# tabular data Published articles for tabular data. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## Cutpurse: Decomposing Explanation-Guided Model Extraction on Tabular Data DevFeed: [Cutpurse: Decomposing Explanation-Guided Model Extraction on Tabular Data]() Original publisher: [Read original article]() Author: Seyma Selcan Magara, Hanru Zhang, Fatme El Zahraa Hamad, Nico Pfeifer, Mete Akg\\"un Published: 2026-11-09T05:00:00Z Content type: article Language: en Sources: [arXiv: Cryptography and Security](), [arXiv: Machine Learning]() Topics: [Machine learning](), [Cryptography](), [Structured-data]() Tags: [adversary](), [api](), [apis](), [attack](), [classifiers](), [cs-cr](), [cs-lg](), [lime](), [model](), [shap](), [tabular-data]() ### AI overview Across six datasets and five target families, feature-attribution rankings gave no systematic advantage for extracting tabular classifiers. Lime's discretization edges improved fidelity by up to 41 percentage points on continuous features, but they reflect quantiles of the explainer's background data and an attacker-side grid reproduced their gains without an explanation API. Cutpurse combines a self-computed grid with diverse query generation and boundary search; when bin edges were withheld, it improved fidelity by 16.3 points on continuous-feature targets and exceeded the undefended published attack by 7.0 points. These results apply to default Lime and Shap configurations under in-domain auxiliary-input, probability-output, and same-family surrogate assumptions; they do not establish the same outcome for other explainers, target classes, or data-free attackers. ### Source excerpt arXiv:2611.09375v1 Announce Type: new Abstract: Explanation APIs are often treated as increasing model-extraction risk, but an end-to-end attack comparison cannot show whether an explanation reveals model-specific information or merely supplies a convenient next-query coordinate. We study this distinction for explanation-guided extraction of tabular classifiers. We separate an explanation into an attribution channel, which ranks influential features, and a coordinate channel, which supplies candidate feature values for perturbing those features. We then test whether the useful bin edges are privileged to the service or reproducible from the adversary's own data. Across six datasets and five target families, using attribution rankings to choose which features to perturb provides no systematic extraction advantage. Lime's discretization bin edges can improve fidelity by up to 41 percentage points on continuous features, but those edges are quantiles of the explainer's background data rather than properties of the target model. In controlled comparisons, an attacker-side grid reproduces the useful bin-edge gains without querying an explanation API. We use these findings to derive Cutpurse, which combines a self-computed grid with diverse query generation and boundary search. Against a service that withholds bin edges, Cutpurse improves fidelity by 16.3 percentage points on continuous-feature targets. It also exceeds the undefended published attack by 7.0 percentage points. Thus, withholding default Lime bin edges alone does not prevent effective extraction under the in-domain auxiliary-input, probability-output, and same-family surrogate assumptions studied here. Our results concern default Lime and Shap configurations for tabular classifiers. They do not establish the same conclusion for other explainers, target classes, or data-free attackers. ## Exploring the Privacy--Size Tradeoff for Differentially Private Self-Attentive Variational Autoencoders DevFeed: [Exploring the Privacy--Size Tradeoff for Differentially Private Self-Attentive Variational Autoencoders]() Original publisher: [Read original article]() Author: Rohit Sadavarte, Dhruvi Kothari, Sparsh Gupta, Eric Van Oeveren Published: 2026-11-05T05:00:00Z Content type: article Language: en Sources: [arXiv: Cryptography and Security]() Topics: [Differential Privacy](), [VAE]() Tags: [attention](), [autoencoders](), [cs-ce](), [cs-cr](), [cs-lg](), [dataset](), [datasets](), [differential-privacy](), [privacy](), [size](), [synthetic](), [synthetic-data](), [tabular-data](), [utility](), [vae]() ### AI overview SAVANT is a tabular-data variational autoencoder combining self-attention with DP-SGD. The study compares it with PrivBayes and MST across privacy budgets and dataset sizes on two public datasets, measuring the downstream classifier utility lost when trained on synthetic instead of real data. Under the same differential privacy guarantee, more training data generally improves utility. The authors report that increased data size offsets much of the cost of stricter privacy for PrivBayes and MST; a noise-scaling argument suggests DP-SGD trades data for privacy budget less favorably. Two membership-inference attacks, assessed against a measured null, found no aggregate membership signal. ### Source excerpt arXiv:2611.05582v1 Announce Type: cross Abstract: Data tends to be most useful when it is shared, and the data most worth sharing is often the data that privacy obligations restrict. Synthetic data from a generator trained under differential privacy (DP) is one way to reconcile the two, but its utility depends on the privacy budget $\varepsilon$ and on how much data the generator sees - a parameter that most data sharing applications of synthetic data control. We present SAVANT, a self-attentive variational autoencoder for tabular data trained with DP-SGD that, to our knowledge, is the first to combine attention, a VAE, and DP in one tabular model. We evaluate it against two marginal mechanisms, PrivBayes and MST, jointly over the privacy budget and the dataset size on two public datasets, measuring the utility a downstream classifier loses by training on synthetic rather than real data. We find that, under the same DP guarantee, more data generally means more utility. For the marginal mechanisms it absorbs most of what a strict budget costs, and a noise-scaling argument suggests DP-SGD exchanges data fo budget at a worse rate. Two independent membership-inference attacks, judged against a measured null, found no aggregate membership signal. ## TabPFN-3: Technical Report DevFeed: [TabPFN-3: Technical Report]() Original publisher: [Read original article]() Author: L\\'eo Grinsztajn, Klemens Fl\\"oge, Oscar Key, Felix Birkel, Philipp Jund, Brendan Roof, Mihir Manium, Shi Bin Hoo, Magnus B\\"uhler, Anurag Garg, Dominik Safaric, Jake Robertson, Benjamin J\\"ager, Simo Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [AI Foundation Models](), [Benchmark](), [LLM optimization](), [AI Factory]() Tags: [api](), [beats](), [bench](), [cs-lg](), [first](), [foundation-model](), [foundation-models](), [gradient](), [inference-speed](), [integrations](), [internet](), [kv-cache](), [llms](), [scaling](), [shap](), [stat-ml](), [state](), [synthetic-data](), [tabular-data](), [time-series]() ### AI overview TabPFN-3 is a tabular foundation model pretrained exclusively on synthetic data. The report says it handles datasets up to 1 million training rows, runs up to 20 times faster than TabPFN-2.5, and, on TabArena, a single forward pass outperforms tuned and ensembled baselines while leading the speed-performance frontier. It reports gains on datasets with many classes and results above eight-hour-tuned gradient-boosted trees for datasets up to 1 million rows and 200 features. TabPFN-3-Plus (Thinking) adds test-time compute scaling; the API variant beats non-TabPFN models on TabArena and is reported 10 times faster than AutoGluon 1.5 extreme. Specialized results include TabPFN-TS-3 ranking second on fev-bench, relational-data performance on RelBenchV1, and faster SHAP computation. Reduced KV cache and row chunking enable inference on 1 million rows using one H100. These are abstract-level reported benchmark claims; the supplied text does not provide evaluation details. ### Source excerpt arXiv:2605.13986v3 Announce Type: replace Abstract: Tabular data underpins most high-value prediction problems in science and industry, and TabPFN has driven the foundation model revolution for this modality. Designed with feedback from our users, TabPFN-3 builds on this foundation to scale state-of-the-art performance to datasets with 1M training rows and substantially reduce training and inference time. Pretrained exclusively on synthetic data from our prior, TabPFN-3 dramatically pushes the frontier of tabular prediction and brings substantial gains on time series, relational, and tabular-text data. On the standard tabular benchmark TabArena, a forward pass of TabPFN-3 outperforms all other models, including tuned and ensembled baselines, by a significant margin, and pareto-dominates the speed/performance frontier. On more diverse datasets, TabPFN-3 ranks first on datasets with many classes, and beats 8-hour-tuned gradient-boosted-tree baselines on datasets up to 1M training rows and 200 features. TabPFN-3 introduces test-time compute scaling to tabular foundation models. Our API offering TabPFN-3-Plus (Thinking) exploits this to beat all non-TabPFN models by over 200 Elo on TabArena, rising to 420 Elo on