Regularized High-Dimensional Additive Tensor Autoregressive Model
本文提出了一种正则化加性张量自回归模型，通过行、列和管方向的时间依赖性的加性交互作用来提高解释性和减少计算负担，并估计转移矩阵的低秩加稀疏模式。
0 citationsRead paper
Selected work
Representative Papers
Learning Auditable Classifier Models: Source-Disjoint Tree Ensembles
This study addresses the challenges of auditing tree-based models and the limited expressiveness of interpretable alternatives in regulated domains by proposing RPTE, a three-stage learning framework. By leveraging source-disjoint constraints and separated coefficient estimation, RPTE constructs sparse models decomposable into sums of named rule contributions, thereby ensuring inherent auditability by design. Evaluated across twelve clinical benchmarks, RPTE achieves accuracy comparable to mainstream ensemble methods while reducing audit units by 9 to 87 times relative to XGBoost and maintaining lower complexity than Explainable Boosting Machines (EBM). Consequently, this approach effectively reconciles high predictive performance with low auditing costs and strong interpretability, offering a viable solution for high-stakes applications where model transparency and regulatory compliance are paramount.
0 citationsRead paper
Designing Compact Neural Architectures via Neuron Gating and Mixed Activation
This study addresses the computational expense and trade-offs between compactness and performance in neural architecture search by proposing a general bilevel optimization framework. Through three scalable continuous relaxation formulations, discrete decisions are transformed into differentiable problems, enabling efficient architecture search at both neuron and activation levels. The proposed method outperforms DARTS on MNIST and CIFAR-10 benchmarks, achieving high accuracy with significantly fewer parameters. These results effectively validate its advantages in model compression and performance enhancement, establishing a new paradigm for constructing lightweight yet high-performance neural networks.
0 citationsRead paper
LP-NAS: Linear Programming-based Neural Architecture Search
This study addresses the inefficiency and limited generalization inherent in differentiable neural architecture search (NAS) by proposing a novel linear programming-based framework. The method constructs a linear programming model utilizing validation loss gradients and training loss Hessians to determine architecture update directions, thereby ensuring parameter optimality while significantly enhancing generalization capability. Experimental results demonstrate that this framework achieves faster convergence and superior early-stage validation performance. Notably, it outperforms DARTS and its variants on both CIFAR and ImageNet benchmarks, establishing an efficient and robust approach to architecture search.
0 citationsRead paper
Complexity-Budgeted, Interaction-Aware Interpretable Model for Tabular Data
This work addresses a critical limitation of traditional interpretable models, which often overlook interaction effects that emerge only through specific joint configurations of variables, leading to significant information loss. To overcome this, the authors propose the IAIML framework—the first approach to explicitly model pairwise interactions within compact interpretable models. IAIML identifies key interactions via adaptive discretization, finite-grid interaction scoring, and an interaction-aware feature admission mechanism, while controlling model complexity through an interpretability budget. Evaluated across 40 datasets, IAIML achieves an average AUC only 1.4% lower than gradient-boosted trees, with interpretability components reduced by 14–28 times. Notably, it substantially outperforms existing baselines in scenarios dominated by strong interactions, effectively balancing predictive accuracy and model interpretability.
0 citationsRead paper
Recent publications
Latest Papers
Regularized High-Dimensional Additive Tensor Autoregressive Model
本文提出了一种正则化加性张量自回归模型，通过行、列和管方向的时间依赖性的加性交互作用来提高解释性和减少计算负担，并估计转移矩阵的低秩加稀疏模式。
0 citationsRead paper
Learning Auditable Classifier Models: Source-Disjoint Tree Ensembles
This study addresses the challenges of auditing tree-based models and the limited expressiveness of interpretable alternatives in regulated domains by proposing RPTE, a three-stage learning framework. By leveraging source-disjoint constraints and separated coefficient estimation, RPTE constructs sparse models decomposable into sums of named rule contributions, thereby ensuring inherent auditability by design. Evaluated across twelve clinical benchmarks, RPTE achieves accuracy comparable to mainstream ensemble methods while reducing audit units by 9 to 87 times relative to XGBoost and maintaining lower complexity than Explainable Boosting Machines (EBM). Consequently, this approach effectively reconciles high predictive performance with low auditing costs and strong interpretability, offering a viable solution for high-stakes applications where model transparency and regulatory compliance are paramount.
0 citationsRead paper
Designing Compact Neural Architectures via Neuron Gating and Mixed Activation
This study addresses the computational expense and trade-offs between compactness and performance in neural architecture search by proposing a general bilevel optimization framework. Through three scalable continuous relaxation formulations, discrete decisions are transformed into differentiable problems, enabling efficient architecture search at both neuron and activation levels. The proposed method outperforms DARTS on MNIST and CIFAR-10 benchmarks, achieving high accuracy with significantly fewer parameters. These results effectively validate its advantages in model compression and performance enhancement, establishing a new paradigm for constructing lightweight yet high-performance neural networks.
0 citationsRead paper
LP-NAS: Linear Programming-based Neural Architecture Search
This study addresses the inefficiency and limited generalization inherent in differentiable neural architecture search (NAS) by proposing a novel linear programming-based framework. The method constructs a linear programming model utilizing validation loss gradients and training loss Hessians to determine architecture update directions, thereby ensuring parameter optimality while significantly enhancing generalization capability. Experimental results demonstrate that this framework achieves faster convergence and superior early-stage validation performance. Notably, it outperforms DARTS and its variants on both CIFAR and ImageNet benchmarks, establishing an efficient and robust approach to architecture search.
0 citationsRead paper
Complexity-Budgeted, Interaction-Aware Interpretable Model for Tabular Data
This work addresses a critical limitation of traditional interpretable models, which often overlook interaction effects that emerge only through specific joint configurations of variables, leading to significant information loss. To overcome this, the authors propose the IAIML framework—the first approach to explicitly model pairwise interactions within compact interpretable models. IAIML identifies key interactions via adaptive discretization, finite-grid interaction scoring, and an interaction-aware feature admission mechanism, while controlling model complexity through an interpretability budget. Evaluated across 40 datasets, IAIML achieves an average AUC only 1.4% lower than gradient-boosted trees, with interpretability components reduced by 14–28 times. Notably, it substantially outperforms existing baselines in scenarios dominated by strong interactions, effectively balancing predictive accuracy and model interpretability.
0 citationsRead paper