Efficient on-device learning for resource-constrained devices - #326

Open

christophe0606 wants to merge 19 commits into

mainfrom

Commits on Aug 14, 2026

Preliminary experimentations with on-device machine learning
christophe0606
committed
9f21f26
autodiff : Added new operators and CMSIS-DSP optimizations.
christophe0606
committed
98ea3bd
dsppp : Added new transpose_view operator and used it in backward pass of fully connected
christophe0606
committed
4c903e3
autodiff : Added Iris classification example
christophe0606
committed
aa7b998
autodiff : Added a dropout operator
christophe0606
committed
bbcf453
autodiff : added matrix multiply
christophe0606
committed
a4591dd
dsppp : Add a new lazy matvec operator to enable fusion on mat x vector
christophe0606
committed
4e49102
autodiff: Remove debug code in mat mul operator
christophe0606
committed
3d99e99
Don't force cmsis toolbox version in dsppp csolution file
christophe0606
committed
53ea969
Update autodiff documentation
christophe0606
committed
48456ad
Add more error handling to autodiff_iris.cpp
christophe0606
committed
f88cddb
Added f16 implementation of autodiff and iris example now uses f16
christophe0606
committed
f907d42
Refactor validation in autodiff operators
christophe0606
committed
a0800c5
autodiff : Add SGD optimizer and LMS filter example
christophe0606
committed
ab59af9
Added f16 tests for autodiff
christophe0606
committed
17a041d
autodiff : Added quantize / dequantize operators
christophe0606
committed
f398285
autodiff : Update README with new example
christophe0606
committed
f6fbd11
autodiff : Added optimization for quantize / dequantize
christophe0606
committed
b42339d
autodiff : Improve documentation for quantization aware training and Ethos
christophe0606
committed
36f82cf