Erchao Li

Incomplete multi-view feature selection identifies informative original features when some samples lack entire views. Under informative view missingness, selectively observed sample pairs may not represent the target population, so graphs learned from these pairs bias feature evaluation even when observed entries are well reconstructed. We formulate this discrepancy as a population relation shift and propose Doubly Robust Population-Graph Learning for Incomplete Multi-view Feature Selection (DRPG-IMFS). The framework combines cross-fitted observation propensities with target-view affinity predictions from auxiliary views to construct a doubly robust correction of pairwise relations. Under conditional exchangeability and positivity, given a common conditioning set, the raw correction is conditionally mean-calibrated when either nuisance function is correct. The corrected relations are used to construct population graph anchors that guide working completion and constrain iterative graph refinement. Reliability-weighted consensus learning and row-sparse projections then produce feature rankings without class labels. Experiments on 11 multi-view datasets under three simulated view-missingness mechanisms compare DRPG-IMFS with 11 competing methods using five clustering metrics. With completed clustering inputs for all methods, DRPG-IMFS achieves a 16.8% relative improvement over TERUIMUFS in the mean normalized area under the clustering accuracy curve over feature budgets. Real-world validation on TCGA-BRCA demonstrates the benefit of full relation construction under native multi-omics view missingness across both clustering input settings. Code will be released after the review process.

Addressing the challenge that, in real-world dynamic multi-objective optimization problems (DMOPs), the severity of changes between pareto optimal set (PS) varies at different times and exhibits nonlinear characteristics rather than simple translations or rotations—making them difficult for traditional prediction strategies to track accurately—this paper proposes a dynamic multi-objective optimization algorithm via trend-cycle decoupling and hybrid time-series prediction. The algorithm first applies the Hodrick-Prescott (HP) filter to decompose the time-series of historical PS centers into a smooth trend component and a fluctuating cycle component to cope with uncertainty in the severity of changes. Then, an AR(p) model is used to fit the trend sequence and infer the long-term linear direction of PS movement; a long short-term memory (LSTM) network learns the cycle sequence to capture nonlinear variation patterns. By fusing the two prediction results, the center of the PS in the new environment is located, and an initial population is constructed using a manifold-based population generation strategy. Comparative experiments on 13 standard dynamic test functions show that the proposed algorithm achieves an effective trade-off between prediction accuracy and computational cost and demonstrates strong robustness to complex time-varying environments. In particular, in scenarios where the pareto optimal front (PF) undergoes rotation, discontinuity, or time-varying shape (convexity/concavity) due to complex mappings in the decision space, the algorithm maintains notable tracking accuracy and population diversity by precisely capturing the PS evolution trajectory.

Hybrid motion-force control (HMFC) of redundant manipulators is a fundamental yet challenging problem due to structural redundancy and multiple constraints. Traditional approaches exhibit low convergence and sensitivity to time-varying disturbances. To address these limitations, a HMFC approach for redundant manipulators with a robust finite-time gradient neural network (RFTGNN) is proposed. It reformulates the motion-force control problem into a constrained quadratic program (QP) and derives the corresponding Karush-Kuhn-Tucker (KKT) conditions. RFTGNN, implemented with an adaptive scaling mechanism, is established to address the HMFC problem of redundant manipulators. The proposed method enables adaptive adjustments of its convergence rate in response to real-time task errors, ensuring stable finite-time tracking while preserving stability in the presence of noise. Theoretical analyses confirm the finite-time tracking of redundant manipulators under HMFC. The effectiveness of the proposed approach is demonstrated through simulations and physical experiments conducted on a Franka Emika Panda manipulator.

This article studies the formation control of stochastic heterogeneous multiagent systems (HMASs) subject to compound (additive and multiplicative) noises and directed topologies. Due to the intrinsic coupling between the directed topologies and heterogeneous dynamics, classical Lyapunov-based stability techniques become inapplicable. Therefore, we introduce a velocity-decomposition technique and a semidecoupling method to address the formation control of stochastic HMASs. First, the boundedness of agents’ states is established using properties of Hurwitz matrices. Then, by applying the Jordan decomposition of the system matrix, some convergence conditions of the closed-loop system are derived. To the best of our knowledge, this work is the first to study the formation control of stochastic HMASs under compound noises and directed topologies and to derive necessary and sufficient conditions for achieving the desired formation. Finally, numerical simulations and physical experiments are given to validate theoretical results.

Owing to the ability to model competitive behavior, $\boldsymbol{k}$-winners-take-all ($\boldsymbol{k}$WTA) networks play a crucial role in swarm control. However, most existing $\boldsymbol{k}$WTA networks rely on centralized architectures and focus primarily on model noise, while largely neglecting the inevitable communication noise. To address these limitations, this article proposes a distributed communication-noise-resistant $\boldsymbol{k}$WTA (DC-$\boldsymbol{k}$WTA) network and applies it to the multirobot task allocation. Due to the communication noise, the classical methods for deterministic convergence analysis based on LaSalle’s invariance principle are no longer applicable. Therefore, the stochastic stability theory and Itô’s formula are employed to analyze the convergence of the proposed DC-$\boldsymbol{k}$WTA network under additive and multiplicative noise, respectively. Finally, the superiority and effectiveness of the proposed DC-$\boldsymbol{k}$WTA network are demonstrated through simulations and physical experiments.