Experiments

From Robin

(Difference between revisions)
Jump to: navigation, search
Line 1: Line 1:
 +
== Search for the first path? ==
 +
=== Research question ===
 +
Is there anything to gain from performing a proper search for the first path versus just picking a random path and training the weights?
 +
 +
=== Hypothesis ===
 +
I think performance will have the same asymptote, but it will be reached in fewer training iterations. The only thing that might be
 +
influenced by this path selection is that the modules in PathNet might have more interconnected dependencies. Maybe the layers are more
 +
"independent" when the weights are updated as part of multiple paths? This might be important for transferability when learning future tasks.
 +
 +
=== Suggested experiment ===
 +
Performing multiple small multi-task learning scenarios. Two tasks should be enough, but it is necessary to show that modules are reused in each scenario.
 +
Test both picking a path and the full-on search for a path and compare convergence time for the second task.
 +
 +
=== Implemented experiment ===
 +
* Information about the execution of the experiment
 +
 +
=== Results ===
 +
* Plots and table contents showing experiment results
 +
 +
=== Conclusion ===
 +
* What does the results tell me
 +
 +
 +
 +
 +
== Format ==
== Format ==
=== Research question ===
=== Research question ===
Line 7: Line 33:
=== Suggested experiment ===
=== Suggested experiment ===
-
* How are the experiment to be performed?
+
* How is the experiment to be performed?
* What metrics are going to be used?
* What metrics are going to be used?

Revision as of 11:50, 8 November 2017

Contents

Search for the first path?

Research question

Is there anything to gain from performing a proper search for the first path versus just picking a random path and training the weights?

Hypothesis

I think performance will have the same asymptote, but it will be reached in fewer training iterations. The only thing that might be influenced by this path selection is that the modules in PathNet might have more interconnected dependencies. Maybe the layers are more "independent" when the weights are updated as part of multiple paths? This might be important for transferability when learning future tasks.

Suggested experiment

Performing multiple small multi-task learning scenarios. Two tasks should be enough, but it is necessary to show that modules are reused in each scenario. Test both picking a path and the full-on search for a path and compare convergence time for the second task.

Implemented experiment

  • Information about the execution of the experiment

Results

  • Plots and table contents showing experiment results

Conclusion

  • What does the results tell me



Format

Research question

  • What do I want to test/know/research?

Hypothesis

  • What do I think my results will be and why?

Suggested experiment

  • How is the experiment to be performed?
  • What metrics are going to be used?

Implemented experiment

  • Information about the execution of the experiment

Results

  • Plots and table contents showing experiment results

Conclusion

  • What does the results tell me
Personal tools
Front page