Interpretable by Design Tabular Prediction via In-Context Local Linear Learning

Ran Eisenberg; Steven Berger; Ofir Lindenbaum

cs.LGOct 29, 2026

Interpretable by Design Tabular Prediction via In-Context Local Linear Learning

Authors: Ran Eisenberg, Steven Berger, Ofir Lindenbaum

Organizations: Faculty of Engineering, Bar-Ilan University, Ramat Gan, Israel

Abstract

In-context learning (ICL) lets tabular foundation models make predictions without dataset-specific training or fine-tuning. These models achieve strong predictive accuracy, but interpreting how individual features contribute to the prediction remains difficult. In biology, for example, feature contributions can help identify biomarkers or risk genes. Post-hoc interpretability methods assign feature importance scores, but these scores do not indicate what the model actually uses and can lead to explanation errors. We introduce TICL $^3$ (Tabular In-Context Local Linear Learning), an interpretable-by-design predictor that uses a tabular foundation model to guide a simple linear prediction rule. TICL $^3$ first learns a shared linear model to capture dataset-level feature effects. Then, for each query, it retrieves similar training samples using representations from the foundation model and fits a second linear model to the residuals in that neighborhood. The shared model captures broad effects, while the local model corrects errors among similar samples. Their combined coefficients determine each feature contribution to the final query-specific linear score, so the same model makes and explains the prediction. Across 386 classification benchmark-dataset evaluations, TICL $^3$ trails TabICLv2 by 1.1 percentage points and outperforms CatBoost in 266 evaluations. It improves average support F1 over MAPLE by 4.7 points and achieves the best performance on both real-data selected-feature utility metrics.

Figures & tables

Figure 1: Summary of results. (a) Average per-dataset rank in classification and regression over all methods in Tables 2 and 6 ; a subset is shown. (b) Deletion monotonicity and selected-feature utility; the latter averages Gen-SVM and Gen-kmeans, reported separately in Table 5 . (c) Per-dataset cost for 50 predictions and explanations, including method-specific explainer construction. All methods explain the same TICL 3 (global + residual) predictor, so its setup is excluded for every method. Online latency is similar for TICL 3 (global + residual) and the learned explainers (about 27 ms per query); the difference comes from explainer construction. LIME and KernelSHAP are extrapolated from the fixed 10-query-per-dataset subset.

Method	Interpretable by design	Explicit additive prediction	Query-specific deployed rule	ICL-guided locality
TabICLv2	✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗	✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗	✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗	✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗
SHAP/LIME on TabICLv2	✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗	✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗	✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗	✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗
ExplainerPFN	✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗	✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗	✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗	✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗
EBM/NAM	✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓	✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓	✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗	✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗
MAPLE	✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓	✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓	✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓	✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗
LLSPIN	✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓	✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓	✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓	✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗

Table 1: Conceptual positioning of TICL 3 . Interpretable by design : the explanation is part of the deployed predictor; explicit additive prediction : prediction sums inspectable feature terms mapped to input columns; query-specific deployed rule : an explicit feature-level rule is fitted or selected per test point (sample weights or post-hoc explanations alone do not qualify); ICL-guided locality : an ICL representation defines locality. KernelICL uses ICL-derived query weights but no query-specific feature-level rule. ShapPFN sums per-feature contributions computed by its transformer but deploys no explicit feature-level rule; ExplainerPFN estimates attributions without access to a predictor.

Figure 2: Overview of TICL 3 (global + residual). Offline we use a frozen TabICLv2 backbone with cross-fitting to build a reference bank of processed features, class probability targets and learned representations. We fit a shared ElasticNet model to the bank targets. For each query, we then use its representation to select similar reference rows and fit a second ElasticNet model to the shared model’s remaining errors on those rows. We add the shared model and the local correction to give a single linear model for the query. We clip and normalize the class scores from this to make the final prediction. We also use the model coefficients and the centered feature values to compute the contribution of each feature to the class scores.

