Specifically, we show that the latent privacy features are still retained in the obfuscated representations.

The proposed MLP decoder aggregates information from different layers, and thus combining both local attention and global attention to render powerful representations. We show that this simple and lightweight design is the key to efficient segmentation on Transformers.

"shortcut learning": when a predictor relies on shortcuts, i.e., spurious correlations between the inputs

"shortcut learning": when a predictor relies on shortcuts, i.e., spurious correlations between the inputs

But these methods are unable to improve throughput (frames-per-second) on real-life hardware while simultaneously preserving robustness to adversarial perturbations.

In this section, we will give an overview of the related literature in time series forecasting. ARIMA Box & Jenkins ( 1968); Box & Pierce ( 1970) follows the Markov process and build recursive sequential forecasting. Temporal convolutional network (TCN) Sen et al. ( 2019) is another family for sequential tasks. Convolution is a parallelizable operation but expensive in inference. Some works use temporal attention Qin et al. ( 2017) to capture long-range Others use the backbone of Transformer.

Long-term forecasting refers to making predictions based on the history for a long horizon in the future, as opposed to short-term forecasting. Long-term time series forecasting has many key applications in energy, weather, economics, transportation, and so on.