the largest data subset, and outperforms AutoGluon 1.5 extreme while being 10x faster, without using LLMs, real data, internet search or any other model besides TabPFN. TabPFN-3 extends the capabilities of our models, enabling SOTA prediction on relational data (new SOTA foundation model on RelBenchV1) and tabular-text data (SOTA on TabSTAR via TabPFN-3-Plus); and improves existing integrations: a specialized checkpoint, TabPFN-TS-3, ranks 2nd on the time-series benchmark fev-bench, and SHAP-value computation is up to 120x faster. TabPFN-3 achieves this performance while being up to 20x faster than TabPFN-2.5. In addition, a reduced KV cache and row-chunking scale to 1M rows on one H100 with fast inference speed. ## Literati: Towards Anytime Optimal Shape Generalized Trees via AO\* DevFeed: [Literati: Towards Anytime Optimal Shape Generalized Trees via AO\*]() Original publisher: [Read original article]() Author: Nakul Upadhya, Eldan Cohen Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Machine learning](), [Combinatorial optimization]() Tags: [algorithm](), [cs-lg](), [decision-trees](), [graph](), [greedy](), [guarantees](), [induction](), [interpretability](), [machine-learning](), [optimization](), [performance](), [real-world](), [round-robin](), [tabular-data](), [test](), [training](), [trees]() ### AI overview Literati is presented as the first algorithm for optimal induction of Shape Generalized Trees (SGTs), which use learnable univariate shape functions instead of axis-aligned threshold splits. It formulates joint optimization of tree structure and shape-function complexity as an AND/OR graph, solved with AO*. A secondary heuristic for OR-node selection and round-robin exploration of AND-nodes improve anytime performance while preserving optimality. On 24 real-world datasets, the authors report higher training and test accuracy than state-of-the-art tree approaches. ### Source excerpt arXiv:2609.09299v2 Announce Type: replace Abstract: Decision trees are prized for their interpretability and strong performance on tabular data, but popular greedy top-down induction algorithms can yield suboptimal and unnecessarily complex structures. Optimal decision tree methods address this through global optimization, yet remain restricted to axis-aligned threshold splits, which limit the expressivity of each node and often force deep, complex trees to capture non-linear feature effects. Shape Generalized Trees (SGTs) generalize threshold splits to learnable univariate shape functions, improving expressivity and enabling more compact trees. However, existing SGT induction algorithms are greedy and offer no optimality guarantees. In this work, we introduce Literati, the first algorithm for optimal SGT induction. We propose a novel AND/OR graph formulation of the problem that jointly optimizes tree structure and shape function complexity. To solve this AND/OR graph, we develop an AO*-based algorithm with two enhancements that improve anytime performance while preserving optimality: a secondary heuristic for OR-node selection and a round-robin policy for AND-node exploration. Across 24 real-world datasets, Literati achieves higher training and test accuracy than state-of-the-art tree approaches. ## KINDER: Kernel-based Independence for Fair Representation Learning via Prototype-space Erasure DevFeed: [KINDER: Kernel-based Independence for Fair Representation Learning via Prototype-space Erasure]() Original publisher: [Read original article]() Author: Abtin Mogharabin, Alp Toykan Kaplan, Jiaee Cheong, Sinan Kalkan Published: 2026-11-03T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Representation learning](), [Machine learning]() Tags: [classification](), [code](), [cs-lg](), [data](), [decoder](), [deep](), [experiments](), [fairness](), [family](), [feature](), [image](), [kernel](), [machine-learning](), [model](), [optimization](), [prototype](), [random](), [regression](), [stat-ml](), [tabular-data]() ### AI overview KINDER aims to remove sensitive-attribute information from intermediate representations by estimating a sensitive subspace from protected-group prototypes in a random Fourier feature space, then projecting representations onto its orthogonal complement. Unlike approaches that define debiasing through a chosen adversary, probe, or fairness penalty, it targets a selected family of prototype functionals. The abstract says it can be used with deep models across architectures, tasks, and training paradigms. Experiments span supervised and self-supervised learning; classification, regression, and image segmentation; and visual, textual, and tabular data. The excerpt does not provide quantitative results or details about limitations. ### Source excerpt arXiv:2611.00264v1 Announce Type: new Abstract: Concept erasure is a prominent approach to achieving fairness in machine learning through the removal of sensitive attributes from learned representations. Prior concept erasure methods typically define debiasing indirectly through the failure of a chosen adversary, probe, or fairness penalty, which makes the target of erasure dependent on a particular decoder family or optimization setup. We introduce KINDER, which directly removes sensitive-attribute information from intermediate representations with respect to a chosen family of prototype functionals, and is compatible with any deep model containing an intermediate feature space, independent of the architecture, task type, or training paradigm. KINDER operates in a random Fourier feature space, where it estimates a sensitive subspace from protected group prototypes and projects representations onto its orthogonal complement. Thus, it forms an explicit representation cleaning mechanism distinct from prior fairness control studies. We conduct extensive experiments across diverse settings and show the effectiveness of KINDER on (i) supervised and self-supervised settings, (ii) classification, regression, and image segmentation tasks, and (iii) diverse data modalities, including visual, textual, and tabular data. Code and trained models are available at: https://github.com/abtinmU/KINDER ## Kolmogorov-Arnold Networks vs. Multi-Layer Perceptrons: A Rigorous Statistical Benchmark on Clinical Tabular Data DevFeed: [Kolmogorov-Arnold Networks vs. Multi-Layer Perceptrons: A Rigorous Statistical Benchmark on Clinical Tabular Data]() Original publisher: [Read original article]() Author: Rafa{\\l} Wodzi\\'nski, Pawe{\\l} Szczerbiak Published: 2026-11-03T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence](), [arXiv: Machine Learning]() Topics: [Predictive AI](), [Benchmark](), [Hyperparameter optimization](), [Structured-data]() Tags: [architectures](), [bayesian](), [benchmark](), [comparison](), [cross-validation](), [cs-ai](), [cs-lg](), [efficiency](), [hyperparameter-optimization](), [kan](), [latency](), [medical](), [metrics](), [mlp](), [performance](), [precision](), [random](), [tabular-data]() ### AI overview Across seven clinical tabular datasets, nine KAN variants showed no systematic pooled discrimination advantage over a tuned StandardMLP, though TabKAN and ReLUKAN achieved higher AUPRC on severely imbalanced datasets and could improve minority-class detection. The benchmark used repeated stratified cross-validation with nested Bayesian tuning, assessed AUPRC, Brier Score, and efficiency, and applied Bayesian equivalence testing. KANs had 1.20-1.48x longer per-sample inference latency and up to 12.5x more trainable parameters; StandardMLP remained an efficiency-focused baseline, while KANs may suit tasks prioritizing precision on rare positives. ### Source excerpt arXiv:2611.02856v1 Announce Type: new Abstract: Kolmogorov-Arnold Networks (KANs) have recently been proposed as a promising alternative to Multi-Layer Perceptrons (MLPs). However, empirical evaluations on structured medical datasets remain scarce, and commonly reported metrics like AUROC can yield overly optimistic performance estimates on imbalanced clinical cohorts. In this study, we present a standardized benchmark comparing eleven architectures - nine KAN variants (FastKAN, WavKAN, ReLUKAN, ChebyKAN, TaylorKAN, GramKAN, JacobiKAN, LegendreKAN, and TabKAN), a tuned StandardMLP, and a Random Forest baseline - across seven heterogeneous medical tabular datasets spanning oncology, cardiology, nephrology, neurology, endocrinology, and obstetrics. Our experimental protocol employs a 3x5 Repeated Stratified K-Fold cross-validation scheme with nested Bayesian hyperparameter optimization and evaluates discrimination via Area Under the Precision-Recall Curve (AUPRC), calibration through the Brier Score, and computational efficiency. Formal equivalence testing is conducted using the Bayesian Correlated t-Test with a Region of Practical Equivalence (ROPE) at +/- 0.01. The results indicate that on datasets with severe class imbalance, dedicated KAN architectures (particularly TabKAN and ReLUKAN) can improve minority-class detection and achieve higher AUPRC than standard MLPs. Nevertheless, across pooled datasets, KAN variants demonstrate comparable overall predictive discrimination rather than a systematic advantage, while requiring 1.20x to 1.48x longer inference latency per sample and up to 12.5x more trainable parameters. Consequently, StandardMLP remains a strong baseline on the efficiency Pareto frontier, while KAN variants provide a viable alternative when precision on rare positive cases is the primary clinical objective. The code for reproducing the results is available at https://github.com/rafalwodzinski/kan_mlp_comparison. ## Beyond IID: How General Are Tabular Foundation Models, Really? DevFeed: [Beyond IID: How General Are Tabular Foundation Models, Really?]