Difference: Real-BogusClassifications (14 vs. 15)

Revision 152020-10-23 - AshishMahabal

Line: 1 to 1

META TOPICPARENT	name="MLRoad-map"

Added:

>
>

This page is mostly about the RF work led by Umaa Rebbapragada (included in https://arxiv.org/abs/1902.01936).

The currently deployed model is braai that uses deep learning and is published here: https://arxiv.org/abs/1907.11259 with the associated GitHub including notebooks.

Overview

The Real-Bogus classifier scores sources on a scale of 0 (bogus) to 1 (real). It is currently a Random Forest classifier that is built upon 'features', which are a collection of statistics and outputs of the real-time data pipeline. The classifier is trained on a set of labeled data. Labels are provided via two data collection venues: 1) Zooniverse and 2) GROWTH marshall.

Line: 47 to 51

- Classifier error analysis: What do low RB reals look like, what do high RB bogus look like?
- Report feature importance by correlated feature groups

Data Collection

Changed:

<
<

Automated Data Contamination [ Charlotte ] * Find contamination using clustering.
Active Learning to Improve Training Data Selection [ Sara ] * Use active learning to discover potential batches of boguses (and reals, alternatively)

>
>

Automated Data Contamination [ Charlotte ] * Find contamination using clustering.
Active Learning to Improve Training Data Selection [ Sara ] * Use active learning to discover potential batches of boguses (and reals, alternatively)

Other Data Collection Sources (preferably automated)
- Find variables (bogus objects that have multiple alert packets, objects >= n_obs, objects with both positive and negative subtractions?)
- Automate cross matches from relevant catalogs (e.g., TNS)

View topic | History: r15 < r14 < r13 < r12 | More topic actions...