Results support IAIML as a compact, interaction-aware framework appropriate for settings where bounded explanation size and controlled treatment of feature interactions are design requirements.
Abstract
Inherently interpretable classifiers for tabular data typically rely on sparse features, rules, or patterns that users can inspect directly. The marginal feature-screening step common to these methods can discard variables whose predictive value emerges only through joint configurations with other variables. We present Interaction Aware Interpretable Machine Learning (IAIML), a framework that addresses this limitation through three coordinated mechanisms: adaptive per-feature discretization, finite-grid pairwise interaction scoring, and a partitioned explanation budget. Detected interactions are routed through one of two strategies: relaxing the screening filter so that interaction-supported variables enter the pattern search, or constructing explicit pair terms for a sparse downstream classifier. On a 40-dataset panel comprising 24 real-world tabular benchmarks and 16 synthetic interaction stress tests, evaluated under nested cross-validation, IAIML achieves mean AUC within 1.4 points of tuned gradient-boosted ensembles while requiring roughly 14--28 times fewer fitted explanation components. On datasets with strong pairwise interaction structure and low marginal signal, IAIML outperforms all baselines. Among compact interpretable methods, IAIML is comparable to RuleFit in AUC and component count and is less expensive to tune. EBM obtains a small but significant AUC advantage across the full panel, with a substantially larger lookup-table footprint. Performance degrades on datasets requiring higher-order interactions beyond the pairwise scope. Component-isolated ablations confirm that adaptive discretization and interaction-aware admission each contribute incrementally. These results support IAIML as a compact, interaction-aware framework appropriate for settings where bounded explanation size and controlled treatment of feature interactions are design requirements.
Exact interpretable learning is attractive in regulated decision settings, but solver runtime can vary substantially across datasets and solver families. We introduce structural meta-features derived from Feature Interaction Graphs (FIG s) as interpretable signals for solver selection. We construct FIG s from binarized tabular data using pairwise mutual information and extract topology-aware signatures such as density and estimated treewidth. Using a transparent shallow decision-tree selector, we demonstrate that FIG features establish an interpretable structural view of solver behavior, complementing basic, statistical, and landmarking meta-features. Experiments on OpenML classification tasks show that topology-aware profiling exposes meaningful structural variation across datasets, although benchmark saturation prevents clear end-to-end routing gains over strong simple baselines. Our results validate FIG as a principled, interpretable diagnostic tool for algorithm selection in exact learning; its diagnostic relevance becomes apparent on harder instances where solver runtime separation is substantial.
Zhigao Huang, Miao Pan, Yuzhuo Pan et al.· Scientific Reports· 0 citations
Understanding which parameters are influential in Large Language Models (LLMs) is central to improving their efficiency, reliability, and interpretability. We introduce Weight-Adjusted Gradients (WAG), a simple yet effective approach for estimating parameter importance that explicitly captures the interaction between model weights and first-order gradient information and identifies parameters that disproportionately influence model behavior, such as those responsible for collapse phenomena in LLMs. Across a range of models and settings, we show that WAG surfaces a tiny but critical subset of parameters whose modification leads to dramatic degradation in performance, a failure mode that existing importance metrics overlook. These findings reveal a previously underexplored interplay between weights and gradients, suggesting that parameter importance cannot be fully understood through either signal alone. The surprising effectiveness of WAG points to fundamental structural properties of trained networks and motivates new open questions about the role of zeroth-order and first-order information in deep learning. We demonstrate the practical utility of WAG across multiple applications, including expert allocation in mixture-of-expert architectures, parameter-specific unlearning, mixed-precision quantization, and layer selection for knowledge editing. Our results position WAG as a unified approach for analyzing, debugging, and controlling LLMs, and opens new directions for principled model-level interpretation.
