Multiple instance learning: attention to instance classification

Result code in IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F68407700%3A21230%2F25%3A00383614" target="_blank" >RIV/68407700:21230/25:00383614 - isvavai.cz</a>

Result on the web

<a href="https://doi.org/10.1117/12.3045059" target="_blank" >https://doi.org/10.1117/12.3045059</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1117/12.3045059" target="_blank" >10.1117/12.3045059</a>

Result language

angličtina

Original language name

Multiple instance learning: attention to instance classification

Original language description

Multiple instance learning (MIL) is a specific form of weakly-supervised learning where instances with hidden labels are grouped into bags, and only bag labels are observed. MIL models generally fall into one of two classes, focusing on instance or bag classification. Blurring the line between the two classes, an existing attention-based MIL method classifies bags accurately while indicating key instances. We build upon this method and propose to jointly learn a bag and instance classifier, essentially removing the distinction between bag-centric and instance-centric approaches. We performed experiments on the CAMELYON16 dataset of histopathological images and two other image datasets. The experiments showed that our method achieves high bag-level performance, comparable to other competing MIL methods. At the same time, our method outperforms other MIL methods in instance-level classification and, when provided with enough data, achieves results comparable to supervised learning using instance labels.

Czech name

—

Czech description

—

Type

D - Article in proceedings

CEP classification

—

OECD FORD branch

10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

Project

Result was created during the realization of more than one project. More information in the Projects tab.

Continuities

P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)

Publication year

2025

Confidentiality

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Article name in the collection

SPIE Medical Imaging 2025: Image Processing

ISBN

9781510685901

ISSN

1605-7422

e-ISSN

—

Number of pages

9

Pages from-to

—

Publisher name

SPIE

Place of publication

Bellingham (stát Washington)

Event location

San Diego, California,

Event date

Feb 16, 2025

Type of event by nationality

WRD - Celosvětová akce

UT code for WoS article

001487072200029