{"user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","month":"04","_id":"14175","article_processing_charge":"No","language":[{"iso":"eng"}],"publication_status":"published","author":[{"full_name":"Makansi, Osama","first_name":"Osama","last_name":"Makansi"},{"first_name":"Julius von","full_name":"Kügelgen, Julius von","last_name":"Kügelgen"},{"last_name":"Locatello","orcid":"0000-0002-4850-0683","id":"26cfd52f-2483-11ee-8040-88983bcc06d4","full_name":"Locatello, Francesco","first_name":"Francesco"},{"full_name":"Gehler, Peter","first_name":"Peter","last_name":"Gehler"},{"last_name":"Janzing","full_name":"Janzing, Dominik","first_name":"Dominik"},{"full_name":"Brox, Thomas","first_name":"Thomas","last_name":"Brox"},{"last_name":"Schölkopf","first_name":"Bernhard","full_name":"Schölkopf, Bernhard"}],"date_updated":"2023-09-11T09:52:20Z","main_file_link":[{"url":"https://doi.org/10.48550/arXiv.2110.05304","open_access":"1"}],"type":"conference","oa":1,"status":"public","conference":{"name":"ICLR: International Conference on Learning Representations","start_date":"2022-04-25","end_date":"2022-04-29","location":"Virtual"},"extern":"1","day":"25","abstract":[{"lang":"eng","text":"Predicting the future trajectory of a moving agent can be easy when the past trajectory continues smoothly but is challenging when complex interactions with other agents are involved. Recent deep learning approaches for trajectory prediction show promising performance and partially attribute this to successful reasoning about agent-agent interactions. However, it remains unclear which features such black-box models actually learn to use for making predictions. This paper proposes a procedure that quantifies the contributions\r\nof different cues to model performance based on a variant of Shapley values. Applying this procedure to state-of-the-art trajectory prediction methods on standard benchmark datasets shows that they are, in fact, unable to reason about interactions. Instead, the past trajectory of the target is the only feature used for predicting its future. For a task with richer social\r\ninteraction patterns, on the other hand, the tested models do pick up such interactions to a certain extent, as quantified by our feature attribution method. We discuss the limits of the proposed method and its links to causality."}],"external_id":{"arxiv":["2110.05304"]},"date_published":"2022-04-25T00:00:00Z","year":"2022","quality_controlled":"1","title":"You mostly walk alone: Analyzing feature attribution in trajectory prediction","date_created":"2023-08-22T14:02:34Z","oa_version":"Preprint","publication":"10th International Conference on Learning Representations","department":[{"_id":"FrLo"}],"citation":{"mla":"Makansi, Osama, et al. “You Mostly Walk Alone: Analyzing Feature Attribution in Trajectory Prediction.” 10th International Conference on Learning Representations, 2022.","ama":"Makansi O, Kügelgen J von, Locatello F, et al. You mostly walk alone: Analyzing feature attribution in trajectory prediction. In: 10th International Conference on Learning Representations. ; 2022.","short":"O. Makansi, J. von Kügelgen, F. Locatello, P. Gehler, D. Janzing, T. Brox, B. Schölkopf, in:, 10th International Conference on Learning Representations, 2022.","ieee":"O. Makansi et al., “You mostly walk alone: Analyzing feature attribution in trajectory prediction,” in 10th International Conference on Learning Representations, Virtual, 2022.","ista":"Makansi O, Kügelgen J von, Locatello F, Gehler P, Janzing D, Brox T, Schölkopf B. 2022. You mostly walk alone: Analyzing feature attribution in trajectory prediction. 10th International Conference on Learning Representations. ICLR: International Conference on Learning Representations.","chicago":"Makansi, Osama, Julius von Kügelgen, Francesco Locatello, Peter Gehler, Dominik Janzing, Thomas Brox, and Bernhard Schölkopf. “You Mostly Walk Alone: Analyzing Feature Attribution in Trajectory Prediction.” In 10th International Conference on Learning Representations, 2022.","apa":"Makansi, O., Kügelgen, J. von, Locatello, F., Gehler, P., Janzing, D., Brox, T., & Schölkopf, B. (2022). You mostly walk alone: Analyzing feature attribution in trajectory prediction. In 10th International Conference on Learning Representations. Virtual."}}