vabh

function up {
    cd `expr "$PWD" : "^\(.*$1[^/]*\)"`
}

vabh

Chen, Tianqi, Ian Goodfellow, and Jonathon Shlens. "Net2Net: Accelerating Learning via Knowledge Transfer." (2016). [http://arxiv.org/abs/1511.05641]

• Knowledge transfer scheme, from one model to another. Knowledge transfer throug copying weights.

• Aim is to reduce training time and give the 'student' model and good starting point to begin its learning process. Student model begins to learn, from where the teacher left off. [critical in real life when several models are stored. Espcially when 'searching' for the best model]

• Significant for 'life-long' learning, changes in model data become available (more data for categories or addition of categories)

vabh

Train a model
caffe_root/build/tools/caffe train -solver [path_to_solver_file] [Trained models are saved in a location specified in the solver file]

accuracy_cifar.py - generates a file with (iterations, accuracy)
python accuracy_cifar.py [network_definition] [model_prefix] [gpu]
network_deifinition - the .prototxt file which has referenced in the solver. This contains the network which was trained.
model_prefix - this is the prefix with which models were saved throughout the training phase.

For both network_definition and model_prefix, the root folders are specified inside the script.

vabh

local sys = require 'sys'
local image = require 'image'
local paths = require 'paths'

local M = {}


local function loadAndConvertImages(imageList, baseDir, numExamples)
  local data = torch.Tensor(numExamples,3,256,256)
  local labels = torch.Tensor(numExamples)

vabh

vabh

class ConvLSTMCell(nn.Module):
    def __init__(self, in_channels, out_channels, kernel_size, padding=1):
        super(ConvLSTMCell, self).__init__()
        
        self.k = kernel_size
        self.in_channels = in_channels
        self.out_channels = out_channels
        self.padding = padding
        
        self.w_i = nn.Parameter(torch.Tensor(4*out_channels, in_channels, kernel_size, kernel_size))