() Original publisher: [Read original article]() Author: Lennart Purucker, Andrej Tschalzev, Nick Erickson, Gioia Blayer, David Holzm\\"uller, Alan Arazi, Alexander Pfefferle, Mustafa Tajjar, Ga\\"el Varoquaux, Frank Hutter Published: 2026-11-02T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence](), [arXiv: Machine Learning]() Topics: [AI Foundation Models](), [Model Development](), [Predictive AI](), [Open Source Models & Datasets](), [Tool](), [Structured-data]() Tags: [benchmark](), [cs-ai](), [cs-lg](), [datasets](), [evaluation](), [foundation-models](), [framework](), [open-source](), [tabular-data]() ### AI overview BeyondArena evaluates tabular models across IID, temporal, and grouped tasks, varied dataset sizes and feature dimensions, and data with text or high-cardinality features. The authors introduce DataFoundry, a Python framework and metadata schema for curating datasets, to address fragmented benchmark software and protocols. In tests of 11 models on 142 datasets, open-source tabular foundation models were competitive or excelled on datasets up to 100,000 rows; traditional tree-based and deep learning models remained dominant on grouped or large data. The abstract presents this as evidence that standard IID benchmarks miss harder settings where current foundation models are less successful. ### Source excerpt arXiv:2606.30410v2 Announce Type: replace Abstract: Foundation models for predictive machine learning on tabular data have recently gained significant traction in academia and industry. Research communities across disciplines are increasingly evaluating tabular foundation models on diverse datasets and tasks. However, these task- and discipline-specific evaluations remain largely inaccessible to model researchers because benchmark software and evaluation protocols are fragmented. As a result, model researchers rely on standard benchmarks, which are mostly defined for tasks where tabular foundation models already excel. The most challenging scenarios are excluded, limiting meaningful progress in the field by focusing on marginal improvements on IID data rather than on broader, more demanding challenges. To overcome this, we introduce BeyondArena, the first unified holistic benchmark for tabular data that supports diverse task types (IID, temporal, grouped), across sample size and feature dimensionality scales, with diverse feature types (with text, with high cardinality) from a broad range of disciplines. To enable unified benchmarking beyond standard benchmarks, we introduce DataFoundry, a Python framework and metadata schema for curating tabular datasets for predictive machine learning. Our results across 11 models and 142 curated datasets show that current open-source tabular foundation models are competitive or excel on datasets up to 100k rows, while traditional tree-based and deep learning models still dominate on grouped or large data. BeyondArena guides model research for the most demanding challenges in tabular data, enabling progress towards truly foundational tabular models. ## Towards Universal Tabular Embeddings: A Benchmark Across Data Tasks DevFeed: [Towards Universal Tabular Embeddings: A Benchmark Across Data Tasks]() Original publisher: [Read original article]() Author: Liane Vogel, Kavitha Srinivas, Niharika D'Souza, Sola Shirai, Oktie Hassanzadeh, Horst Samulowitz Published: 2026-11-02T05:00:00Z Content type: article Language: en Sources: [arXiv: Databases](), [arXiv: Machine Learning]() Topics: [Embeddings](), [Representation learning](), [AI Foundation Models](), [Machine learning](), [Benchmark]() Tags: [benchmark](), [column](), [comparison](), [cs-db](), [cs-lg](), [domains](), [embeddings](), [foundation](), [foundation-models](), [guidance](), [practical](), [prediction](), [real-world](), [retrieval](), [search](), [semantic](), [semantic-search](), [table](), [tabular-data](), [tasks](), [test]() ### AI overview TEmBed (Tabular Embedding Test Bed) provides a unified benchmark for comparing tabular representation learning models across cell, row, column, and table levels. It addresses the difficulty of comparing methods evaluated in separate task-specific settings. The reported results find that the best model depends on both the task and representation level, so there is no single approach identified as best across them. The abstract says the benchmark offers guidance for choosing embeddings in applications such as table retrieval, semantic search, and table-based prediction, and supports work toward more general-purpose tabular representations. The supplied text contains only the abstract, not detailed results or model-by-model comparisons. ### Source excerpt arXiv:2604.21696v3 Announce Type: replace-cross Abstract: Tabular foundation models aim to learn universal representations of tabular data that transfer across tasks and domains, enabling applications such as table retrieval, semantic search and table-based prediction. Despite the growing number of such models, it remains unclear which approach works best in practice, as existing methods are often evaluated under task-specific settings that make direct comparison difficult. To address this, we introduce TEmBed, the Tabular Embedding Test Bed, a unified benchmark for systematically evaluating tabular embeddings across four representation levels: cell, row, column, and table. Evaluating a diverse set of tabular representation learning models, we show that which model to use depends on the task and representation level. Our results offer practical guidance for selecting tabular embeddings in real-world applications and lay the groundwork for developing more general-purpose tabular representation models. ## Interpretable neural equation learning for high-dimensional materials data DevFeed: [Interpretable neural equation learning for high-dimensional materials data]() Original publisher: [Read original article]() Author: Philip Jacobson, Prasad Iyer, Remi Dingreville, Saaketh Desai Published: 2026-10-30T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Machine learning](), [Neural Network]() Tags: [activation](), [benchmarks](), [cond-mat-mtrl-sci](), [cs-lg](), [datasets](), [experimental](), [framework](), [learning](), [machine-learning](), [materials](), [neural](), [simulations](), [tabular-data]() ### AI overview nn-EQL learns compact candidate equations for sparse, high-dimensional materials data through iterative neural-network training, pruning, and retraining with customizable neuron-level activation functions. It ranks candidates with a minimum description length-inspired score balancing error and sparsity, and diagnoses term contributions, input-scaled sensitivities, and uncertainties. The authors demonstrate it on AI-Feynman benchmarks, finite-element simulations of heterogeneous materials, and high-throughput thin-film deposition data; standard encoder models extend the approach to high-dimensional inputs. The abstract suggests future additions such as physically interpretable representations, multimodal and multi-fidelity data, and adaptive data acquisition, but presents these as directions rather than demonstrated capabilities. ### Source excerpt arXiv:2610.37880v1 Announce Type: new Abstract: Equation learning seeks to discover compact, interpretable equations from data using various machine-learning techniques. However, few workflows discover and interpret learned equations in sparse, high-dimensional materials datasets. We present nn-EQL, an equation-learning framework that generates candidate equations through iterative training, pruning, and retraining of a neural network with customizable, user-specified neuron-level activation functions. Candidate equations are ranked using a minimum description length (MDL)-inspired score, balancing error and sparsity, and diagnosed to assess term-wise contributions, input-scaled sensitivities, and uncertainties. We demonstrate nn-EQL on standard AI-Feynman benchmarks as well as multiscale high-dimensional materials datasets such as finite-element simulations of heterogeneous materials, and high-throughput experimental thin-film deposition data. nn-EQL integrates with tabular data and high-dimensional data via standard encoder models, learning equations and providing insight into learned terms. Looking forward, integrating physically interpretable representations, multimodal and multi-fidelity data, and adaptive data acquisition could enable equation learning to formulate and test mechanistic hypotheses. ## Octopus: Agentic Multi-Table Discovery over Tabular Data Lakes DevFeed: [Octopus: Agentic Multi-Table Discovery over Tabular Data Lakes]() Original publisher: [Read original article]() Author: Wen-Zhi Li, Sainyam Galhotra Published: 2026-10-30T04:00:00Z Content type: article Language: en Sources: [arXiv: Databases]() Topics: [data-engineering](), [Structured-data](), [Agentic AI](), [Large Language Model](), [AI-driven information retrieval](), [schema-evolution](), [hybrid-search]() Tags: [adaptive](), [agent](), [agentic](), [analytics](), [api](), [benchmark](), [claude](), [claude-code](), [code](), [column](), [cs-db](), [data-discovery](), [data-lakes](), [datasets](), [discovery](), [execution](), [experimental](), [framework](), [octopus](), [table](), [tabular-data]() ### AI overview Octopus is a lightweight agentic framework for finding relevant tables when a natural-language question may require information from multiple tables. It combines lightweight offline semantic guidance with query-time access to raw tabular data; an LLM agent orchestrates semantic column search, structured tabular operations, and general-purpose execution, using intermediate observations to locate and verify tables. On a new benchmark spanning four independent-discovery and two join-based datasets, Octopus achieved the highest macro-F1 on all six. The abstract reports lower offline preprocessing overhead than Pneuma and lower online latency and API cost than Claude Code. With the same NL2SQL agent, it also produced substantially higher downstream question-answering F1 than Claude Code. No scores or evaluation details are provided in the supplied excerpt. ### Source excerpt arXiv:2601.02304v2 Announce Type: replace Abstract: Tabular data discovery aims to identify the tables relevant to a natural-language question from a large collection of heterogeneous data. Existing systems typically assume that a question is answered by a single table and rely on retrieval pipelines constructed through substantial offline preprocessing. These assumptions become restrictive when relevant information is distributed across multiple tables. In this paper, we present Octopus, a lightweight agentic framework for multi-table data discovery. Octopus follows a thin-harness design that combines lightweight offline semantic guidance with adaptive access to raw tabular data. At query time, an LLM agent dynamically orchestrates semantic column search, structured tabular operations, and general-purpose execution to locate and verify relevant tables based on intermediate observations. We further introduce a benchmark that explicitly evaluates multi-table data discovery, covering both settings where multiple tables independently answer a question and settings where multiple tables must be joined. Experimental results across four independent-discovery datasets and two join-based datasets show that Octopus achieves the highest macro-F1 on all six datasets. It also substantially reduces offline preprocessing overhead relative to Pneuma and online latency and API cost relative to Claude Code. When coupled with the same NL2SQL agent, Octopus yields substantially higher downstream question-answering F1 than Claude Code, showing that improved table discovery provides more complete evidence for downstream analytics. Code is available at https://github.com/wenzhilics/octopus. ## TACO-Graph: Automated Construction of Text-Atributed Graphs DevFeed: [TACO-Graph: Automated Construction of Text-Atributed Graphs]() Original publisher: [Read original article]() Author: Amiri Hayes, Kristina Wicke Published: 2026-10-28T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Graph Mining](), [Graphs](), [Artificial Intelligence](), [Representation learning](), [Dataset development]() Tags: [automated](), [cs-lg](), [datasets](), [decision-trees](), [evaluation](), [gnn](), [graph](), [graphrag](), [machine-learning](), [open-source](), [tabular-data](), [text]() ### AI overview TACO-Graph helps select how to turn multi-attribute tabular data into text-attributed graphs (TAGs) for a target task. It enumerates valid graph constructions, scores them with fast proxy evaluations on small representative subsets, and trains an interpretable decision tree to predict which choices perform better. Across five datasets covering citation networks, product reviews, and bibliographic relationships, the trees predicted construction quality on separate evaluation samples with mean Spearman ρ above 0.9 for node classification and above 0.8 for GraphRAG retrieval. The abstract reports results for both GNN and GraphRAG tasks and says the pipeline is available as an open-source library for evaluation on new datasets. ### Source excerpt arXiv:2610.36650v1 Announce Type: new Abstract: Graph-structured representations of text data have become foundational across machine learning, yet it remains unclear how to systematically construct a graph from a given dataset for a target task. Choices about which entities become nodes, which relationships become edges, and what text each node contains can substantially affect downstream performance. In practice, finding optimal constructions often requires expert knowledge or manual trial and error. We introduce TACO-Graph, a decision tree framework for learning which graph construction choices are most effective for a given dataset and task. Specifically, we focus on text-attributed graphs (TAGs) constructed from multi-attribute tabular data. TACO-Graph enumerates valid TAG representations, scores each using fast proxy evaluation on small representative subsets, and trains an interpretable decision tree on those scores. Across five datasets spanning citation networks, product reviews, and bibliographic relationships, our decision trees predict construction quality on separately drawn evaluation samples with mean Spearman $\rho$ above $0.9$ for node classification and above $0.8$ for GraphRAG retrieval. Our trees systematically identify which constructions perform better across both GNN and GraphRAG tasks, and the full pipeline is released as an open-source library to support evaluation on new tasks and datasets at https://github.com/AmiriHayes/text_attributed_graphs. TACO-Graph provides a systematic and interpretable guide for determining how to convert tabular data to text-attributed graphs for downstream tasks. ## Interpretable Synthetic Medical Tabular Data Generation for Clinical Decision Support Using Fuzzy Cognitive Maps DevFeed: [Interpretable Synthetic Medical Tabular Data Generation for Clinical Decision Support Using Fuzzy Cognitive Maps]() Original publisher: [Read original article]() Author: Michael Vasilakakis (Department of Computer Science and Biomedical Informatics, University of Thessaly, Lamia, Greece), Dimitris K. Iakovidis (Department of Computer Science and Biomedical Informatics Published: 2026-10-28T04:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence](), [arXiv: Machine Learning]() Topics: [Machine learning](), [Structured-data](), [Artificial Intelligence]() Tags: [activation](), [causality](), [clinical-decision-support](), [coherence](), [cpu](), [cs-ai](), [cs-lg](), [data](), [data-types](), [datasets](), [deep](), [dependencies](), [disease](), [domain](), [edge](), [evaluation](), [gaussian](), [generation](), [maps](), [medical](), [synthetic](), [tabular-data]() ### AI overview The paper proposes generating synthetic medical records with Fuzzy Cognitive Maps (FCMs): clinical features are represented as interpretable fuzzy sets, dependencies are encoded through fuzzy-set intersections, and records are produced by propagating random linguistic activations to convergence before defuzzification. The method handles mixed data types and health-record constraints. On UCI benchmarks under Train-on-Synthetic-Test-on-Real evaluation, it reports Heart Disease accuracy up to 0.81 and AUROC up to 0.90, matching or exceeding TVAE and Gaussian Copula while running on CPU. Reported fidelity reaches 0.91 KS Complement and 0.95 Correlation Similarity; DCR Baseline Protection scores are reported as consistently higher than TVAE. These are abstract-level claims, not evidence of performance beyond the evaluated benchmarks. ### Source excerpt arXiv:2610.00391v2 Announce Type: replace-cross Abstract: Synthetic medical tabular data generation has become essential for developing and validating computer-based medical systems (CBMSs) when real clinical data is restricted due to privacy, ethical, or data availability limitations. Existing probabilistic and deep generative models often lack interpretability and fail to preserve clinically meaningful dependencies, limiting their suitability for safety-critical applications. This paper proposes a novel application of Fuzzy Cognitive Maps (FCMs) in a framework for synthetic medical tabular data generation with explicit causality and privacy preservation. Clinical features are described using linguistically interpretable fuzzy sets, and inter-feature dependencies are encoded as FCM edge weights computed from fuzzy set intersections. Synthetic patient records are generated by propagating randomly initialized linguistic activation vectors through the FCM until convergence, followed by defuzzification to produce clinically coherent numerical values. The approach natively handles mixed data types, and domain constraints common in health records. Experimental