Family	Method	CC-18 $\uparrow$	TabArena-Lite $\uparrow$	TALENT $\uparrow$	BeyondArena $\uparrow$
Tabular foundation models	TabICLv2	$\mathbf{0.872\pm 0.154}$	$\mathbf{0.880\pm 0.087}$	$\mathbf{0.830\pm 0.148}$	$\mathbf{0.839\pm 0.132}$
Tabular foundation models	TabPFNv2	$\underline{0.865\pm 0.157}$	$\mathbf{0.880\pm 0.086}$	$\underline{0.825\pm 0.149}$	$\underline{0.835\pm 0.137}$
Interpretable by design	TICL 3 (global + residual)	$0.856\pm 0.161$	$\underline{0.877\pm 0.089}$	$0.817\pm 0.160$	$0.833\pm 0.134$
	TICL 3 (regional anchors)	$0.855\pm 0.162$	$0.874\pm 0.087$	$0.814\pm 0.164$	$0.834\pm 0.136$
	MAPLE	$0.830\pm 0.164$	$0.862\pm 0.091$	$0.790\pm 0.166$	$0.820\pm 0.137$
	EBM	$0.840\pm 0.159$	$0.865\pm 0.093$	$0.779\pm 0.186$	$0.825\pm 0.136$

Table 2: Predictive performance on CC-18, TabArena-Lite, TALENT, and BeyondArena under Section 4.1 . In that order, TabICLv2 uses 8/4/4/4 estimators and TabPFNv2 uses 8/8/4/8; both TICL 3 variants use one. The TICL 3 rows use a five-fold cross-fitted bank and the configurations in Section 3 and Appendix E.5 . TICL 3 (regional anchors) is grouped by its predictor form, but its predicted-class explanation is all zero for 73.3% of synthetic and 43.8% of real-data queries (Appendix E.5 ). The

\pm

values describe variation across dataset means.

Method	Mean Acc. $\uparrow$	Fit (s) $\downarrow$	Predict (s) $\downarrow$
TabICLv2 (8 estimators)	$\mathbf{0.8735}$	$\underline{0.950}$	$5.462$
TabICLv2 (1 estimator)	$\underline{0.8710}$	$\mathbf{0.485}$	$\mathbf{0.745}$
TabPFNv2	$0.8641$	$3.139$	$9.660$
TICL 3 (global + residual)	$0.8557$	$5.515$	$6.849$
TICL 3 -Fast (global + residual)	$0.8609$	$5.842$	$1.495$
TICL 3 (regional anchors)	$0.8555$	$6.736$	$\underline{0.850}$

Table 3: CC-18 accuracy and mean fit/prediction time over 69 datasets and three seeds. The default TabICLv2 row uses eight estimators; the matched row uses one. Configuration and timing details are in Appendix E.5 .

Interpretable-by-design predictors (own deployed accuracy)
	E1		E2		E3		E4
Method	F1 $\uparrow$	Acc. $\uparrow$	F1 $\uparrow$	Acc. $\uparrow$	F1 $\uparrow$	Acc. $\uparrow$	F1 $\uparrow$	Acc. $\uparrow$
TICL 3 (global + residual)	$0.512$	$\underline{0.663}$	$0.540$	$\underline{0.675}$	$0.554$	$\mathbf{0.787}$	$0.548$	$0.697$
MAPLE	$0.408$	$\mathbf{0.665}$	$0.480$	$\mathbf{0.684}$	$0.455$	$0.760$	$0.624$	$\mathbf{0.712}$
EBM	$0.439$	$0.636$	$0.361$	$0.653$	$0.494$	$0.690$	$0.541$	$0.688$
Logistic (ERM)	$0.422$	$0.535$	$0.343$	$0.649$	$0.491$	$0.656$	$0.493$	$0.632$

Table 4: Generating-support F1 and accuracy on LSPIN E1–E4 (10 seeds; 100 queries per task–seed). Blocks distinguish own-predictor from masked- TabICLv2 accuracy; E4 uses group-conditional support. An all-zero coefficient vector selects no features and receives support F1 zero.

