Ensemble methods come in two families — many copies of one kind of model
(homogeneous), and a mix of different kinds (heterogeneous) — and the folklore about
when each one wins is quoted far more often than it is measured. This work measures it:
the same prediction and feature-selection tasks, run across both families of
tree-based ensembles, under conditions where the comparison is actually fair.
It sits in the breeding platform's research direction (P.001) as the feature-selection
and model-interpretation component: which variables survive when several models are
asked at once, and whether the answer is stable enough to act on.
The paper's bibliographic record is in the academic section; this page is about the
work itself — the question it asks, the line it belongs to, and what can and cannot
be said about it yet. The two do not overlap: the academic section answers "what was
published", this one answers "what is being done".
Submitted to ACML 2026 (OpenReview Submission 132) and in review: the reviewers'
reports came back and the author rebuttal was filed in September 2026. The author list
is ten people across two colleges, with 林海波 and 马晓东 among the corresponding
authors.
Seventh of ten authors. On a paper of that size the honest description of one person's
share is narrow, and this page does not improve on the contribution record — that
record, not the website of one of the authors, is the place that decides the wording.
What this page is waiting for
The method, the benchmark and the result belong here once the review settles and the
camera-ready wording exists. Until then this page says only what can be said without
guessing: the title, the setting, and where it sits in the larger project.
The paper is under active review. Turning an unsettled result into a stated conclusion
is the easiest mistake to make on a page like this.