evaluation on UCI medical benchmark datasets demonstrates competitive performance under a Train-on-Synthetic-Test-on-Real (TSTR) protocol. The proposed method achieves accuracy of up to 0.81 and AUROC of up to 0.90 on the Heart Disease dataset, matching or exceeding TVAE and Gaussian Copula baselines while running exclusively on CPU. Fidelity metrics including KS Complement (up to 0.91) and Correlation Similarity (up to 0.95) confirm strong statistical coherence, and DCR Baseline Protection scores consistently exceed those of TVAE, confirming adequate privacy guarantees. These results demonstrate that causally grounded, interpretable fuzzy modeling offers a computationally efficient and transparent alternative to deep generative models for trustworthy synthetic data generation in CBMSs. ## SynthGuard-ReleaseBench: Locked-Audit Evidence for Synthetic Tabular Data Releases DevFeed: [SynthGuard-ReleaseBench: Locked-Audit Evidence for Synthetic Tabular Data Releases]() Original publisher: [Read original article]() Author: Jeffery Opoku, David Banahene Published: 2026-10-28T04:00:00Z Content type: article Language: en Sources: [arXiv: Cryptography and Security]() Topics: [audit](), [Synthetic Data Generation](), [Differential Privacy](), [Structured-data]() Tags: [audit](), [benchmark](), [control](), [cs-cr](), [framework](), [generation](), [generator](), [health](), [lower-bounds](), [mechanism](), [model](), [models](), [panel](), [privacy](), [prototype](), [queries](), [records](), [release](), [releases](), [stat-me](), [synthetic](), [tabular-data]() ### AI overview SynthGuard-ReleaseBench evaluates whether synthetic tabular data releases are supported for a specified use, population, and threat model. It locks the candidate panel, tolerances, and audit schedule before evaluation, compares real-trained and synthetic-trained workflows on protected data, and reports simultaneous finite-sample bounds for bounded loss gaps. Controls help distinguish favorable, unfavorable, and excluded audit outcomes; utility, empirical privacy risk, mechanism claims, and human release authority remain separate. Across the reported studies, transparent baselines pass some audits, while compact learned models fail at declared budgets; a health-table case is excluded after its negative control passes. Repeated post-audit scaling admits the learned models at larger fit sizes while rejecting a dependence-destroying control, but withdraws a finer single-seed ordering. The framework also presents sample-size rules and certificates, plus lower bounds on audit reconstruction and panel-size correction. It provides use-specific release evidence, not a claim that generators are private, safe, or deployment-ready. ### Source excerpt arXiv:2608.14753v2 Announce Type: replace Abstract: Synthetic tabular data are often judged by realism, privacy, or downstream-task scores. Those scores do not answer whether a proposed release is supported for a named use, population, and threat model. We introduce SynthGuard-ReleaseBench, an audit framework that locks the use, candidate panel, tolerances, and audit schedule before evaluation. It compares real-trained and synthetic-trained workflows on protected data, gives simultaneous finite-sample bounds for bounded loss gaps, requires controls, and keeps utility, empirical privacy risk, mechanism claims, and human release authority separate. Across four American Community Survey studies, five non-ACS records, two chronological diagnostics, and a sealed prototype, the benchmark retains favorable, unfavorable, and excluded outcomes. Transparent baselines pass some locked audits; compact learned models fail under the declared budgets; a health-table case is excluded because its negative control passes. A post-audit scaling arm, repeated across three generation seeds, shows the same locked criterion admitting those learned models once they are fit on enough data while still rejecting a dependence-destroying control at every size, so the criterion discriminates rather than merely rejects; the same repetition withdraws a finer single-seed ordering. The theory adds a pre-audit sample-size rule, variance-adaptive and anytime-valid certificates that tighten the bound two to ten times on the same locked evidence, a temporal certificate for time-ordered audits, and two lower bounds: ordinary bounded queries reconstruct a protected audit once the query budget reaches its size, and the panel-size correction is necessary rather than conservative. The contribution is a reproducible workflow for use-specific release evidence, not a claim that any generator is private, safe, or deployment-ready. ## PRISMA: Differentially Private Synthetic In-Car Telemetry via Differentiable Simulation DevFeed: [PRISMA: Differentially Private Synthetic In-Car Telemetry via Differentiable Simulation]() Original publisher: [Read original article]() Author: Kilian Zeiseweis, Nele Sophie Br\\"ugge, Marven Kummerfeldt, Ali Raza, Esfandiar Mohammadi Published: 2026-10-27T04:00:00Z Content type: article Language: en Sources: [arXiv: Cryptography and Security]() Topics: [Cryptography](), [Representation learning](), [telemetry](), [Decoding](), [clustering]() Tags: [algorithm](), [car](), [cs-cr](), [distribution](), [diversity](), [driving](), [energy](), [intrinsic-dimensionality](), [model](), [physics](), [power-consumption](), [prisma](), [private](), [profile](), [simulation](), [synthesis](), [synthetic](), [tabular-data](), [telemetry](), [trace]() ### AI overview PRISMA synthesizes private in-car energy consumption traces by encoding them into a low-dimensional latent space, applying the differentially private tabular synthesis algorithm AIM to that representation, then decoding samples with a differentiable physics-based simulation using public information. The authors say this final step uses no additional privacy budget. They report that at ε = 0.5, PRISMA outperforms prior tabular and marginal-based DP synthesis baselines in their experiments, preserving diversity across power consumption profiles rather than collapsing to a few modes. ### Source excerpt arXiv:2610.35521v1 Announce Type: new Abstract: Differentially private synthesis of in-car energy consumption data is challenging because the output is high-dimensional, and known lower bounds show that the privacy-utility trade-off degrades with the square root of the number of output dimensions. However, we observe that the intrinsic dimensionality of in-car energy consumption is much lower than that of the raw trace: given a small set of contextual personal variables-such as start point, end point, the car model, and the driving behavior-the consumption profile can be largely explained by publicly available map data (including elevation) combined with a physical model of vehicle energy use. We exploit this low intrinsic dimensionality in PRISMA, a Privacy-preserving In-car Synthesis and Modelling Algorithm. PRISMA involves three stages: (i) energy consumption traces are encoded into a low-dimensional latent space; (ii) the distribution of latent vectors is synthesized using the differentially private tabular data synthesis algorithm AIM, and (iii) samples drawn from this distribution are decoded back into full energy consumption traces via a differentiable physics-based simulation, a step that consumes no additional privacy budget since it uses only public information. By using a compact, physically-grounded latent representation of the traces, PRISMA significantly reduces the effective output dimension. Our experimental results show that PRISMA achieves for a competitive $\varepsilon = 0.5$ a better data synthesis performance than prior tabular and marginal-based DP synthesis baselines by preserving the diversity of the original power consumption profiles without collapsing to a small set of modes. ## Bridging Search Intent and Tabular Data in Dataset Search DevFeed: [Bridging Search Intent and Tabular Data in Dataset Search]() Original publisher: [Read original article]() Author: Wenjing Lin, Madelon Hulsebos Published: 2026-10-27T04:00:00Z Content type: article Language: en Sources: [arXiv: Databases]() Topics: [AI-driven information retrieval](), [Embeddings](), [text2sql]() Tags: [abstract](), [also](), [analysis](), [arxiv](), [based](), [bridging](), [capture](), [combined](), [components](), [content](), [cs-db](), [cs-ir](), [customizable](), [data](), [data-analysis](), [dataset](), [embedding](), [embeddings](), [example](), [existing](), [search](), [tabular-data]() ### AI overview The paper presents two natural-language components for dataset search. Hypothetical Schema Embeddings (HySE) uses a language model to create task-aligned hypothetical table schemas for embedding-based retrieval. On KaggleDS, a corpus of 6,503 tables and task queries, it improves recall over keyword and embedding baselines and a dense retriever fine-tuned on that corpus. On NQ-Tables, it raises Recall@10 by 6.0 percentage points over dense retrieval baselines. FlexPrune converts natural-language dataset constraints into SQL predicates over a customizable metadata index; combined with HySE, it improves Recall@10 by 6.0 points over dense retrieval