Method	Deletion Spearman $\uparrow$	Gen-SVM $\uparrow$	Gen-kmeans $\uparrow$	Neighbour stab. $\downarrow$	Online ms/query $\downarrow$
TICL 3 (global + residual)	$\underline{0.470}$	$\mathbf{0.886}$	$\mathbf{0.697}$	$0.032$	$\mathbf{26.974}$
KernelSHAP	$0.384$	$0.708$	$0.555$	$0.037$	$53{,}686.935$
LIME	$0.449$	$0.813$	$0.601$	$0.032$	$14{,}607.496$
FastSHAP	$0.408$	$0.608$	$0.637$	$0.025$	$27.197$
IG on surrogate	$0.417$	$\underline{0.861}$	$\underline{0.669}$	$0.031$	$27.482$
GradientSHAP on surrogate	$0.446$	$0.856$	$0.666$	$0.031$	$27.560$

Table 5: Explanation quality and online prediction-plus-explanation cost on 10 CC-18 datasets. We evaluate quality using 500 queries; all rows use the same 422 queries with valid values for deletion and 493 for the other metrics. We exclude undefined values rather than setting them to zero. We average runtime across datasets, giving each dataset equal weight and excluding one-off setup. For LIME and KernelSHAP attribution time, we use a fixed subset of 100 queries (10 per dataset, three repeats); for the common predictor latency and the remaining rows, we use 50 queries per dataset.

Family	Method	Mean nRMSE $\downarrow$	Mean $R^{2}$ $\uparrow$
Tabular foundation models	TabICLv2	$\mathbf{0.479\pm 0.300}$	$\mathbf{0.681}$
	TabICLv2 (1 estimator)	$\underline{0.482\pm 0.303}$	$\underline{0.676}$
	TabPFNv2	$0.506\pm 0.310$	$0.648$
Interpretable by design	TICL 3 (global + residual)	$0.515\pm 0.287$	$0.653$
	TICL 3 (regional anchors)	$0.521\pm 0.275$	$0.654$
	MAPLE	$0.571\pm 0.298$	$0.586$

Table 6: TALENT regression on 119 splits. nRMSE uses the test-target standard deviation;

\pm

reports across-dataset variation.

Appendix figures & tables3 assets

Supplementary material from the paper’s appendix.

Appendix

Method	Fixed configuration	Selection/validation
TabICLv2	TabICL; 8/4/4/4 estimators; packaged defaults otherwise.	No tuning or validation selection.
TabPFNv2	TabPFN; 8/8/4/8 estimators; native multiclass model for at most 10 classes and one-vs-rest above 10; packaged defaults otherwise.	No tuning or validation selection.
KernelICL	Repository implementation with four TabICL estimators; $k$ NN uses $k=20$ and Gaussian uses scale 4.	Fixed a priori; no calibration or search.
MAPLE	scikit-learn random forest with 200 trees, max_features =0.5, minimum leaf size 10; per-query ridge penalty $0.001$ .	Stratified 80/20 training/validation split; retained count chosen by validation RMSE from candidates up to $\min(D,100)$ , using at most 50 validation queries.
EBM	InterpretML; 128 bins, 3000 maximum rounds, learning rate 0.01, and no interactions.	Fixed; package-internal fitting only, with no external search.
NAM	One hidden layer of width 32 per feature; at most 300 epochs; batch size 256; AdamW with learning rate $10^{-3}$ and weight decay $10^{-4}$ .	Stratified 80/20 training/validation split; validation-loss early stopping with patience 30.

Appendix

Table 7: Configurations for the predictive baselines in Table 2 . Suite order for ensemble counts is CC-18/TabArena-Lite/TALENT/BeyondArena. “Defaults” means the named package’s released default configuration except for the explicitly listed settings.

