Spatio-temporal trajectories play a vital role in various spatio-temporal
data mining tasks. Developing a versatile trajectory learning approach that can
adapt to different tasks while ensuring high accuracy is crucial. This requires
effectively extracting movement patterns and travel purposes embedded in
trajectories. However, this task is challenging due to limitations in the size
and quality of available trajectory datasets. On the other hand, pre-trained
language models (PLMs) have shown great success in adapting to different tasks
by training on large-scale, high-quality corpus datasets. Given the
similarities between trajectories and sentences, there is potential in
leveraging PLMs to enhance the development of a versatile and effective
trajectory learning method. Nevertheless, vanilla PLMs are not tailored to
handle the unique spatio-temporal features present in trajectories and lack the
capability to extract movement patterns and travel purposes from them.
To overcome these obstacles, we propose a model called PLM4Traj that
effectively utilizes PLMs to model trajectories. PLM4Traj leverages the
strengths of PLMs to create a versatile trajectory learning approach while
addressing the limitations of vanilla PLMs in modeling trajectories. Firstly,
PLM4Traj incorporates a novel trajectory semantic embedder that enables PLMs to
process spatio-temporal features in trajectories and extract movement patterns
and travel purposes from them. Secondly, PLM4Traj introduces a novel trajectory
prompt that integrates movement patterns and travel purposes into PLMs, while
also allowing the model to adapt to various tasks. Extensive experiments
conducted on two real-world datasets and two representative tasks demonstrate
that PLM4Traj successfully achieves its design goals. Codes are available at
this https URL

通过利用预训练语言模型（PLM）建立一个名为 PLM4Traj 的模型，该模型能够适应不同任务，有效地提取出嵌入在轨迹中的运动模式和旅行目的，从而解决了原始 PLMs 在建模轨迹时的不足，为轨迹学习提供了通用的解决方案。

PLM4Traj：使用预训练语言模型识别轨迹的运动模式与旅行目的

PLM4Traj: Cognizing Movement Patterns and Travel Purposes from  Trajectories with Pre-trained Language Models

Recent approaches to render photorealistic views from a limited set of
photographs have pushed the boundaries of our interactions with pictures of
static scenes. The ability to recreate moments, that is, time-varying
sequences, is perhaps an even more interesting scenario, but it remains largely
unsolved. We introduce DCT-NeRF, a coordinatebased neural representation for
dynamic scenes. DCTNeRF learns smooth and stable trajectories over the input
sequence for each point in space. This allows us to enforce consistency between
any two frames in the sequence, which results in high quality reconstruction,
particularly in dynamic regions.

本论文提出了 DCT-NeRF，一种基于坐标的神经表示方法，用于动态场景的高质量重建，它学习空间中每个点的平稳和稳定轨迹，并强制执行序列中任意两帧之间的一致性。