followed by tag-based filtering. ### Source excerpt arXiv:2610.33936v1 Announce Type: cross Abstract: Dataset search remains a bottleneck in data analysis workflows, as existing systems largely rely on keyword matching or shallow retrieval that fail to capture user intent. To facilitate intent-driven search, we present two retrieval components to equip multi-faceted dataset search interfaces through natural language. First, we propose Hypothetical Schema Embeddings (HySE) that uses a language model to generate task-aligned hypothetical table schemas which are used for embedding-based retrieval, bridging the gap between intent and table content. We introduce KaggleDS, a corpus comprising 6,503 tables and diverse task-queries derived from Kaggle, and show that HySE consistently improves recall over strong keyword and embedding-based baselines, as well as over a dense table retriever fine-tuned on KaggleDS. We also show that HySE generalizes to end-to-end question answering on, for example, the NQ-Tables dataset, a corpus of 169,898 Wikipedia tables with human-written queries, where HySE improves Recall@10 by +6.0 pp over dense retrieval baselines. Second, we introduce FlexPrune, which translates dataset constraints expressed in natural-language into SQL predicates over a customizable metadata index, enabling filtering across arbitrary metadata attributes. Combined with HySE, FlexPrune increases Recall@10 by +6.0 pp over dense retrieval followed by tag-based filtering. Our results show that natural-language-based retrieval supporting task and metadata queries offers a viable alternative to schema-level keyword search in dataset search. ## Why Does Counterintuitive Likelihood Behavior Rarely Emerge in Tabular Anomaly Detection? DevFeed: [Why Does Counterintuitive Likelihood Behavior Rarely Emerge in Tabular Anomaly Detection?]() Original publisher: [Read original article]() Author: Donghwan Kim, Junghun Phee, Hyunsoo Yoon Published: 2026-10-26T04:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence](), [arXiv: Machine Learning]() Topics: [Structured-data](), [AIOps for Kubernetes]() Tags: [anomaly-detection](), [behavior](), [correlation](), [cs-ai](), [cs-lg](), [cv](), [data](), [datasets](), [embedding](), [experiments](), [feature](), [flow](), [general](), [generative](), [generative-models](), [image](), [models](), [nlp](), [performance](), [practical](), [tabular-data](), [theoretical]() ### AI overview The abstract reports that normalizing-flow likelihood scoring rarely assigns anomalously high likelihoods to anomalies in tabular data, unlike the frequently observed pattern in images. The authors define a domain-agnostic formulation for detecting and evaluating this behavior, then test 47 tabular datasets against 13 baselines and 10 CV/NLP embedding datasets from ADBench. They find the behavior consistently rare in general tabular data and report competitive anomaly-detection performance from normalizing-flow likelihood. Their theoretical and empirical analyses attribute the difference from images to dimensionality and feature correlation. ### Source excerpt arXiv:2602.09593v2 Announce Type: replace Abstract: Deep generative models with tractable likelihoods, exemplified by normalizing flows, offer an effective basis for anomaly detection through likelihood-based scoring. We demonstrate that, unlike in the image domain where deep generative models frequently assign higher likelihoods to anomalous data, such counterintuitive behavior occurs rarely in tabular settings. We first introduce a domain-agnostic formulation that enables consistent detection and evaluation of the counterintuitive phenomenon, addressing the lack of a precise definition in prior work. Through extensive experiments on 47 tabular datasets benchmarked against 13 baseline models and on 10 CV/NLP embedding datasets from ADBench, we show that the phenomenon is consistently rare in general tabular data and that normalizing flow likelihood achieves competitive anomaly detection performance. We further provide theoretical and empirical analyses attributing this disparity to differences in dimensionality and feature correlation between tabular and image data. Therefore, our results indicate that likelihood-based detection with normalizing flows is a practical approach for tabular anomaly detection. ## Tabular Foundation Models for Credit Risk Prediction: A Dedicated Review and Empirical Comparison DevFeed: [Tabular Foundation Models for Credit Risk Prediction: A Dedicated Review and Empirical Comparison]() Original publisher: [Read original article]() Author: Bart Baesens, Andreas Goethals, Stefan Lessmann, Simon De Vos, Cristi\\'an Bravo, David Martens, Victor Medina-Olivares, Christophe Mues, Maria Oskarsd\\'ottir, Seppe vanden Broucke, Tony Van Gestel, Ti Published: 2026-10-26T04:00:00Z Content type: comparison Language: en Sources: [arXiv: Machine Learning]() Topics: [AI Foundation Models](), [Machine learning](), [Finance](), [Hyperparameter optimization]() Tags: [benchmark](), [boosting](), [classification](), [comparison](), [cs-lg](), [financial](), [foundation-models](), [hyperparameter-optimization](), [machine](), [machine-learning](), [modeling](), [performance](), [portfolios](), [prediction](), [predictive-modeling](), [pretraining](), [review](), [risk](), [small](), [tabular-data](), [tasks]() ### AI overview The benchmark reports that TabPFN-3 and TabICLv2 achieve the strongest average results for credit risk prediction among the evaluated methods, with the largest gains in small-sample and strongly imbalanced settings. It evaluates 33 classification methods on 14 Probability of Default datasets and 27 regression methods on 7 Loss Given Default datasets. The foundation models use fixed pretrained parameters, without downstream hyperparameter optimization or task-specific updates. The supplied text is the abstract, so it does not provide details about individual datasets, metrics, or experimental conditions. ### Source excerpt arXiv:2605.18147v3 Announce Type: replace Abstract: Predictive models are central to credit risk management, where the accuracy of default-probability and loss estimates affects the profitability of lending and the stability of the financial system. Decades of benchmarking research have consolidated the state of the art, with gradient-boosting models often representing the strongest machine-learning benchmark, while linear and logistic models remain important regulatory and industry reference models. The recent advent of tabular data foundation models introduces a new paradigm for predictive modeling. Rather than fitting a model to a target dataset, they use in-context learning to integrate information from the target dataset with prior knowledge acquired through large-scale pretraining on diverse collections of (synthetic) tabular data. We conjecture that access to information beyond the target data is beneficial in small-data settings and may help address longstanding challenges, including low-default portfolios. Whether this promise holds in practice remains an open question. This paper benchmarks state-of-the-art tabular foundation models against a broad set of classical and advanced machine learning competitors on two core tasks: Probability of Default (PD) and Loss Given Default (LGD) modeling. We evaluate $33$ classification methods across $14$ PD datasets and $27$ regression methods across $7$ LGD datasets, spanning a range of performance indicators and experimental conditions. The most recent tabular foundation models, in particular TabPFN-3 and TabICLv2, achieve the strongest average results, with the largest gains in small-sample and strongly imbalanced settings. These results are notable because these models are used with their pretrained parameters fixed and without downstream hyperparameter optimization or task-specific parameter updates. ## When Does Self-Supervised Pretraining Help Tabular Models? A Study of Label Scarcity and Missing Data DevFeed: [When Does Self-Supervised Pretraining Help Tabular Models? A Study of Label Scarcity and Missing Data]() Original publisher: [Read original article]() Author: Sahand Mazrouei Published: 2026-10-26T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Self-supervised learning](), [Machine learning](), [Model Development]() Tags: [cs-lg](), [experiments](), [missing-data](), [performance](), [pretraining](), [self-supervised-learning](), [study](), [tabular-data](), [variance]() ### AI overview Across 14 tabular tasks, mask-and-recover self-supervised pretraining did not show a statistically significant overall gain over training from scratch in low-label settings. Its largest gains appeared on datasets with little or no recorded numeric missingness; it more often degraded performance on datasets with native numeric missingness. At 10% labels, Random Forest averaged 0.9015 AUC versus about 0.8954 for SSL. Under test-time MCAR and MNAR missingness, SSL had higher average AUC, but neither improvement was significant after correction. Mask-and-recover also showed no significant performance difference from simplified VIME, SCARF, or SubTab using the same encoder. The reported results are limited by single-run or