Strong denoising of financial time-series
Keywords: mutual learning, regularization
TL;DR: a regularization scheme for auto-encoders where two networks compare related but different input data
Abstract: In this paper we introduce a method for improving the signal to noise ratio of financial data. The approach relies on combining a target variable with different context variables and using auto-encoders (AEs) to learn reconstructions of the combined inputs. The idea is to seek agreement among multiple AEs which are trained on related but different inputs for which they are forced to find common ground. The training process is set up as a conversation where models take turns at producing a prediction (speaking) or reconciling own predictions with the output of the other AE (listening), until an agreement is reached. This leads to "mutual regularization" among the AEs. Unlike standard regularization which relies on including a complexity penalty into the loss function, the proposed method uses the partner network to detect and amend the lack of generality in the data representation. As only true regularities can be agreed upon by the AEs, the replication of noise is costly and will therefore be avoided.
Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 5158
Loading