Author name cluster

Cheng Tang

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

7 papers

2 author rows

AAAI Conference 2026 Conference Paper

CloserToMe: A Unified Framework for Accurate and Transferable Latency Prediction Across Heterogeneous Devices

Cheng Tang
Guochong Sui
Wenqi Lou
Zihan Wang
Jiayi Tuo
Wenqian Xie
Yinkang Gao
Yixuan Zhu

Hardware accelerators such as GPUs, NPUs, and FPGAs are essential to meeting AI’s computational demands. With the proliferation of heterogeneous devices across cloud and edge, various model optimization techniques adapt to diverse hardware characteristics through operator transformations and structural modifications. Accurate, efficient latency prediction enables rapid selection of optimal strategies across hardware backends. Many existing methods treat hardware as a black-box executor, directly regressing latency without explicitly modeling the intricate interactions between neural network (NN) structures and device-specific execution behaviors. To address these challenges, we introduce a new modeling perspective that captures the interaction between neural architectures and hardware execution. To capture device-specific characteristics, we propose two complementary modeling strategies. The Device Behavior Signature Selector (DBSel) characterizes hardware execution behavior by selectively probing a small set of representative architectures, forming a compact, workload-driven profile. In parallel, we construct capability vectors that capture the hierarchical memory of each device and compute characteristics, providing a structured abstraction of its architectural capacity. To unify both behavioral and structural views, we introduce the Hardware–Operation Dialogue Module (HODM), which models fine-grained interactions between neural operators and hardware properties. Together, these components empower CloserToMe to deliver accurate and transferable latency predictions across unseen and diverse platforms.

PDF Details DOI

EAAI Journal 2026 Journal Article

Evaluating a novel incremental-input neural network for multivariate air temperature forecasting

Zhenyu Song
Shuangyu Song
Shuangbao Song
Lixing Tan
Cheng Tang
Junkai Ji

Air temperature prediction (ATP) plays a crucial role in meteorological applications, such as agricultural planning, disaster forecasting, and energy management. However, the existing methods often struggle with the challenges posed by nonstationary and nonlinear time series data. In this paper, we introduce a novel incremental-input neural network (IINN) model that is designed to improve the accuracy and stability of multivariate ATP processes. By leveraging an incremental-input mechanism, the IINN addresses key challenges such as gradient vanishing and explosion while enhancing the robustness and nonlinear modelling capacity of the model for use with high-dimensional datasets. Comprehensive evaluations conducted on the Seoul metropolitan summer temperature dataset demonstrate that the IINN achieves state-of-the-art performance across two forecasting horizons. Specifically, compared with the best-performing baseline model, the IINN produces a 6. 1% MSE improvement for the minimum temperature ( T m i n ) and a 5. 8% improvement for the maximum temperature ( T m a x ). Thus, this work provides a significant step forward in the field of air temperature forecasting, offering a lightweight, efficient, and interpretable solution for modelling complex, nonstationary time series. The proposed approach offers a new and practical paradigm for modelling multivariate temperature time series and shows strong potential for broader applications in environmental forecasting scenarios.

Details DOI

EAAI Journal 2025 Journal Article

MFLSCI: Multi-granularity fusion and label semantic correlation information for multi-label legal text classification

Chunyun Meng
Yuki Todo
Cheng Tang
Li Luan
Zheng Tang

Multi-label text classification tasks face challenges such as sample diversity, complexity, and the need for effective utilization of label correlations. In this paper, we propose a model that integrates multi-granularity fusion of text sequence features and label semantic correlation information. Our model leverages graph convolutional networks to extract label semantic correlation, which enhances classification performance for samples with similar labels and addresses label omission issues. Additionally, text convolutional neural networks are employed to extract multi-granularity sense group features from text sequences, calculate their similarity with semantic correlation label distributions, and dynamically adjust the similarity between text context and label information. This approach tackles the limitations of feature extraction in short texts and label confusion. We replace the original multi-hot label encoding in model training with a label distribution that fuses text multi-granularity sense group features and label correlation information, using a more precise encoding method for soft alignment based on label probability distributions. This enhances the model’s resilience to noisy data, avoiding the issue of assigning high-confidence probabilities to incorrect categories due to hard-coded supervision. Our model’s performance improvement on noisy datasets significantly surpasses that achieved by label smoothing. Extensive experiments on three legal text datasets and two generalized multi-label datasets demonstrate the model’s excellent performance. Our approach is applicable in various real-world scenarios, such as legal judgment prediction, news categorization, and recommendation systems, where accurate multi-label classification is crucial. Ablation and experiments on noisy datasets validate the model’s effectiveness and robustness.

