Difference between revisions of "LightR"

From Robowiki
Jump to navigation Jump to search
m (Towards differentiable programming)
m (Learned models -> learned systems)
Line 10: Line 10:
 
; Design principle
 
; Design principle
 
: Strategy light, machine learning heavy.  
 
: Strategy light, machine learning heavy.  
 +
 +
; Central goal
 +
: Learned models -> learned systems
  
 
; Planned experiments
 
; Planned experiments
Line 15: Line 18:
 
:: Multiple hand-tuned danger models -> Expert model & gate model
 
:: Multiple hand-tuned danger models -> Expert model & gate model
 
:: Hand-crafted features with naive KNN -> Search-based sequence model
 
:: Hand-crafted features with naive KNN -> Search-based sequence model
:: Slow networks -> Knowledge distill & quantization aware training
 
 
:: Offline pre-training & online fine-tuning of everything above.  
 
:: Offline pre-training & online fine-tuning of everything above.  
 
: Towards differentiable programming:
 
: Towards differentiable programming:
:: Directly optimizing max escape angle & prior probability of getting hit (distancing)
+
:: Directly optimizing prior probability of getting hit (max escape angle & distancing)
 
:: Per-instance level optimization of the above (mea & distancing as part of network)
 
:: Per-instance level optimization of the above (mea & distancing as part of network)
  

Revision as of 07:21, 20 March 2022

LightR Sub-pages:
Version History

This page is under construction. For recent activities, see Version History.


Design principle
Strategy light, machine learning heavy.
Central goal
Learned models -> learned systems
Planned experiments
Towards deep learning:
Multiple hand-tuned danger models -> Expert model & gate model
Hand-crafted features with naive KNN -> Search-based sequence model
Offline pre-training & online fine-tuning of everything above.
Towards differentiable programming:
Directly optimizing prior probability of getting hit (max escape angle & distancing)
Per-instance level optimization of the above (mea & distancing as part of network)