Welcome to EDAboard.com

Welcome to our site! EDAboard.com is an international Electronics Discussion Forum focused on EDA software, circuits, schematics, books, theory, papers, asic, pld, 8051, DSP, Network, RF, Analog Design, PCB, Service Manuals... and a whole lot more! To participate you need to register. Registration is free. Click here to register now.

What is meant by Manhattan update rule in neural networks ?

Status
Not open for further replies.

vetriece86@gmail.com

Junior Member level 2
Joined
Nov 10, 2014
Messages
20
Helped
0
Reputation
0
Reaction score
0
Trophy points
1
Activity points
146
Could anyone explain me about

1. What is meant by Manhattan update rule ?

2. How to use Manhattan update rule in delta rule ?

Thank you.
 

SunnySkyguy

Advanced Member level 5
Joined
Sep 26, 2007
Messages
6,744
Helped
1,675
Reputation
3,348
Reaction score
1,644
Trophy points
1,413
Location
Richmond Hill, ON, Canada
Activity points
50,737
Been there.. But never tried it ;)

https://www.helwan.edu.eg/university/staff/~yhifny/publications/dnn_crf_MH.pdf

it seems no delta on estimate just polarity.
To converge the estimate STEP SIZE is incremented each time.

seems like a simple but slow method but resilient

like dynamic RAM Virtual Memory size reserve estimates.... Adapt with margin , grow by increasing larger steps at some rate until converge on desired value

, stable, resilient , wide range

not sure if step size reset to base increment on polarity toggle.. Probably
 

vetriece86@gmail.com

Junior Member level 2
Joined
Nov 10, 2014
Messages
20
Helped
0
Reputation
0
Reaction score
0
Trophy points
1
Activity points
146
Even I too read that Pdf. But I didnt understand. If possible, can you explain me in short about Manhattan update rule.
 

Status
Not open for further replies.

Part and Inventory Search

Welcome to EDABoard.com

Sponsor

Top