single-seed comparisons. ### Source excerpt arXiv:2608.24381v2 Announce Type: replace Abstract: In this work, we conduct an empirical study on self-supervised learning (SSL) for tabular data. Our experiments examine limited label regimes (that is, 1%, 5%, 10%, and 20% of the labels) and highlight limitations and opportunities in current tabular SSL. Contrary to what is expected, our mask-and-recover approach shows its largest improvements for datasets that have little to no recorded numeric missingness, whereas it more often leads to degradation on datasets that are characterized by native numeric missingness (post hoc split). Random Forest achieves an average AUC of 0.9015 across the 14 tasks for the 10% labels setting, whereas SSL shows a performance of ~0.8954 (RF: 1 draw; SSL: 3-draw mean). Moreover, the gain of SSL compared to training from scratch exhibits a high variance across tasks and is not significant (1 run; Wilcoxon p = 0.626, Holm-adjusted p = 1.000, for both the 5% and 10% settings). On the other hand, we observe that SSL-pretrained models achieve a higher average AUC when MNAR and MCAR missingness were injected at test-time. In comparison to models trained from scratch, our SSL-pretrained models show an improvement of +0.0245 (MCAR) and +0.0418 AUC (MNAR) under test-time missingness (single seed), with 11 (MCAR; plus one tie) and 8 (MNAR) positive results across the 14 tasks. However, after Holm-Bonferroni correction neither improvement is significant (Holm-adjusted p = 0.118 for MCAR and 0.518 for MNAR). Finally, when comparing mask-and-recover with three other approaches, namely, simplified VIME, SCARF and SubTab implementations using an identical encoder, we observe no significant difference in performance (Holm-adjusted p = 0.459 for VIME, 1.000 for SCARF, 1.000 for SubTab). ## TabClustPFN: A Prior-data Fitted Network for Tabular Data Clustering DevFeed: [TabClustPFN: A Prior-data Fitted Network for Tabular Data Clustering]() Original publisher: [Read original article]() Author: Tianqi Zhao, Guanyang Wang, Yan Shuo Tan, Qiong Zhang Published: 2026-10-23T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [clustering](), [Neural Network]() Tags: [architecture](), [art](), [arxiv](), [benchmark](), [cardinality](), [clustering](), [code](), [context](), [cs-lg](), [data](), [dataset](), [deep](), [inference](), [invariant](), [network](), [network-architecture](), [partition](), [pass](), [performance](), [permutation](), [tabular-data]() ### AI overview TabClustPFN performs tabular clustering in a single pass, jointly inferring cluster assignments and the number of clusters without per-dataset optimization. It combines a hybrid pretraining prior intended to capture varied real-tabular geometry, separate networks for partition and cardinality inference, and a permutation-invariant SoftARI objective that avoids label-ordering conventions. The abstract says prior PFN clustering methods can require known cluster counts or depend on unstable label ordering and restrictive synthetic priors. On a curated benchmark of 44 real-world tabular datasets, TabClustPFN reportedly achieves state-of-the-art performance against classical, deep, and amortized baselines, with runtime comparable to efficient baselines. ### Source excerpt arXiv:2601.21656v5 Announce Type: replace Abstract: Prior-data Fitted Networks (PFNs) have reframed supervised tabular learning as single-pass in-context inference without per-dataset optimization. Extending this paradigm to clustering is appealing yet fundamentally more challenging, due to absent supervision, unknown cluster cardinality, and label switching inherent to partition outputs. Existing PFN-based clustering methods address these challenges only partially, either requiring known cardinality as input or relying on unstable label-ordering conventions and overly restrictive synthetic priors. We introduce TabClustPFN, a clustering PFN that resolves all challenges jointly through co-designed prior, objective, and architecture. Our hybrid pretraining prior captures heterogeneous real-tabular geometry; our decoupled partition inference network--cardinality inference network architecture jointly infers cluster assignments and cardinality in a single pass; and our SoftARI training objective is permutation-invariant by construction, eliminating the need for any label-ordering convention. On a 44 curated real-world tabular benchmark, TabClustPFN achieves state-of-the-art clustering performance against classical, deep, and amortized baselines, with runtime comparable to efficient baselines. Code is available at https://github.com/Tianqi-Zhao/TabClustPFN. ## In-Context Bias Propagation in LLM-Based Tabular Data Generation DevFeed: [In-Context Bias Propagation in LLM-Based Tabular Data Generation]() Original publisher: [Read original article]() Author: Pol G. Recasens, Alberto Gutierrez, Jordi Torres, Josep. Ll Berral, Javier Carnerero-Cano, Anisa Halimi, Kieran Fraser Published: 2026-10-22T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Large Language Model](), [Adversarial attacks](), [LLM security](), [Data Quality](), [Structured-data](), [prompt]() Tags: [adversarial](), [bias](), [classifiers](), [context](), [cs-lg](), [data](), [data-augmentation](), [dataset](), [examples](), [fairness](), [generation](), [global](), [groups](), [language](), [language-models](), [large-language-models-llms](), [llm](), [llms](), [malicious](), [performance](), [tabular-data]() ### AI overview The study finds that statistical bias in examples supplied to LLMs for in-context tabular data generation can distort the resulting synthetic data, even when the bias is mild. It also describes an adversarial scenario in which a contributor injects bias through a subset of examples, harming downstream classifier fairness for a targeted protected subgroup. Preprocessing examples and instruction-based defenses can attenuate disparity, but the models remain sensitive to adversarial prompts. These findings identify a vulnerability in synthetic data pipelines for sensitive domains. ### Source excerpt arXiv:2506.09630v3 Announce Type: replace Abstract: Large Language Models (LLMs) are increasingly used for synthetic tabular data generation through in-context learning (ICL), offering a practical solution for data augmentation in data scarce scenarios. While prior work has shown the potential of LLMs to improve downstream task performance through augmenting underrepresented groups, these benefits often assume access to a subset of unbiased in-context examples, representative of the real dataset. In real-world settings, however, data is frequently noisy and demographically skewed. In this paper, we systematically study how statistical biases within in-context examples propagate to synthetic tabular data, showing that even mild in-context biases lead to global statistical distortions. We further introduce an adversarial scenario where a malicious contributor can inject bias into the synthetic dataset via a subset of in-context examples, ultimately compromising the fairness of downstream classifiers for a targeted and protected subgroup. Finally, we evaluate prompt-side mitigation strategies based on preprocessing in-context examples and instruction-based defenses, demonstrating that while such techniques can attenuate disparity, the inherent sensitivity of LLMs to adversarial prompts remains a persistent challenge. Our findings highlight a critical new vulnerability in LLM-based data generation pipelines within sensitive domains. ## When Retain Constraints Conflict: Mitigating Forget-Retain Interference in Tabular Data DevFeed: [When Retain Constraints Conflict: Mitigating Forget-Retain Interference in Tabular Data]() Original publisher: [Read original article]() Author: Zijie Liu, Jinhao Duan, Bingqi Shang, Xinming An, Sijia Liu, Tianlong Chen Published: 2026-10-22T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Machine learning](), [Structured-data](), [Large Language Model](), [schema design]() Tags: [cs-lg](), [language-models](), [llm](), [machine-unlearning](), [medical](), [oracle](), [schema](), [tabular-data]() ### AI overview Serialized tabular records can make forgetting one row conflict with preserving nearby retained rows that share schema slots, attribute ranges, and high-signal features. The paper proposes Conflict-Aware Unlearning (CAU), which identifies retained rows that conflict most with the forget set and relaxes their preservation constraints. On clinical and non-medical tasks involving sample-level and feature-level unlearning, CAU reportedly comes closer to a retraining oracle while maintaining predictive utility and retain-region behavior. The abstract attributes reliable tabular LLM unlearning partly to how retain constraints are constructed. It does not provide quantitative results or implementation details. ### Source excerpt arXiv:2609.06786v2 Announce Type: replace Abstract: Machine unlearning aims to remove the influence of designated training data while preserving model utility, but its behavior on tabular data remains underexplored. This gap is important because tabular prediction is widely used in high-stakes domains and is increasingly