Initial run results

Created on October 25|Last edited on November 16
Comment
﻿
Overall resultsResults obtained over 20 runs of each of the eight configurations we considered.
Experimental settings: Adam with default settings (NB: Includes weight decay); minibatches of size 500; training for 500 epochs. 
Expansion: Start with 2 hidden units, and expand into 4.
﻿
Validation accuracy during training
Validation accuracy during training
Computing group metrics from first 10 groups
020406080Epoch0.20.30.40.50.60.7Validation accuracy
run_type: U[copy, random; FR:True] W[copy, random; FR:True]   val_accuracy_pretrain
run_type: U[copy, random; FR:False] W[copy, copy; FR:False]   val_accuracy_pretrain
run_type: U[random, random; FR:False] W[copy, copy; FR:False]   val_accuracy_pretrain
run_type: U[random, random; FR:False] W[copy, random; FR:False]   val_accuracy_pretrain
run_type: U[copy, copy; FR:False] W[copy, copy; FR:False]   val_accuracy_pretrain
run_type: U[copy, random; FR:False] W[copy, random; FR:False]   val_accuracy_pretrain
run_type: U[copy, random; FR:True] W[copy, random; FR:True]   val_accuracy_expand
run_type: U[copy, random; FR:False] W[copy, copy; FR:False]   val_accuracy_expand
run_type: U[random, random; FR:False] W[copy, copy; FR:False]   val_accuracy_expand
run_type: U[random, random; FR:False] W[copy, random; FR:False]   val_accuracy_expand
run_type: U[copy, copy; FR:False] W[copy, copy; FR:False]   val_accuracy_expand
run_type: U[copy, random; FR:False] W[copy, random; FR:False]   val_accuracy_expand
exp8-large (20 runs)120
﻿
Effective rank﻿
Run set160
﻿
Most interesting runs; heatmaps﻿
﻿
Run set4
﻿
﻿
Add a comment