Skip to main content

Table 1 Main features of multi-task metric learning methods

From: A review on multi-task metric learning

Name

Year

Multi-task Strategy

Solver

Dimension Reduction

Side-information

Regularizer

mt-LMNN

2010

Shared composition

Projected gradient descent

No

Triplets

Frobenius norm

TML

2010

Task relationship learning

Alternating Optimization

No

Pairs

Task covariance

mtMLCS

2011

Shared subspace

Gradient descent

Yes

Triplets

-

M2SL

2012

Shared composition

Coordinate gradient descent

No

Pairs

Frobenius norm

GPmtML

2012

Geometry preserving

Alternating optimization

No

Triplets

Von Neumann divergence

mt-SCML

2014

Shared sparse representation

Regularized dual averaging

Yes

Triplets

2/1 norm

MtMCML

2014

Graph regularization

Alternating optimization

No

Pairs

Laplacian

TMTL

2015

Metric weighted sum

Direct calculation

No

Covariance

-

online-SMDM

2016

Shared composition

Online projected gradient descent

No

Pairs

Frobenius norm

CP-mtML

2016

Coupled projection

Stochastic gradient projection

Yes

Pairs

-

DMML

2016

Shared subnetwork

Sub-gradient descent

No

Pairs

-

HMML

2017

Shared composition

Not mentioned

No

Triplets

Trace norm

mtDCML

2017

Shared network

Gradient descent

No

Pairs

-