1: \begin{abstract}

2: We study decentralized non-convex finite-sum minimization problems described over a network of nodes, where each node possesses a local batch of data samples. {\color{black}In this context, we analyze a single-timescale randomized incremental gradient method, called~\textbf{\texttt{GT-SAGA}}.} \textbf{\texttt{GT-SAGA}} is computationally efficient as it evaluates one component gradient per node per iteration and achieves provably fast and robust performance by leveraging node-level variance reduction and network-level gradient tracking. For general smooth non-convex problems, we show the almost sure and mean-squared convergence of \textbf{\texttt{GT-SAGA}} to a first-order stationary point and further describe regimes of practical significance where it outperforms the existing approaches and achieves a network topology-independent iteration complexity respectively. When the global function satisfies the Polyak-\L ojaciewisz condition, we show that~\textbf{\texttt{GT-SAGA}} exhibits linear convergence to an optimal solution in expectation and describe regimes of practical interest where the performance is network topology-independent and improves upon the existing methods. {\color{black}Numerical experiments are included to highlight the main convergence aspects of \textbf{\texttt{GT-SAGA}} in non-convex settings.} 

3: \end{abstract}

4: