CORTEXA
← Browse
arxivcs.LG2026-07-06

Parameter-Free Encoders Remain Viable for RDB Foundation Models

Linjie Xu, David Wipf

Given a relational database (RDB) storing heterogeneous tabular information, how can we predict missing (or future) values in some target column of interest? As the space of potential targets is vast across enterprise settings, it is preferable to avoid learning a new model from scratch each time there is a new prediction task. Frozen foundation models based on RDB-specific encoders provide a viable solution, but ideal design remains an open question. On the one hand, it has recently been argued that certain parameter-free subgraph encoders combined with single-table foundation models can achieve near SOTA performance, with no RDB-specific pre-training required. Meanwhile, other contemporary studies advocate for parameterized encoders pre-trained to exploit observable labels for learning task-specific representations. To address this ambiguity, we analyze RDB encoder properties specifically when labels are present as inputs, proving limitations on the potential efficacy of trainable encoder parameters. As empirical validation, we demonstrate that considerably simpler parameter-free encoders are still capable of strong performance across many relevant benchmarking tasks.

View free PDFSource page

Related papers

arxivcs.LG2026-07-02

Zeus: Towards Tuning-Free Foundation Model for Time Series Analysis

Yisong Fu, Zezhi Shao, Chengqing Yu, Yujie Li, Yongjun Xu, Xueqi Cheng, et al.

We present Zeus, a unified tuning-free Time Series Foundation Model (TSFM) that delivers superior performance across diverse analysis tasks without any task-specific fine-tuning. Unlike prior studies that primarily focus on zero-shot forecasting but require task-specific tuning f…

View free PDFSource page
arxivcs.LG2026-07-23

Parameter-free Adaptive Sparse Attention via Compression-Based Content Selection

Debarshi Kundu, Swaroop Ghosh, Vasant Honavar

Data-adaptive sparse attention masks substantially outperform fixed patterns (e.g., BigBird and Longformer) and can even exceed dense attention on long sequences. Existing adaptive approaches---including SBM-Transformer, Dynamic Mask Attention, and NSA---typically require additio…

View free PDFSource page
arxivcs.CVcs.LGcs.MM2026-07-20

LFM: Leveraging Foundation Models for Source-Free Universal Domain Adaptation

Jing Li, Pan Liu, Meng Zhao, Wanli Xue, Yanhong Yang, Xu Cheng, et al.

Source-free universal domain adaptation (SF-UniDA) adapts a pre-trained source model to an unlabeled target domain under both covariate and label shifts, without access to source data. However, existing SF-UniDA methods rely on inefficient techniques such as threshold tuning and…

View free PDFSource page
arxivcs.LGcs.AI2026-07-18

Principled Direction-Free Intrinsic Motivation through Model-Free Epistemic Free-Energy Estimators

Alireza Furutanpey, Schahram Dustdar

Across environments with mixed sources of uncertainty, unsupervised reinforcement learning requires intrinsic motivation that does not precommit to a particular direction of surprise. Surprise minimization is scoped by design to ``unstable'' environments. Prediction-error curiosi…

View free PDFSource page
arxivcs.LGcs.AI2026-07-04

When Does Small Data Work? Accuracy and Efficiency Trade-offs Between Tabular Foundation Models and Conventional Methods for Crowd-State Classification at Hajj and Umrah

AlJawharh S. AlOtaibi, Mohamed Eltahir, Jude AlSubaie

Learning from few labeled examples is a central challenge in tabular machine learning, and it becomes the binding constraint in domains where labeling is costly, such as crowd monitoring during Hajj and Umrah. Tabular foundation models, which predict from only a handful of exampl…

View free PDFSource page