-
This Looks Better than That: Better Interpretable Models with ProtoPNeXt
Authors:
Frank Willard,
Luke Moffett,
Emmanuel Mokel,
Jon Donnelly,
Stark Guo,
Julia Yang,
Giyoung Kim,
Alina Jade Barnett,
Cynthia Rudin
Abstract:
Prototypical-part models are a popular interpretable alternative to black-box deep learning models for computer vision. However, they are difficult to train, with high sensitivity to hyperparameter tuning, inhibiting their application to new datasets and our understanding of which methods truly improve their performance. To facilitate the careful study of prototypical-part networks (ProtoPNets), w…
▽ More
Prototypical-part models are a popular interpretable alternative to black-box deep learning models for computer vision. However, they are difficult to train, with high sensitivity to hyperparameter tuning, inhibiting their application to new datasets and our understanding of which methods truly improve their performance. To facilitate the careful study of prototypical-part networks (ProtoPNets), we create a new framework for integrating components of prototypical-part models -- ProtoPNeXt. Using ProtoPNeXt, we show that applying Bayesian hyperparameter tuning and an angular prototype similarity metric to the original ProtoPNet is sufficient to produce new state-of-the-art accuracy for prototypical-part models on CUB-200 across multiple backbones. We further deploy this framework to jointly optimize for accuracy and prototype interpretability as measured by metrics included in ProtoPNeXt. Using the same resources, this produces models with substantially superior semantics and changes in accuracy between +1.3% and -1.5%. The code and trained models will be made publicly available upon publication.
△ Less
Submitted 20 June, 2024;
originally announced June 2024.
-
ProtoEEGNet: An Interpretable Approach for Detecting Interictal Epileptiform Discharges
Authors:
Dennis Tang,
Frank Willard,
Ronan Tegerdine,
Luke Triplett,
Jon Donnelly,
Luke Moffett,
Lesia Semenova,
Alina Jade Barnett,
Jin Jing,
Cynthia Rudin,
Brandon Westover
Abstract:
In electroencephalogram (EEG) recordings, the presence of interictal epileptiform discharges (IEDs) serves as a critical biomarker for seizures or seizure-like events.Detecting IEDs can be difficult; even highly trained experts disagree on the same sample. As a result, specialists have turned to machine-learning models for assistance. However, many existing models are black boxes and do not provid…
▽ More
In electroencephalogram (EEG) recordings, the presence of interictal epileptiform discharges (IEDs) serves as a critical biomarker for seizures or seizure-like events.Detecting IEDs can be difficult; even highly trained experts disagree on the same sample. As a result, specialists have turned to machine-learning models for assistance. However, many existing models are black boxes and do not provide any human-interpretable reasoning for their decisions. In high-stakes medical applications, it is critical to have interpretable models so that experts can validate the reasoning of the model before making important diagnoses. We introduce ProtoEEGNet, a model that achieves state-of-the-art accuracy for IED detection while additionally providing an interpretable justification for its classifications. Specifically, it can reason that one EEG looks similar to another ''prototypical'' EEG that is known to contain an IED. ProtoEEGNet can therefore help medical professionals effectively detect IEDs while maintaining a transparent decision-making process.
△ Less
Submitted 3 December, 2023;
originally announced December 2023.
-
Centromere reference models for human chromosomes X and Y satellite arrays
Authors:
Karen H. Miga,
Yulia Newton,
Miten Jain,
Nicolas Altemose,
Huntington F. Willard,
W. James Kent
Abstract:
The human genome remains incomplete, with multi-megabase sized gaps representing the endogenous centromeres and other heterochromatic regions. These regions are commonly enriched with long arrays of near-identical tandem repeats, known as satellite DNAs, that offer a limited number of variant sites to differentiate individual repeat copies across millions of bases. This substantial sequence homoge…
▽ More
The human genome remains incomplete, with multi-megabase sized gaps representing the endogenous centromeres and other heterochromatic regions. These regions are commonly enriched with long arrays of near-identical tandem repeats, known as satellite DNAs, that offer a limited number of variant sites to differentiate individual repeat copies across millions of bases. This substantial sequence homogeneity challenges available assembly strategies, and as a result, centromeric regions are omitted from ongoing genomic studies. To address this problem, we present a locally ordered assembly across two haploid human satellite arrays on chromosomes X and Y, resulting in an initial linear representation of 3.83 Mb of centromeric DNA within an individual genome. To further expand the utility of each centromeric reference sequence, we evaluate sites within the arrays for short-read mappability and chromosome specificity. As satellite DNAs evolve in a concerted manner, we use these centromeric assemblies to assess the extent of sequence variation among 372 individuals from distinct human populations. In doing so, we identify two ancient satellite array variants in both X and Y centromeres as determined by array length and sequence composition. This study provides an initial linear representation and comprehensive sequence characterization of a regional centromere and establishes a foundation to extend genomic characterization to these sites as well as to other repeat-rich regions within complex genomes.
△ Less
Submitted 17 September, 2013; v1 submitted 28 June, 2013;
originally announced July 2013.
