Anomaly categorization & design of synthetic evaluation dataset

Result code in IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F68407700%3A21230%2F16%3A00236674" target="_blank" >RIV/68407700:21230/16:00236674 - isvavai.cz</a>

Result on the web

<a href="https://github.com/breznak/neural.benchmark" target="_blank" >https://github.com/breznak/neural.benchmark</a>

DOI - Digital Object Identifier

—

Result language

angličtina

Original language name

Anomaly categorization & design of synthetic evaluation dataset

Original language description

We design a categorisation of anomalies into distinct classes and create synthetic datasets that aim on a single anomaly category, allowing us to stress specific features of our anomaly detection models, this is in contrast with commonly available rea-world (annotated) datasets. We are aiming to thouroughly benchmark and compare ML algorithms (with current focus on HTM), by designing specialized synthetic datasets that stress a single feature and can be well evaluated and understood. For users able to decide where each algorithm has its strong/weak-spots and help them decide in application for real-world problems. This can also work as a benchmark to evaluate development impact of proposed changes to the algorithms. Goals of this project include: This repository should be a collection of datasets ( real-world, synthetic); papers; algorithm implementations (with initial focus on HTM from NuPIC, but we will gladly include any other algorithms/results.); results (as CSV, image); collection of ideas in the Issues

Czech name

—

Czech description

—

Type

A - Audiovisual production

CEP classification

IN - Informatics

OECD FORD branch

—

Project

—

Continuities

S - Specificky vyzkum na vysokych skolach

Publication year

2016

Confidentiality

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

ISBN

—

Place of publication

—

Publisher/client name

—

Version

—

Carrier ID

—