FL	Federated Learning
HAR	Human Activity Recognition
i.i.d	Independent and identically distributed
KD	Knowledge Distillation
E	The number of local epochs
R	The number of global communication rounds
${\beta }^r$	An integer used to seed the permutation of $D_p$ at round r
${\alpha }^r$	constant $\in [0,1]$ used to calculate $D_{Aug}^r$
C	a set of clients/devices participating in FL
$C_i$	The $i^{th}$ client
$f_i$	Independently designed deep model of the $i^{th}$ client
$f_i^{*}$	$f_i$ minus the last softmax activation layer
$D_i$	The $i^{th}$ local dataset of the $i^{th}$ client
$\{X^{\left[j\right]},Y^{\left[j\right]}\}$	The $j^{th}$ sample pair in a local dataset
$D_t$	test dataset (shared)
$D_p$	the public dataset (shared)
$\left\{X_P^{\left[j\right]}\right\}$	The $j^{th}$ sample in $D_p$
$D_d^r$	$D_p$ permuted using the seed ${\beta }^r$
$D_{Aug}^r$	augmented public dataset. $D_p$ and $D_d^r$ weighted by ${\alpha }^r$.
$P_i^r$	is the accuracy of $f_i$ on $D_t$ at global round r
$S_i^r$	soft labels of $i^{th}$ client on $D_{Aug}^r$ at global round r
$S^r$	soft labels aggregated by the server at global round r