Details DOI

ICML Conference 2025 Conference Paper

Robust Offline Reinforcement Learning with Linearly Structured f-Divergence Regularization

Cheng Tang
Zhishuai Liu
Pan Xu 0002

The Robust Regularized Markov Decision Process (RRMDP) is proposed to learn policies robust to dynamics shifts by adding regularization to the transition dynamics in the value function. Existing methods mostly use unstructured regularization, potentially leading to conservative policies under unrealistic transitions. To address this limitation, we propose a novel framework, the $d$-rectangular linear RRMDP ($d$-RRMDP), which introduces latent structures into both transition kernels and regularization. We focus on offline reinforcement learning, where an agent learns policies from a precollected dataset in the nominal environment. We develop the Robust Regularized Pessimistic Value Iteration (R2PVI) algorithm that employs linear function approximation for robust policy learning in $d$-RRMDPs with $f$-divergence based regularization terms on transition kernels. We provide instance-dependent upper bounds on the suboptimality gap of R2PVI policies, demonstrating that these bounds are influenced by how well the dataset covers state-action spaces visited by the optimal robust policy under robustly admissible transitions. We establish information-theoretic lower bounds to verify that our algorithm is near-optimal. Finally, numerical experiments validate that R2PVI learns robust policies and exhibits superior computational efficiency compared to baseline methods.

Details

EAAI Journal 2020 Journal Article

A novel machine learning technique for computer-aided diagnosis

Cheng Tang
Junkai Ji
Yajiao Tang
Shangce Gao
Zheng Tang
Yuki Todo

The primary motivation of this paper is twofold: first, to employ a heuristic optimization algorithm to optimize the dendritic neuron model (DNM) and second, to design a tidy visual classifier for computer-aided diagnosis that can be easily implemented on a hardware system. Considering that the backpropagation (BP) algorithm is sensitive to the initial conditions and can easily fall into local minima, we propose an evolutionary dendritic neuron model (EDNM), which is optimized by the gbest-guided artificial bee colony (GABC) algorithm. The experiments are performed on the Liver Disorders Data Set, the Wisconsin Breast Cancer Data Set, the Haberman’s Survival Data Set, the Diabetic Retinopathy Debrecen Data Set and Hepatitis Data Set, and the effectiveness of our model was rigorously validated in terms of the classification accuracy, the sensitivity, the specificity, the F_measure, Cohen’s Kappa, the area under the receiver operating characteristic curve (AUC), convergence speed and the statistical analysis of the Wilcoxon signed-rank test. Moreover, after training, the EDNM can simplify its neural structure by removing redundant synapses and superfluous dendrites by the neuronal pruning mechanism. Finally, the simplified structural morphology of the EDNM can be replaced by a logic circuit (LC) without sacrificing accuracy. It is worth emphasizing that once implemented by an LC, the model has a significant advantage over other classifiers in terms of speed when handling big data. Consequently, our proposed model can serve as an efficient medical classifier with excellent performance.

Details DOI

NeurIPS Conference 2019 Conference Paper

Exponentially convergent stochastic k-PCA without variance reduction

Cheng Tang

We present Matrix Krasulina, an algorithm for online k-PCA, by gen- eralizing the classic Krasulina’s method (Krasulina, 1969) from vector to matrix case. We show, both theoretically and empirically, that the algorithm naturally adapts to data low-rankness and converges exponentially fast to the ground-truth principal subspace. Notably, our result suggests that despite various recent efforts to accelerate the convergence of stochastic-gradient based methods by adding a O(n)-time variance reduction step, for the k- PCA problem, a truly online SGD variant suffices to achieve exponential convergence on intrinsically low-rank data.

PDF Details

NeurIPS Conference 2018 Conference Paper

When do random forests fail?

Cheng Tang
Damien Garreau
Ulrike von Luxburg

Random forests are learning algorithms that build large collections of random trees and make predictions by averaging the individual tree predictions. In this paper, we consider various tree constructions and examine how the choice of parameters affects the generalization error of the resulting random forests as the sample size goes to infinity. We show that subsampling of data points during the tree construction phase is important: Forests can become inconsistent with either no subsampling or too severe subsampling. As a consequence, even highly randomized trees can lead to inconsistent forests if no subsampling is used, which implies that some of the commonly used setups for random forests can be inconsistent. As a second consequence we can show that trees that have good performance in nearest-neighbor search can be a poor choice for random forests.

PDF Details