Trajectory Analysis and Representation Learning

The field of trajectory analysis is moving towards more efficient and effective methods for similarity computation, representation learning, and generation. Researchers are exploring new approaches to capture the complexities of trajectory data, including the use of contrastive learning, graph-prototypical frameworks, and variational autoencoders. These innovations are leading to significant improvements in accuracy, robustness, and computational efficiency. Notable papers in this area include MovSemCL, which proposes a movement-semantics contrastive learning framework for trajectory similarity computation, and Pathlet Variational Auto-Encoder, which introduces a deep generative model for robust trajectory generation. Other noteworthy papers include Region-Point Joint Representation, ST-ProC, Blurred Encoding, and GeoPTH, which all contribute to the advancement of trajectory analysis and representation learning.

Sources

MovSemCL: Movement-Semantics Contrastive Learning for Trajectory Similarity

Linear time small coresets for k-mean clustering of segments with applications

Region-Point Joint Representation for Effective Trajectory Similarity Learning

ST-ProC: A Graph-Prototypical Framework for Robust Semi-Supervised Travel Mode Identification

Blurred Encoding for Trajectory Representation Learning

Pathlet Variational Auto-Encoder for Robust Trajectory Generation

GeoPTH: A Lightweight Approach to Category-Based Trajectory Retrieval via Geometric Prototype Trajectory Hashing

Built with on top of