TabArena launches a dynamic, updatable benchmarking system for tabular ML that shows boosted trees remain competitive, deep learning matches them under larger budgets with ensembling, foundation models excel on small data, and cross-model ensembles advance SOTA while flagging validation overfitting.
InterpretML: A unified framework for machine learning interpretability.arXiv preprint arXiv:1909.09223, 2019
18 Pith papers cite this work, alongside 322 external citations. Polarity classification is still indexing.
18
Pith papers citing it
322
external citations · external index
citation-role summary
background 3
citation-polarity summary
roles
background 3polarities
background 3representative citing papers
TFM-Retouche is an architecture-agnostic input-space residual adapter that improves tabular foundation model accuracy on 51 datasets by learning input corrections through the frozen backbone, with an identity guard to fall back to the original model.
Agentic-imodels evolves scikit-learn regressors via an autoresearch loop to jointly boost predictive performance and LLM-simulatability, improving downstream agentic data science tasks by up to 73% on the BLADE benchmark.
TabDistill distills feature interactions from tabular foundation models via post-hoc attribution and inserts them into GAMs, yielding consistent predictive gains.
GPT-4 exceeds the USMLE passing score by more than 20 points and outperforms both GPT-3.5 and the medically fine-tuned Med-PaLM on the MultiMedQA benchmarks.
FlagGAM builds sparse univariate rule bases from features and feeds them into a restricted additive model, achieving competitive accuracy with superior robustness to missingness and noise on tabular benchmarks.
DEM distills XGBoost into a residual decision tree with a new fidelity metric for interpretable anomaly detection in WBAN data, reporting AUC 0.9964 and 0.9047 with 0.17ms inference.
Introduces a Riesz basis for explicit generalized functional ANOVA decomposition under input dependence and an associated data-driven estimation procedure.
Corrected empirical limits show the most massive galaxies never exceed the theoretical baryonic maximum of 0.16 times halo virial mass, keeping observations consistent with LambdaCDM at all redshifts.
ParamBoost improves GAMs by fitting piecewise cubic polynomials via gradient boosting and supports constraints for continuity, monotonicity, convexity, and feature interactions.
A new pipeline for interpretable heterogeneous regression that combines response-informed random Fourier features, PCA embedding, GMM soft clustering, and cluster-specific spline GAMs.
Adds a trainable feature selection layer to NAM and NBM to cut computational cost, enable two-input interaction networks in high dimensions, and match or exceed state-of-the-art GAM performance.
NRI is a pretrained model for zero-shot logical rule induction that uses domain-agnostic statistical literal encodings and a parallel slot-based decoder.
Gradient boosting produces risk scores with competitive accuracy but 60% fewer rules on classification tasks and 16% fewer on time-to-event tasks than regression-based methods like AutoScore.
Zero-shot vision-language models are unreliable and vary widely for depression screening, and explainability-based fairness interventions often trade away accuracy without reliable fairness gains.
A radiation-bias correction using generalized additive models is calibrated on collocated low-cost and professional sensors then applied to Swiss urban networks to reassess tropical nights and heat warnings.
This survey organizes intrinsic interpretability approaches for LLMs into five categories—functional transparency, concept alignment, representational decomposability, explicit modularization, and latent sparsity induction—while discussing challenges and future directions.
Machine learning assists with image denoising and soliton detection in cold-atom quantum simulators while addressing the need for model interpretability.
citing papers explorer
-
TabArena: A Living Benchmark for Machine Learning on Tabular Data
TabArena launches a dynamic, updatable benchmarking system for tabular ML that shows boosted trees remain competitive, deep learning matches them under larger budgets with ensembling, foundation models excel on small data, and cross-model ensembles advance SOTA while flagging validation overfitting.
-
TFM-Retouche: A Lightweight Input-Space Adapter for Tabular Foundation Models
TFM-Retouche is an architecture-agnostic input-space residual adapter that improves tabular foundation model accuracy on 51 datasets by learning input corrections through the frozen backbone, with an identity guard to fall back to the original model.