Tabular deep learning is still challenging in real-world settings. Many datasets include both numerical and categorical variables, substantial missingness, and a need for not only strong classification performance but also interpretability and reliable probability estimates. DA2-Net is proposed to address this problem through a dual-branch architecture. It combines an interpretable additive pathway for feature-wise main effects with a selective self-attention pathway for higher-order interactions. In this design, features are ranked using additive contribution magnitude, uncertainty, and missingness-aware scaling. Only a Top-K subset is then passed to a single multi-head self-attention block. The final prediction is obtained through uncertainty-aware gated fusion. The model is also supported by sparsity, stability, and Brier-based calibration regularization. This allows it to balance expressive interaction modeling with transparency and robustness under incomplete data. DA2-Net is evaluated on four public binary tabular benchmarks, namely AdultIncome, DefaultCredit, HeartDisease, and BankMarketing, under controlled Missing Completely At Random (MCAR) missingness levels of 0.0, 0.1, 0.2, and 0.3. The evaluation uses 5-fold stratified cross-validation repeated across three random seeds. This produces 15 runs for each dataset and missingness condition, and 128 evaluation blocks in total across AUC, AUPRC, ACC, F1, sensitivity, specificity, Brier score, and Expected Calibration Error (ECE). Across this benchmark, DA2-Net achieves the best overall mean rank with 3.078 ± 2.044, ahead of SAINT-Lite at 3.980 ± 2.624. It achieves or shares the best result in all 16 AUC blocks, 13 of 16 AUPRC blocks, 10 of 16 ACC blocks, 11 of 16 Brier blocks, and 7 of 16 ECE blocks. These results show that its main strength lies in robust ranking-based discrimination and strong overall probability quality under missingness. It also shows a favorable practical-efficiency profile in the current benchmark, remaining more compact and inference-efficient than the main transformer-like baselines. Epoch-wise loss analysis also shows stable convergence across all four datasets. The binary cross-entropy (BCE) term drives the optimization, while the auxiliary regularizers act as controlled refinements. The ablation study further confirms that the interaction branch is essential. Removing it in the AdditiveOnly variant causes the clearest degradation in both predictive and calibration metrics. In contrast, removing the gate or the auxiliary regularization terms leads only to minor changes. A sensitivity analysis also supported the selected interaction subset size k=10 and spline knot count K=8 as balanced settings, while additive shape-function visualizations provided direct qualitative evidence for feature-wise interpretability.
Mücahit Cihan· Gazi university journal of s...· 0 citations
Given a model that is already trained, which features does it rely on causally versus spuriously? Existing methods require access to the training procedure and cannot answer this post-hoc. We introduce the \textbf{Normalised Sensitivity Ratio~(NSR)}, a post-hoc, model-agnostic diagnostic for this question under a structured-shift regime: environments differ primarily in the mean of spurious features while the causal mechanism and causal marginals remain stable, as in multi-site clinical data or multi-batch genomics. Within this regime, causal features induce constant model sensitivity across environments while spurious features track shift. NSR formalises this as the squared coefficient of variation of per-environment sensitivity. Under a linear structural causal model (SCM) with $K\ge3$ non-degenerate environments, NSR achieves exact identification (Theorem~1). We fully characterise failure: weak shifts ($O(\varepsilon^4)$ collapse), degenerate geometry, and proxy attenuation ($O((1-\alpha)^4)$), giving practitioners quantitative criteria for assessing whether the regime holds. Finite-sample rates are $O_p(n^{-1})$ under the null and $O_p(n^{-1/2})$ under the alternative. Experiments confirm all theoretical predictions on synthetic data (area under the ROC curve [AUROC] $= 1.000$ under conditions satisfying the regime), show consistent rankings across five model families (Kendall $\tau\ge0.529$), and recover six of eight causal features on bike-sharing data (Precision@7 $= 0.75$) without modifying any trained model.
Athanasios Vlontzos, G. Papanastasiou, Bernhard Kainz et al.· 0 citations
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 Data Foundry, 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 existing tabular foundation models excel on tiny- to medium-sized IID data, while traditional tree-based and deep learning models still dominate on non-IID, large, and high-dimensional datasets. BeyondArena guides model research for the most demanding challenges in tabular data, enabling progress towards truly foundational tabular models.
Lennart Purucker, Andrej Tschalzev, Nick Erickson et al.· arXiv.org· 2 citations
Interpretability methods for neural network activations span a wide cost spectrum, from cheap, training-free techniques (such as linear probes, PCA, SVD) to more expensive training-based ones (such as SAEs and activation oracles). Training-based methods are typically more powerful, in part because they leverage large activation datasets during training. This raises a natural question - do they actually surface insights that go beyond what is recoverable from the training dataset itself? To address this, we equip an LLM agent with a vector database of activations paired with their textual contexts, along with tools for manipulating activations - projecting out directions in latent space, computing activation differences and averages. The agent iteratively queries the database, forms hypotheses from the retrieved samples, and validates them by constructing linear probes. We call this method HARP, for Hypothesis-driven Agentic Retrieval and Probing. Despite not involving any training, HARP outperforms both activation oracles and SAE-based agents on concept discovery, concept detection, model steering, and secret elicitation. The training-free design also makes HARP substantially cheaper and more flexible: new datasets can be indexed on demand whenever existing ones prove insufficient. More broadly, our results suggest that current training-based methods do not yet extract insights beyond their training data, and motivate benchmarks that explicitly require interpretability methods to demonstrate such insights. We release our code at https://github.com/SriramB-98/HARP