Skip to main content
Search

# Gaussian Mixture Model

4 related articles

Subscribe to updates
  1. arXiv

    Every generative model eventually fails a large enough test: SIGMA profiles fit uncertainty and keeps the same model trustworthy on test sets an order of magnitude larger

    A generative model fitted to a finite training set is never exactly the true density, so any sufficiently powerful goodness-of-fit test will eventually reject it once the test sample is large enough, even when the model family is correct. SIGMA writes the parameter uncertainty of a Gaussian mixture model as a tractable covariance from its score functions and Fisher information, then promotes the conventional point-null test to a composite null in which that uncertainty is profiled as a nuisance parameter. On a two-dimensional toy model and on a QCD background benchmark in a learned jet embedding, including the uncertainty keeps the same model trustworthy on test sets an order of magnitude larger than the standard test allows.

  2. Sepsis coagulopathy screening misses its highest-risk patients: 72-hour INR and PTT trajectories recover a group with nearly double the mortality

    Sepsis coagulopathy screening still rests on a single admission lab set, with platelets, INR and SOFA combined into one cross-sectional threshold. Among 3,138 sepsis patients in MIMIC-IV, clustering the joint 72-hour trajectories of platelet count, INR and PTT yielded a high-risk subphenotype with 27.8% mortality, of whom the admission SIC score flagged only 18.1%. The gradient reproduced in an independent refit of 12,828 eICU-CRD patients, where 86.3% of the high group remained SIC-negative.

  3. medRxiv

    Urine metabolites flag cancer without labels: a four-metabolite panel keeps most of supervised separation across eight cancers

    A four-metabolite urine and plasma panel separated cancer cases from controls across eight cancer types, and a model that never used outcome labels still reached an AUC of 0.925 against 0.991 for supervised learning.

  4. arXiv

    LSL gates weight updates to the current data's support with a GMM, letting a 7B-parameter LLM keep 96.6% of pretrained capability after multi-phase finetuning

    The authors propose Local Support Learning (LSL), which frames catastrophic forgetting as a geometric problem in each weight matrix's input space and uses a Gaussian Mixture Model gate to enable a weight adapter only on the support of the current phase's training distribution, so that LLMs of up to 7 billion parameters can be finetuned across multiple phases without access to prior data while learning new tasks at full capacity and retaining pretrained and previously finetuned capabilities.

You’re all caught up