Global and local components
Configuration	Acc. (%) $\uparrow$	Nonzero expl. (%) $\uparrow$	Deletion Spearman $\uparrow$	Gen-SVM (%) $\uparrow$	Gen-kmeans (%) $\uparrow$
Local ElasticNet only ( $k=400$ , matched)	90.18	68.0	0.5207	70.72	55.77
Global ElasticNet only	76.13	100.0	0.6097	95.40	67.80
TICL — primary ( $k=400$ , 300 iterations)	89.90	100.0	0.5297	88.80	71.40
Local reference configuration
Local ElasticNet reference ( $k=100$ )	90.29	41.8	0.2807	53.03	52.97

Appendix

Table 8: Component and configuration ablations. Bold identifies the primary configuration, not column-wise best results. Iteration caps apply to the local correction; shared global fits retain 300 iterations. A nonzero explanation has a finite contribution vector with maximum absolute contribution above

10^{-12}

. Explanation metrics use configuration-specific valid populations and equally weight contributing datasets. Gen-SVM targets each predictor’s own classes; Gen-kmeans uses true labels. Evaluation counts appear in Table 9 .

Configuration	Valid deletion /500	Contributing datasets
Matched local K400	336	9 / 9 / 9
Global-only	500	10 / 10 / 10
TICL — primary	488	10 / 10 / 10
Local K100	209	7 / 6 / 6
TICL ( $k=30$ , 300 iterations)	493	10 / 10 / 10
TICL ( $k=400$ , 30 iterations)	488	10 / 10 / 10

Appendix

Table 9: Evaluation populations for the ablation. Every configuration has 500 attempted explanation queries and full-test accuracy on all ten datasets. Dataset counts refer to deletion Spearman, Gen-SVM and Gen-kmeans, respectively.

Interpretable ML

Interpretable machine learning develops models and explanations that make predictions, evidence, and decision processes understandable to people.

See all 310 papers

TabSHAP
Apr 22, 2026Aryan Chaudhary, Prateek Agarwal, Tejasvi AlladiFeature Attribution SHAP Feature Attribution
Department of Computer Science Birla Institute of Technology and Science India
TabICLv2: A better, faster, scalable, and open tabular foundation model
Feb 11, 2026Jingang Qu, David Holzmüller, Gaël Varoquaux +1Synthetic Data Pretraining Tabular Foundation Models
SODA Team, INRIA Saclay, Palaiseau, France · Probabl, France
Real-Time Explanations for Tabular Foundation Models
Mar 31, 2026Luan Borges Teodoro Reis Sena, Francisco Galuppo AzevedoNeural Network Interpretability Tabular Foundation Models
Kunumi Institute, Belo Horizonte, Minas Gerais, Brazil · Universidade Federal de Minas Gerais, Belo Horizonte, Minas Gerais, Brazil

Apr 22, 2026J/K move · Enter open · S save

TabSHAP

Aryan Chaudhary, Prateek Agarwal, Tejasvi Alladi

Department of Computer Science Birla Institute of Technology and Science India

Feature Attribution SHAP Feature Attribution LLM Interpretability Tabular ML

Open paper AskPDF arXiv

Large Language Models (LLMs) fine-tuned on serialized tabular data are emerging as powerful alternatives to traditional tree-based models, particularly for heterogeneous or context-rich datasets. However, their deployment in high-stakes domains is hindered by a lack of faithful interpretability; existing methods often rely on global linear proxies or scalar probability shifts that fail to capture the model's full probabilistic uncertainty. In this work, we introduce TabSHAP, a model-agnostic interpretability framework designed to directly attribute local query decision logic in LLM-based tabular classifiers. By adapting a Shapley-style sampled-coalition estimator with Jensen-Shannon divergence between full-input and masked-input class distributions, TabSHAP quantifies the distributional impact of each feature rather than simple prediction flips. To align with tabular semantics, we mask at the level of serialized key:value fields (atomic in the prompt string), not individual subword tokens. Experimental validation on the Adult Income and Heart Disease benchmarks demonstrates that TabSHAP isolates critical diagnostic features, achieving significantly higher faithfulness than random baselines and XGBoost proxies. We further run a distance-metric ablation on the same test instances and TabSHAP settings: attributions are recomputed with KL or L1 replacing JSD in the similarity step (results cached per metric), and we compare deletion faithfulness across all three.

Abstract

Interpretable ML

Explore similar work

TabSHAP

TabICLv2: A better, faster, scalable, and open tabular foundation model

Real-Time Explanations for Tabular Foundation Models