1: \begin{abstract}

2: 

3: Ant Colony Optimization (ACO) is renowned for its effectiveness in solving Traveling Salesman Problems, yet it faces computational challenges in CPU-based environments, particularly with large-scale instances.

4: In response, we introduce a Tensorized Ant Colony Optimization (TensorACO) to utilize the advancements of GPU acceleration.

5: As the core, TensorACO fully transforms ant system and ant path into tensor forms, a process we refer to as \emph{tensorization}.

6: For the tensorization of ant system, we propose a preprocessing method to reduce the computational overhead by calculating the probability transition matrix.

7: In the tensorization of ant path, we propose an index mapping method to accelerate the update of pheromone matrix by replacing the mechanism of sequential path update with parallel matrix operations.

8: Additionally, we introduce an Adaptive Independent Roulette (AdaIR) method to overcome the challenges of parallelizing ACO's selection mechanism on GPUs.

9: Comprehensive experiments demonstrate the superior performance of TensorACO achieving up to 1921$\times$ speedup over standard ACO.

10: Moreover, the AdaIR method further improves TensorACO's convergence speed by 80\% and solution quality by 2\%. Source codes are available at \href{https://github.com/EMI-Group/tensoraco}{https://github.com/EMI-Group/tensoraco}.

11: \end{abstract}

12: