Retrieval-Augmented Generation (RAG) enhances large language models (LLMs) by integrating up-to-date external knowledge, yet real-world web environments present unique challenges. These limitations manifest as two key challenges: pervasive misinformation in the web environment, which introduces unreliable or misleading content that can degrade retrieval accuracy, and the underutilization of web tools, which, if effectively employed, could enhance query precision and help mitigate this noise, ultimately improving retrieval results in RAG systems. To address these issues, we propose WebFilter, a novel RAG framework that generates source-restricted queries and filters out unreliable content. This approach combines a retrieval filtering mechanism with a behavior- and outcome-driven reward strategy, optimizing both query formulation and retrieval outcomes. Extensive experiments demonstrate that WebFilter improves answer quality and retrieval precision, outperforming existing RAG methods on both in-domain and out-of-domain benchmarks.

Reinforcement learning (RL) has emerged as a powerful framework to improve the reasoning performance of large language models (LLMs), with approaches such as Group Relative Policy Optimization (GRPO) showing promising results. However, GRPO and its variants struggle with collapsed groups (i.e., all-correct or all-incorrect completions), leading to zero-variance rewards and ineffective gradient signals. Moreover, focusing solely on final answer correctness while ignoring the reasoning process, along with rigid length penalties, can hinder training stability and output quality. To address these issues, we introduce TAPO, a reinforcement learning framework that enhances optimization signals by modifying sampled completions within training groups. TAPO incorporates three core techniques: (1) Dynamic Teacher Injection (DTI), which selectively injects high-quality or adversarial examples to restore effective gradient signals in collapsed groups; (2) Perturbed Answer Injection (PAI), which makes partially correct completions to provide contrastive supervision separating reasoning correctness but wrong answer from the trajectories; and (3) InfoLen-Aware Reward Shaping, a fine-grained reward strategy that penalizes outputs based on both length and semantic redundancy, encouraging concise yet informative responses. Extensive experimental results demonstrate that TAPO significantly improves the mathematical reasoning capabilities of LLMs across multiple challenging benchmarks, outperforming the GRPO baseline by a substantial margin. Component-wise ablations further validate the contribution of each proposed technique.

As one of the most basic specifications for many types of sensors, sensor resolution has been largely overlooked in a multitude of state estimation studies. Under the binary-encoding mechanism, this article deals with the outlier-resilient state estimation problem of Markov jump neural networks (MJNNs) with sensor resolution. An improved binary-encoding procedure, capable of assigning distinct bit lengths to different MJNN modes, is proposed to accommodate diverse physical constraints and limited network resources. Building on this procedure, a mode-dependent state estimation scheme embedded with a saturation function is put forward to alleviate the by-effects of external disturbances, decoding errors, and measurement outliers. Sufficient conditions are derived to guarantee the exponential ultimately boundedness of estimation errors. Lastly, simulation experiments are carried out to demonstrate the applicability of the proposed method.

Jun Yin, Pengyu Zeng, Haoyuan Sun, Yuqin Dai, Han Zheng, Miao Zhang, Yachao Zhang, Shuai Lu. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

In recent years, architectural design automation has made significant progress, but the complexity of open-world environments continues to make residential design a challenging task, often requiring experienced architects to perform multiple iterations and human-computer interactions.Therefore, assisting ordinary users in navigating these complex environments to generate and edit residential design is crucial.In this paper, we present the CARD framework, which leverages a system of specialized crossmodal agents to adapt to complex open-world environments.The framework includes a pointbased cross-modal information representation (CMI-P) that encodes the geometry and spatial relationships of residential rooms, a crossmodal residential generation model, supported by our customized Text2FloorEdit model, that acts as the lead designer to create standardized floor plans, and an embedded expert knowledge base for evaluating whether the designs meet user requirements and residential codes, providing feedback accordingly.Finally, a 3D rendering module assists users in visualizing and understanding the layout.CARD enables cross-modal residential generation from freetext input, empowering users to adapt to complex environments without requiring specialized expertise.

Residential design is a complex and open-ended problem that requires designers to integrate diverse types of input information while adhering to stringent energy consumption standards. However, most current research in this field focuses on generating floor plans from a limited set of input types, often neglecting to incorporate energy-related physical constraints. Existing approaches are limited by: (1) the lack of multimodal datasets in this domain, (2) the absence of comprehensive residential energy consumption data, and (3) the challenges associated with effectively integrating multiple input types into a unified model. To address these challenges, we propose MRED-14, the first large-scale Multimodal Residential Energy Dataset, comprising 14 input types, including energy consumption values, vector drawings, and textual descriptions, paired with 41,280 high-quality residential floor plans that have been scored and annotated by human experts. Based on this dataset, we introduce the LER-net model, which can flexibly adapt to various input types and generate low-energy residential floor plans. Experimental results demonstrate that LER-net outperforms existing models, achieving state-of-the-art performance under the same input conditions. In addition, the energy consumption of the generated floor plans is reduced by 5.1% compared to the actual residential designs. Further expert evaluations confirm the LER-net model's feasibility for use in residential design.

In this article, the issue of event-triggered multiasynchronous$H_{\infty }$control for Markov jump systems with transmission delay is concerned. In order to reduce sampling frequency, multiple event-triggered schemes (ETSs) are introduced. Then hidden Markov model (HMM) is employed to describe multiasynchronous jumps among subsystems, ETSs, and controller. Based on the HMM, the time-delay closed-loop model is constructed. In particular, when triggered data are transmitted over networks, a large transmission delay may cause disorder of transmission data such that the time-delay closed-loop model cannot be developed directly. To overcome this difficulty, a packet loss schedule is presented and the unified time-delay closed-loop system is obtained. By the use of the Lyapunov–Krasovskii functional method, sufficient conditions with the controller design are formulated for guaranteeing the$H_{\infty }$performance of the time-delay closed-loop system. Finally, the effectiveness of the proposed control strategy is demonstrated by two numerical examples.

With the ubiquitous IoT sensors and enormous real-time data generation, data privacy is becoming a critical societal concern. State-of-the-art privacy protection methods all demand significant hardware overhead due to computation-insensitive algorithms and divided sensor/security architecture. In this paper, we propose a generic time-domain circuit architecture that protects raw data by enabling a differentially-private compressed sensing (DP-CS) algorithm secured by physical unclonable functions (PUF). To address privacy concerns and hardware overhead at the same time, a robust unified PUF and time-domain mixed-signal (TD-MS) module are designed, where PUF enables private and secure entropy generation. To evaluate the proposed design against a digital baseline, we performed experiments based on synthesized circuits and SPICE simulation and measured a 2.9x area reduction and 3.2x energy gains. We also measured high-quality PUF generation with TD-MS circuit with a inter-die Hamming distance of 52% and a low intra-die Hamming distance of 2.8%. Furthermore, we performed attack and algorithm performance measurements demonstrating the proposed design preserves data privacy even under attack, and the machine learning performance has minimal degradation (within 2%) compared to the digital baseline.

The neural-network (NN)-based state estimation issue of Markov jump systems (MJSs) subject to communication protocols and deception attacks is addressed in this article. For relieving communication burden and preventing possible data collisions, two types of scheduling protocols, namely: 1) the Round-Robin (RR) protocol and 2) weighted try-once-discard (WTOD) protocol, are applied, respectively, to coordinate the transmission sequence. In addition, considering that the communication channel may suffer from mode-dependent probabilistic deception attacks, a hidden Markov-like model is proposed to characterize the relationship between the malicious signal and system mode. Then, a novel adaptive neural state estimator is presented to reconstruct the system states. By taking the influence of deception attacks into performance analysis, sufficient conditions under two different scheduling protocols are derived, respectively, so as to ensure the ultimately boundedness of the estimate error. In the end, simulation results testify the correctness of the adaptive neural estimator design method proposed in this article.

This article investigates event-triggered and self-triggered$\mathcal {L}_{\infty }$control problems for the Markov jump stochastic nonlinear systems subject to denial-of-service (DoS) attacks. When attacks prevent system devices from obtaining valid information over networks, a new switched model with unstable subsystems is constructed to characterize the effect of DoS attacks. On the basis of the switched model, a multiple Lyapunov function method is utilized and a set of sufficient conditions incorporating the event-triggering scheme (ETS) and restriction of DoS attacks are provided to preserve$\mathcal {L}_{\infty }$performance. In particular, considering that ETS based on mathematical expectation is difficult to be implemented on a practical platform, a self-triggering scheme (STS) without mathematical expectation is presented. Meanwhile, to avoid the Zeno behavior resulted from general exogenous disturbance, a positive lower bound is fixed in STS in advance. In addition, the exponent parameters are designed in STS to reduce triggering frequency. Based on the STS, the mean-square asymptotical stability and almost sure exponential stability are both discussed when the system is in the absence of exogenous disturbance. Finally, two examples are given to substantiate the effectiveness of the proposed method.

The security control issue of nonlinear networked systems is considered in this article on the basis of interval type-2 fuzzy modeling strategy. For avoiding communication congestion, a stochastic scheduling strategy called Markovian communication protocol is introduced to coordinate the sensor transmission order. An asynchronous observer is designed for estimating the unmeasured states via the hidden Markov model. In addition, a more comprehensive scenario on deception attack is considered, in which attacks occur both in the sensor–observer and controller–actuator communication channels with different types of deception signals. In view of slack matrix approach and stochastic analysis technique, some sufficient conditions for ensuring the ultimately boundedness of the resulting closed-loop system are obtained. In the end, simulations show the validity of the proposed protocol-based fuzzy control method.

This article is concerned with the neural network (NN)-based event-triggered control problem for discrete-time networked Markov jump systems with hybrid cyberattacks and unmeasured states. The event-triggered mechanism (ETM) is used to reduce the communication load, and a Luenberger observer is introduced to estimate the unmeasured states. Two kinds of cyberattacks, denial-of-service (DoS) attacks and deception attacks, are investigated due to the vulnerability of cyberlayer. For the sake of mitigating the impact of these two types of cyberattacks on system performance, the ETM under DoS jamming attacks is discussed first, and a new estimation of such mechanism is given. Then, the NN technique is applied to approximate the injected false information. Some sufficient conditions are derived to guarantee the boundedness of the closed-loop system, and the observer and controller gains are presented by solving a set of matrix inequalities. The effectiveness of the presented control method is demonstrated by a numerical example.

In this article, the event-triggered resilient$\mathcal {L}_{\infty }$control problem is concerned for the Markov jump systems in the presence of denial-of-service (DoS) jamming attacks. First, a fixed lower bound-based event-triggering scheme (ETS) is presented in order to avoid the Zeno problem caused by exogenous disturbance. Second, when DoS jamming attacks are involved, the transmitted data are blocked and the old control input is kept by using the zero-order holder (ZOH). On the basis of this process, the effect of DoS attacks on ETS is further discussed. Next, by utilizing the state-feedback controller and multiple Lyapunov functions method, some criteria incorporating the restriction of DoS jamming attacks are proposed to guarantee the$\mathcal {L}_{\infty }$control performance of the event-triggered Markov closed-loop jump system. In particular, the bounded transition rates rather than the exact ones are taken into account. That is appropriate for the practical environment in which transition rates of the Markov process are difficult to measure accurately. Correspondingly, some criteria are proposed to obtain state-feedback gains and event-triggering parameters simultaneously. Finally, we provide two examples to show the effectiveness of the proposed method.

This article is concerned with$H_\infty$control problem for discrete-time Takagi–Sugeno fuzzy Markov jump systems with event-triggering scheme (ETS) and denial-of-service (DoS) attacks. Aperiodic DoS attacks characterized by duration and frequency are introduced. In the light of DoS attacks and event-triggered fuzzy controller, the switched fuzzy Markov jump closed-loop system is established. Particularly, in order to address the issue of mismatched behavior between membership functions of fuzzy system and fuzzy controller, a novel ETS consisting of membership functions is developed. Then, with the help of multiple Lyapunov function method and iterative technique, some sufficient conditions are provided to ensure the$H_\infty$performance of the resulting closed-loop system. Subsequently, the explicit parameter design of controller and ETS is provided. Finally, an example is employed to verify the validity of the proposed theoretical method.

The stability issue of stochastic hybrid systems against energy-constrained denial-of-service (DoS) attacks is investigated in this article. A sampled-data-based Round-Robin protocol, which sends measurement data cyclically according to the predetermined transmission sequence, is introduced to avoid communication network congestion. Additionally, a new switched time-delay stochastic closed-loop system with an unstable subsystem is established by discussing the effect of DoS attacks. Then, the stability of the closed-loop system is discussed in light of the piecewise Lyapunov–Krasovskii functional method and stochastic analysis technique. Finally, two examples are included to illustrate the correctness and applicability of the proposed method.

This article investigates stabilization problems of networked event-triggered switched systems under multiasynchronous switching. Different from the existing asynchronous literature, a novel problem is considered in this article, i.e., the event-triggering scheme and controller both have independent switching delays relative to the system, which is called multiasynchronous switching. An adaptive event-triggering scheme with switching structure is utilized to achieve better adjustment in triggering frequency. A systematic framework for analyzing multiasynchronous switching stability and solving controller gains is established. Different Lyapunov functionals are employed, and new tight bound conditions on average dwell time are constructed. Besides, an active packet loss approach for handling packets disorder issues caused by large transmission delays is presented. Since the activation instants of the system, triggering scheme, and controller are mutually staggered with the data updating instants of the actuator, a novel analytical method aiming at the coupling effect is proposed. Finally, the validity of the adopted method in this article is demonstrated.

This paper studies the problem of event-triggered and self-triggered H∞controls for uncertain switched linear systems with exogenous disturbances whose magnitude is bounded by the system state's norm. With the proposed schemes, the control task is carried out only when the triggering condition is met. That leads to changing and adaptive interexecution intervals and can further reduce the system resource costs to a certain extent. Moreover, to ensure the H∞control performance of the event-triggered switched system, a proof which makes certain that the occurrence of Zeno problem can be prevented is first provided. Then, by adopting the average dwell time method, a set of sufficient conditions for H∞performance analysis is developed. Subsequently, the codesign of controller gains and event-triggering schemes is provided. On the basis of the eventtriggered control, a solution to the self-triggered control problem is then presented. It can only utilize the current sampled data to predict the next triggered instant. Finally, in order to test the effectiveness of the proposed methods, numerical simulations are performed.