Phylogenetic inference is a challenging generative modeling problem because phylogenetic trees combine continuous branch lengths with discrete topological structure. Bayesian methods provide a principled framework for representing uncertainty, but exploring the resulting posterior distribution over trees remains computationally difficult.<\/p>\n\n\n\n
In this talk, I will present PhylaFlow, a hybrid flow-matching framework that learns transport directly in Billera\u2013Holmes\u2013Vogtmann (BHV) tree space. PhylaFlow is trained on geodesic paths from random trees toward short-run posterior samples, coupling continuous branch-length dynamics with learned boundary events that change tree topology. Across eight phylogenetic posterior benchmarks, PhylaFlow reaches posterior-relevant regions more effectively than classical initialization strategies and improves finite-budget Bayesian refinement. I will also discuss a split-guided PhylaFlow-MCMC variant, preliminary sequence-conditioned inference, and the broader opportunity to combine geometric generative models with classical scientific inference. This work represents an initial step toward scalable, differentiable phylogenetics.<\/p>\n\n\n\n
Yasha Ektefaie is an Eric and Wendy Schmidt Fellow and Postdoctoral Associate in the Sabeti Lab at the Broad Institute of MIT and Harvard. He received his PhD from Harvard Medical School, where he was co-advised by Maha Farhat and Marinka Zitnik. His research develops machine-learning methods for biological inference and discovery, spanning differentiable phylogenetics, generative modeling, drug discovery, and AI agents for science. His recent work, PhylaFlow, was selected as one of two Best Academic Papers at the ICML 2026 GenBio Workshop.<\/p>\n","protected":false},"excerpt":{"rendered":"
Phylogenetic inference is a challenging generative modeling problem because phylogenetic trees combine continuous branch lengths with discrete topological structure. Bayesian methods provide a principled framework for representing uncertainty, but exploring the resulting posterior distribution over trees remains computationally difficult. In this talk, I will present PhylaFlow, a hybrid flow-matching framework that learns transport directly in […]<\/p>\n","protected":false},"featured_media":1184233,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","msr_hide_image_in_river":0,"footnotes":""},"research-area":[13556],"msr-video-type":[270340],"msr-locale":[268875],"msr-post-option":[],"msr-session-type":[],"msr-impact-theme":[],"msr-pillar":[],"msr-episode":[],"msr-research-theme":[],"class_list":["post-1184232","msr-video","type-msr-video","status-publish","has-post-thumbnail","hentry","msr-research-area-artificial-intelligence","msr-video-type-msr-new-england-generative-modeling-sampling-seminar","msr-locale-en_us"],"msr_download_urls":"","msr_external_url":"https:\/\/youtu.be\/m4KVm3SkkGY","msr_secondary_video_url":"","msr_video_file":"http:\/\/0","_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-video\/1184232","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-video"}],"about":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-video"}],"version-history":[{"count":1,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-video\/1184232\/revisions"}],"predecessor-version":[{"id":1184234,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-video\/1184232\/revisions\/1184234"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media\/1184233"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1184232"}],"wp:term":[{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1184232"},{"taxonomy":"msr-video-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-video-type?post=1184232"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1184232"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=1184232"},{"taxonomy":"msr-session-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-session-type?post=1184232"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=1184232"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=1184232"},{"taxonomy":"msr-episode","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-episode?post=1184232"},{"taxonomy":"msr-research-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-theme?post=1184232"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}