Guardado en:
Detalles Bibliográficos
Autores principales: Shin, Changho, Yan, Xinya, Jo, Suenggwan, Cho, Sungjun, Chaudhuri, Shourjo Aditya, Sala, Frederic
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:https://arxiv.org/abs/2503.18693
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866912292384276480
author Shin, Changho
Yan, Xinya
Jo, Suenggwan
Cho, Sungjun
Chaudhuri, Shourjo Aditya
Sala, Frederic
author_facet Shin, Changho
Yan, Xinya
Jo, Suenggwan
Cho, Sungjun
Chaudhuri, Shourjo Aditya
Sala, Frederic
contents Language models often struggle with temporal misalignment, performance degradation caused by shifts in the temporal distribution of data. Continuously updating models to avoid degradation is expensive. Can models be adapted without updating model weights? We present TARDIS, an unsupervised representation editing method that addresses this challenge. TARDIS extracts steering vectors from unlabeled data and adjusts the model's representations to better align with the target time period's distribution. Our experiments reveal that TARDIS enhances downstream task performance without the need for fine-tuning, can mitigate temporal misalignment even when exact target time period data is unavailable, and remains efficient even when the temporal information of the target data points is unknown at inference time.
format Preprint
id arxiv_https___arxiv_org_abs_2503_18693
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle TARDIS: Mitigating Temporal Misalignment via Representation Steering
Shin, Changho
Yan, Xinya
Jo, Suenggwan
Cho, Sungjun
Chaudhuri, Shourjo Aditya
Sala, Frederic
Machine Learning
Language models often struggle with temporal misalignment, performance degradation caused by shifts in the temporal distribution of data. Continuously updating models to avoid degradation is expensive. Can models be adapted without updating model weights? We present TARDIS, an unsupervised representation editing method that addresses this challenge. TARDIS extracts steering vectors from unlabeled data and adjusts the model's representations to better align with the target time period's distribution. Our experiments reveal that TARDIS enhances downstream task performance without the need for fine-tuning, can mitigate temporal misalignment even when exact target time period data is unavailable, and remains efficient even when the temporal information of the target data points is unknown at inference time.
title TARDIS: Mitigating Temporal Misalignment via Representation Steering
topic Machine Learning
url https://arxiv.org/abs/2503.18693