The ALICE experiment will run with continuous readout at interaction rates of up to 50 kHz in Pb-Pb collisions during Run 3 of the LHC. In order to achieve this goal, a new data processing scheme and software are developed. This scheme strongly relies on GPUs (Graphics Processing Unit) for fast online and offline calibration and reconstruction as well as on efficient data compression. On the hardware side, the Time Projection Chamber (TPC), among other detector systems, received major upgrades to its readout chambers and readout electronics. The multiwire proportional chambers were replaced by stacks of four Gas Electron Multiplier foils to allow for continuous readout while keeping the ion backflow below 1%, minimizing space-charge effects from amplification ions entering the drift volume. Nevertheless, significant space-point distortions due to space charge are expected at the highest interaction rates in Pb-Pb collisions. In addition, space-charge density fluctuations lead to distortion fluctuations which have to be corrected on time scales of 10 ms in order to preserve the intrinsic tracking resolution of the TPC. While the average space-charge distortions can be corrected using information from external detectors as a reference, data-driven machine learning algorithms and convolutional neural networks are foreseen to provide the correction for the distortion fluctuations.