-
Agentic-imodels: Evolving agentic interpretability tools via autoresearch
Agentic-imodels evolves scikit-learn regressors via an autoresearch loop to jointly boost predictive performance and LLM-simulatability, improving downstream agentic data science tasks by up to 73% on the BLADE benchmark.
-
Selecting Feature Interactions for Generalized Additive Models by Distilling Foundation Models
TabDistill distills feature interactions from tabular foundation models via post-hoc attribution and inserts them into GAMs, yielding consistent predictive gains.
-
Capabilities of GPT-4 on Medical Challenge Problems
GPT-4 exceeds the USMLE passing score by more than 20 points and outperforms both GPT-3.5 and the medically fine-tuned Med-PaLM on the MultiMedQA benchmarks.
-
FlagGAM: Rule-Basis Generalized Additive Models for Explainable Tabular Prediction
FlagGAM builds sparse univariate rule bases from features and feeds them into a restricted additive model, achieving competitive accuracy with superior robustness to missingness and noise on tabular benchmarks.
-
DEM: A Distilled Explanation Model for Interpretable Anomaly Detection in Physiological Sensor Networks
DEM distills XGBoost into a residual decision tree with a new fidelity metric for interpretable anomaly detection in WBAN data, reporting AUC 0.9964 and 0.9047 with 0.17ms inference.
-
Generalized Functional ANOVA in Closed-Form: A Unified View of Additive Explanations
Introduces a Riesz basis for explicit generalized functional ANOVA decomposition under input dependence and an associated data-driven estimation procedure.
-
Empirical estimates of how massive galaxies can be in {\Lambda}CDM
Corrected empirical limits show the most massive galaxies never exceed the theoretical baryonic maximum of 0.16 times halo virial mass, keeping observations consistent with LambdaCDM at all redshifts.
-
ParamBoost: Gradient Boosted Piecewise Cubic Polynomials
ParamBoost improves GAMs by fitting piecewise cubic polynomials via gradient boosting and supports constraints for continuity, monotonicity, convexity, and feature interactions.
-
Cluster-Based Generalized Additive Models Informed by Random Fourier Features
A new pipeline for interpretable heterogeneous regression that combines response-informed random Fourier features, PCA embedding, GMM soft clustering, and cluster-specific spline GAMs.
-
Neural Additive and Basis Models with Feature Selection and Interactions
Adds a trainable feature selection layer to NAM and NBM to cut computational cost, enable two-input interaction networks in high dimensions, and match or exceed state-of-the-art GAM performance.
-
A Foundation Model for Zero-Shot Logical Rule Induction
NRI is a pretrained model for zero-shot logical rule induction that uses domain-agnostic statistical literal encodings and a parallel slot-based decoder.
-
Gradient Boosted Risk Scores
Gradient boosting produces risk scores with competitive accuracy but 60% fewer rules on classification tasks and 16% fewer on time-to-event tasks than regression-based methods like AutoScore.
-
FAIR_XAI: Improving Multimodal Foundation Model Fairness via Explainability for Wellbeing Assessment
Zero-shot vision-language models are unreliable and vary widely for depression screening, and explainability-based fairness interventions often trade away accuracy without reliable fairness gains.
-
Revisiting urban heat indices in Switzerland using low-cost measurement networks
A radiation-bias correction using generalized additive models is calibrated on collocated low-cost and professional sensors then applied to Swiss urban networks to reassess tropical nights and heat warnings.
-
Towards Intrinsic Interpretability of Large Language Models:A Survey of Design Principles and Architectures
This survey organizes intrinsic interpretability approaches for LLMs into five categories—functional transparency, concept alignment, representational decomposability, explicit modularization, and latent sparsity induction—while discussing challenges and future directions.
-
Can machine learning for quantum-gas experiments be explainable?
Machine learning assists with image denoising and soliton detection in cold-atom quantum simulators while addressing the need for model interpretability.