adapted to language models through record serialization and schema-aware prompting. We identify a key challenge that distinguishes tabular unlearning from unlearning in free-form text or other modalities: schema-induced forget-retain overlap. In serialized tabular data, records share fixed column-name/value slots, similar attribute ranges, and common output spaces. Consequently, a forget row may have nearby retain rows that rely on the same high-signal attributes, causing retain preservation to oppose the update required for forgetting. Motivated by this failure mode, we propose Conflict-Aware Unlearning (CAU), a schema-aware approach that reduces forget-retain interference by relaxing preservation constraints on retained rows that most conflict with the forget set. Across sample-level and feature-level unlearning on clinical and non-medical tabular tasks, CAU more closely matches a retraining oracle while maintaining predictive utility and retain-region behavior. Our results show that reliable tabular LLM unlearning depends not only on the forgetting objective, but also on how retain constraints are constructed. ## Zero-Shot Tabular Learning via Hypergraph Reasoning with LLMs DevFeed: [Zero-Shot Tabular Learning via Hypergraph Reasoning with LLMs]() Original publisher: [Read original article]() Author: Geon Lee, Kyungho Kim, Dongwon Choi, Juyeon Kim, Shinhwan Kang, Kijung Shin Published: 2026-10-21T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Machine learning](), [Artificial Intelligence](), [Large Language Model](), [llm-reasoning](), [few-shot]() Tags: [classification](), [cost](), [cs-ai](), [cs-lg](), [data](), [dataset](), [dependencies](), [few-shot](), [finance](), [healthcare](), [inference](), [learning](), [llms](), [prompting](), [reasoning](), [rules](), [tabular-data](), [zero-shot]() ### AI overview TABHGR+ reduces the inference cost of zero-shot tabular classification by using a constant number of LLM calls, independent of dataset size. Given column and task descriptions, an LLM generates rules that assign weak labels and group semantically related rows into a hypergraph. The method refines noisy labels through hypergraph label propagation and uses label-adaptive hyperedge weights to improve propagation. The abstract reports higher classification accuracy than zero-shot and strong few-shot baselines; it provides no quantitative results or experiment details. ### Source excerpt arXiv:2610.23923v1 Announce Type: new Abstract: Tabular data are prevalent across diverse domains such as healthcare and finance. Zero-shot tabular learning aims to make predictions without labeled data. However, existing LLM-based approaches rely on row-wise prompting, resulting in high inference cost and limited use of inter-row dependencies. In this work, we propose TABHGR+, an efficient relation-aware zero-shot framework for tabular data that leverages LLM-generated rules to label and group rows. Specifically, based on column and task descriptions, we use LLMs to generate rules that assign weak labels to all rows. To refine these noisy weak labels, we construct a hypergraph by grouping semantically related rows using LLM-generated rules and perform hypergraph-based label propagation. Lastly, we further enhance the LLM-generated hypergraph for propagation, which is potentially sub-optimal, by assigning label-adaptive weights to hyperedges. By requiring only a constant number of LLM calls independent of dataset size, TABHGR+ substantially reduces inference cost while achieving higher classification accuracy than zero-shot and strong few-shot baselines. ## LightSplit: Practical Privacy-Preserving Split Learning via Orthogonal Projections (Full Version) DevFeed: [LightSplit: Practical Privacy-Preserving Split Learning via Orthogonal Projections (Full Version)]() Original publisher: [Read original article]() Author: Mert Cihangiroglu, Alessandro Pegoraro, Phillip Rieger, Antonino Nocera, Ahmad-Reza Sadeghi Published: 2026-10-21T04:00:00Z Content type: article Language: en Sources: [arXiv: Cryptography and Security](), [arXiv: Machine Learning]() Topics: [Machine learning](), [distributed-training](), [Security & Privacy](), [Fine-tuning]() Tags: [attacks](), [benchmarks](), [communication](), [compression](), [cs-cr](), [cs-lg](), [fine-tuning](), [image-classification](), [information-theory](), [language-model](), [learning](), [lightweight](), [privacy](), [projection](), [split](), [tabular-data]() ### AI overview LightSplit uses a fixed orthogonal random projection at the split-learning cut layer to send lower-dimensional activations, reducing communication and limiting information available for reconstruction. The server uses lifted representations, and the method adds no trainable client components while preserving exact gradient propagation. Because the projection is non-invertible, some client representation is discarded. In the reported benchmarks, compression up to 32x retained at least 94% of baseline accuracy on CIFAR-10, FashionMNIST, GTSRB, and UCI HAR, 86% on CIFAR-100, and 91% on average with non-IID data. At 8x, it retained 92% of ImageNet-1k top-1 accuracy and increased GPT-2 perplexity by at most 1.15x. Against reconstruction attacks, the strongest co-trained attack lost up to 87.3% of SSIM; UCI HAR subject re-identification fell by more than half. ### Source excerpt arXiv:2605.13265v2 Announce Type: replace Abstract: Split learning (SL) enables collaborative training by partitioning a neural network across clients and a central server. However, the cut-layer interface introduces a key challenge: high-dimensional activations incur substantial communication overhead while exposing representations vulnerable to reconstruction attacks. Existing approaches typically address efficiency or privacy in isolation. We propose LightSplit, which limits information exposure and reduces communication overhead via a lightweight fixed orthogonal random projection at the cut layer. Based on Shannon's information theory, this projection acts as an information bottleneck that restricts instance-specific information and suppresses exploitable per-sample signals. With low-dimensional projections sent instead of raw activations, the server operates on lifted representations without requiring changes to existing SL architectures. Adding no trainable client components, LightSplit remains lightweight and suitable for edge devices while preserving end-to-end differentiability via exact gradient propagation. As the projection is non-invertible, part of the original representation is irreversibly discarded at the client, so LightSplit reduces the information available for reconstruction. We evaluate LightSplit on seven benchmarks spanning image classification, tabular data, and language-model fine-tuning, under IID and non-IID client data. At up to 32x compression it retains at least 94% of baseline accuracy on CIFAR-10, FashionMNIST, GTSRB, and UCI HAR, 86% on CIFAR-100, and 91% on average under non-IID client data; at 8x it retains 92% of ImageNet-1k top-1 and raises GPT-2 perplexity by at most 1.15x. Against state-of-the-art reconstruction attacks LightSplit resists inversion; the strongest, co-trained attack loses up to 87.3% of its reconstruction SSIM, and subject re-identification on UCI HAR drops by more than half. ## Devavrat Shah's research and Ikigai Labs use tabular data for real-time forecasting and decision-making DevFeed: [Devavrat Shah's research and Ikigai Labs use tabular data for real-time forecasting and decision-making]() Original publisher: [Read original article]() Author: David Chandler | Laboratory for Information and Decision Systems Published: 2026-07-14T20:25:00Z Content type: article Language: en Sources: [MIT AI News]() Topics: [Artificial Intelligence](), [decision-making](), [Structured-data](), [AI Models](), [data](), [Time Series](), [Computer science](), [Electrical engineering and computer science (EECS)](), [Temporal data](), [systems]() Tags: [ai](), [ai-and-decision-making](), [ai-in-business-planning](), [ai-models](), [algorithms](), [artificial-intelligence](), [business-and-management](), [business-modeling](), [celonis-context-model](), [computer-science-and-technology](), [data](), [data-systems](), [decision-making](), [devavrat-shah](), [electrical-engineering-and-computer-science-eecs](), [faculty](), [forecasting](), [idss](), [ikigailabs](), [laboratory-for-information-and-decision-systems-lids](), [mit-eecs](), [mit-idss](), [mit-intellectual-property](), [mit-lids](), [mit-schwarzman-college-of-computing](), [predictions](), [profile](), [real-time](), [research](), [school-of-engineering](), [startups](), [structured](), [tabular-data](), [time-series-data]() ### AI overview MIT Professor Devavrat Shah's research led to a foundation model for tabular and time-series enterprise data. Developed with Ikigai Labs, the system continuously tests predictions against real outcomes to support large-scale forecasting and decision-making. ### Source excerpt Through research and entrepreneurship, Professor Devavrat Shah is helping to design methods that can handle constant decision-making using limited computational resources. [Next page]()