknown and evaluation relies on BME-induced metrics.","pith_extraction_headline":"Tropical axial attention replaces standard attention with max-plus operators to learn phylogenetic distances aligned with tree geometry."},"references":{"count":36,"sample":[{"doi":"","year":2018,"title":"Capacitated dynamic programming: Faster knap- sack and graph algorithms","work_id":"c6442afe-f36b-46b6-87bd-a52fb39c9a2b","ref_index":1,"cited_arxiv_id":"1802.06440","is_internal_anchor":true},{"doi":"","year":2025,"title":"J., Duchene, D., and Yoshida, R","work_id":"ffd8dde2-f3d8-49bc-b0c2-c88c670feb54","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1006/aama.2001","year":2001,"title":"Permutations with Restricted Patterns and Dyck Paths","work_id":"7fb02a36-d1fa-4604-af34-ad96f8215592","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Blassel, L., Sauvage, N., Barrat-Charlaix, P., Boussau, B., Lartillot, N., and Jacob, L. (2025). Likelihood-free inference of phylogenetic tree posterior distributions.arXiv preprint arXiv:2510.12976","work_id":"c3a0cc9c-b928-445d-b2eb-ea20994f09f7","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Chen, Y., Huang, J., Yang, C., Hsu, K., and Liu, H. (2023). A comprehensive phyloge- netic analysis of sars-cov-2: Utilizing a novel and convenient in-house rt-pcr method for characterization without ","work_id":"deb74b7c-9e1d-4e88-9c16-7e8cf0644f47","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":36,"snapshot_sha256":"236eba6fe6807d6c7a161013ca4b6aac07f6c9a91f015d5f9c2eea5cd4b6189a","internal_anchors":1},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}