Bridging Visual Communication and Data Exploration through Pose-Driven Query Synthesis
Keywords: Code Completion, Transfer Learning
TL;DR: This paper introduces R-KinetiQuery, a groundbreaking framework for domain-adaptive sign language to SQL query translation
Abstract: SQL is widely used for managing relational databases and conducting interactive data analysis. Now, various natural language interfaces have emerged, designed to simplify the process of crafting SQL queries by translating natural language commands into executable SQL-Code. However, the communication preferences of the deaf and hard-of-hearing community have been largely overlooked.
This paper introduces R-KinetiQuery, a groundbreaking framework for domain-adaptive sign language to SQL query translation, underpinned by a rigorous mathematical foundation synthesizing functional analysis, ergodic theory, and information geometry. At its core, R-KinetiQuery addresses the fundamental challenge of domain adaptation in the context of multimodal language translation, specifically tailored to bridge the gap between sign language communication and database query languages. A key innovation lies in our application of ergodic theory to analyze the long-term behavior of R-KinetiQuery under domain shift. We establish an ergodic theorem for the model's time-averaged operator, demonstrating its convergence to the expected behavior across domains. This result provides a robust foundation for the model's stability and adaptability in non-stationary environments. Our information-theoretic analysis reveals a deep connection between R-KinetiQuery and the Information Bottleneck principle. We derive a variational bound that explicitly quantifies the trade-off between compression and prediction in the model's latent representation, providing insights into its domain-invariant feature learning.
Empirically, we demonstrate R-KinetiQuery's superior performance on a diverse set of domain adaptation tasks, consistently outperforming state-of-the-art baselines. Our experiments span a wide range of domain shifts, from subtle variations in sign language dialects to dramatic changes in database schemas and query complexities.
Primary Area: generative models
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.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
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: 13045
Loading