This paper introduces InceptionTime, a deep learning ensemble for Time Series Classification (TSC) that aims to bridge the gap between the accuracy of state-of-the-art algorithms like HIVE-COTE and the scalability of deep learning models. InceptionTime is inspired by the Inception-v4 architecture and consists of five deep Convolutional Neural Network (CNN) models, each with a unique set of randomly initialized weights. The core of InceptionTime is the Inception module, which applies multiple filters of varying lengths to the input time series, allowing the network to automatically extract relevant features from both long and short time series. The authors demonstrate that InceptionTime achieves state-of-the-art accuracy on the UCR archive, a benchmark for TSC, while also being significantly more scalable than HIVE-COTE. InceptionTime can learn from 1,500 time series in one hour and 8 million time series in 13 hours, a capability far beyond the reach of HIVE-COTE. The paper also includes a detailed analysis of the architectural hyperparameters and the characteristics of the Inception module, providing insights into why InceptionTime is so successful.This paper introduces InceptionTime, a deep learning ensemble for Time Series Classification (TSC) that aims to bridge the gap between the accuracy of state-of-the-art algorithms like HIVE-COTE and the scalability of deep learning models. InceptionTime is inspired by the Inception-v4 architecture and consists of five deep Convolutional Neural Network (CNN) models, each with a unique set of randomly initialized weights. The core of InceptionTime is the Inception module, which applies multiple filters of varying lengths to the input time series, allowing the network to automatically extract relevant features from both long and short time series. The authors demonstrate that InceptionTime achieves state-of-the-art accuracy on the UCR archive, a benchmark for TSC, while also being significantly more scalable than HIVE-COTE. InceptionTime can learn from 1,500 time series in one hour and 8 million time series in 13 hours, a capability far beyond the reach of HIVE-COTE. The paper also includes a detailed analysis of the architectural hyperparameters and the characteristics of the Inception module, providing insights into why InceptionTime is so successful.