search

snuspl / dolphin

14 stars 2 forks source link

Add multilayer perceptron algorithm. #60

Closed jsryu21 closed 8 years ago

jsryu21 commented 9 years ago

Before implementing convolutional neural network, I suggest that we could implement multiplayer perceptron first.

There're some implementations of multilayer perceptron.

  1. C version : Neural Networks for Face Recognition http://www.cs.cmu.edu/afs/cs.cmu.edu/user/mitchell/ftp/faces.html
  2. Python version : a fully connected neural network https://github.com/jorgenkg/python-neural-network http://arctrix.com/nas/python/bpnn.py

At this present time, there's not gorgeous method to transfer data across partition boundary for horizontal or vertical partitioning, but we could use group communication with master & slave model in dolphin.

bgchun commented 9 years ago

It'd be nice to create a generic nn representation which we can use to create a specific nn. The code should not be specific to mlp.

I also would like to suggest not to be restricted by group communication. It'd be good to assume that you will get a communication layer that allows arbitrary communication and create a design.

On Tue, Jul 7, 2015 at 4:24 PM, jsryu21 notifications@github.com wrote:

Before implementing convolutional neural network, I suggest that we could implement multiplayer perceptron first.

There're some implementations of multilayer perceptron.

  1. C version : Neural Networks for Face Recognition http://www.cs.cmu.edu/afs/cs.cmu.edu/user/mitchell/ftp/faces.html
  2. Python version : a fully connected neural network https://github.com/jorgenkg/python-neural-network http://arctrix.com/nas/python/bpnn.py

At this present time, there's not gorgeous method to transfer data across partition boundary for horizontal or vertical partitioning, but we could use group communication with master & slave model in dolphin.

— Reply to this email directly or view it on GitHub https://github.com/cmssnu/dolphin/issues/60.

Byung-Gon Chun

bgchun commented 9 years ago

A mini-design review item for tomorrow's meeting?

On Tue, Jul 7, 2015 at 5:34 PM, Byung-Gon Chun bgchun@gmail.com wrote:

It'd be nice to create a generic nn representation which we can use to create a specific nn. The code should not be specific to mlp.

I also would like to suggest not to be restricted by group communication. It'd be good to assume that you will get a communication layer that allows arbitrary communication and create a design.

On Tue, Jul 7, 2015 at 4:24 PM, jsryu21 notifications@github.com wrote:

Before implementing convolutional neural network, I suggest that we could implement multiplayer perceptron first.

There're some implementations of multilayer perceptron.

  1. C version : Neural Networks for Face Recognition http://www.cs.cmu.edu/afs/cs.cmu.edu/user/mitchell/ftp/faces.html
  2. Python version : a fully connected neural network https://github.com/jorgenkg/python-neural-network http://arctrix.com/nas/python/bpnn.py

At this present time, there's not gorgeous method to transfer data across partition boundary for horizontal or vertical partitioning, but we could use group communication with master & slave model in dolphin.

— Reply to this email directly or view it on GitHub https://github.com/cmssnu/dolphin/issues/60.

Byung-Gon Chun

Byung-Gon Chun

jsryu21 commented 9 years ago

Yes, and for prototyping.

jsjason commented 8 years ago

Instead of trying to implement a REEF multilayer perceptron, we've decided to start with a naive neural network with only fully connected layers in #66. Thus, this issue won't be addressed.