Fixed-time adaptive output feedback control of stochastic nonlinear systems with actuator faults and unmodeled dynamics

Composite learning adaptive event-triggered output feedback control of linear 2 × 2 hyperbolic PDE systems

Evaluating adaptive and generative AI-based feedback and recommendations in a knowledge-graph-integrated programming learning system

Alterations in the mitogen-activated protein kinase (MAPK) pathway play a central role in colorectal cancer (CRC) tumor biology, therapeutic response, and resistance. RAS oncogenic mutations are present in up to 35% of CRC; they represent constitutively active molecular switches that impair GTP hydrolysis and promote ligand-independent signaling. The development of allele-specific KRAS G12C inhibitors exploits a cryptic pocket to trap the protein in its inactive (OFF) conformation. In metastatic CRC, where EGFR feedback loop limits monotherapy efficacy, combinations of KRAS G12C inhibitors with EGFR blockade and chemotherapy have demonstrated meaningful activity. Parallel efforts allowed the development of RAS(ON) inhibitors that sterically block the interaction with downstream effectors. Co-occurring genomic alterations, MAPK pathway reactivation, and pharmacologic limitations drive primary and acquired resistance. BRAF V600 mutations drive approximately 10% of CRCs; first-generation, type I BRAF inhibitors require combination with upstream EGFR, downstream MEK/ERK blockade to overcome adaptive feedback reactivation. Emerging strategies include paradox breaking RAF inhibitors, type II pan-RAS inhibitors, and immunotherapy combinations, particularly relevant for the immune-activated phenotype of BRAF-mutant microsatellite stable tumors. Beyond RAS and RAF, novel therapeutic avenues (such as antibody-drug conjugates, bispecific antibodies, DNA damage response inhibitors, and tumor microenvironment modulating agents) are reshaping precision oncology in CRC. Integrating multiomics profiling with dynamic biomarkers may enable durable activity and personalized treatment strategies.

Successful goal-directed movements depend on the central nervous system's (CNS) ability to handle diverse physical interactions. The CNS is thought to handle different dynamical contexts through three mechanisms: (i) trial-by-trial adaptation when forces are predictable, (ii) a model-free robust control strategy, and (iii) online adaptation of feedback responses. While each has been studied independently, their relative contributions and the possibility that they are recruited to different extents across contexts is unknown. Here, we quantified all three strategies within the same individuals to examine how CNS exploits them under varying environmental conditions. Participants (19 female, 15 male) performed reaching tasks while interacting with robot-generated force-fields that were either consistent or varied unpredictably. Trial-by-trial adaptation was measured using standard force channels to isolate anticipatory compensation. Robust control was assessed through movement velocity and corrective force magnitude. Online adaptive control was quantified by the temporal alignment between commanded and measured forces within a movement. Results showed that participants improved anticipatory compensation in consistent environments and relied on both robust and online adaptation when perturbations were unpredictable. Crucially, markers of robust control dominated the early movement phase, whereas online adaptation dominated later corrections. This temporal dissociation was confirmed by electromyographic recordings. Markers of robust and online adaptive feedback strategies also statistically predicted participants' ability to adapt across trials in consistent environments, revealing a common trait linking online control and adaptation. These findings reveal a rich and flexible combination of control mechanisms, offering a new framework for understanding the neurophysiological bases of reaching control.Significance Statement Human reaching control is a complex behavior resulting from several mechanisms that orchestrate feedback responses to mechanical perturbations and adaptation to changes in the environment. Here we combine previously studied paradigms to highlight within the same groups of healthy volunteers that three major components are recruited to different extents dependent on the context: unpredictable environment promote concomitant use of robust control and online adaptation whereas predictable environments recruit standard adaptation based on anticipatory compensation. Remarkably, individuals' adaptive capabilities correlated across consistent and inconsistent environments, suggesting a key involvement of adaptive mechanisms in both online control and trial-by-trial adaptation. Robust control, online adaptation, and anticipatory compensation are dissociable behaviorally, and are used to varying levels as a result of individual traits.

This article focuses on stabilizing uncertain nonlinear systems with limited communication resources. Traditional approaches relying on static quantizers or fixed-gain observers face significant limitations. To solve this, an adaptive observer-based quantized output feedback control framework is proposed. A dynamic-gain state observer is developed, with observer gains adjusted by a differential equation to handle nonlinearities and quantization effects. A criterion for choosing quantization parameters is established, linking them to control gains, observer dynamics, and bounded uncertainties. This confines quantization errors and ensures global asymptotic stability of the closed-loop system. Simulations on a robotic manipulator system validate the superiority of the proposed method. The work integrates dynamic observer adaptation and quantizer design, promoting resource-efficient control in bandwidth and resource-constrained applications.

Despite widespread claims that Generative Artificial Intelligence (GenAI) will transform education, longitudinal empirical evidence on its pedagogical integration remains limited. This study examines how GenAI use shapes teaching and learning practices over time. Using a mixed methods longitudinal design, the study draws on 34 semi structured interviews conducted at two time points, six to eight months apart, with 17 secondary school teachers who independently adopted GenAI tools. The analysis was triangulated with 212 GenAI-supported teaching and learning activities. A theory-driven classification based on the SAMR framework was combined with inductive thematic analysis and quantitative pre-post comparisons. The findings, based on a thematic analysis of teacher discourse, reveal differentiated trends in opportunities and challenges. Opportunities related to fostering creativity increased over time, whereas efficiency, workload reduction, and teacher empowerment remained stable. Concerns regarding content quality and inherent biases showed a marginal increase, while references to prohibited or improper use declined. Regarding teaching and learning activities, a significant increase was observed in teaching-related uses of GenAI over time. In addition, a significant increase was identified at the Modification level, indicating a shift toward more advanced forms of pedagogical redesign, particularly through the development of personalized materials, AI-supported instructional planning, and adaptive feedback practices, while learning activities at higher levels remained comparatively stable. Taken together, these findings position the SAMR as a dynamic framework for examining longitudinal patterns of GenAI integration and suggest that GenAI currently accelerates instructional innovation more than it fundamentally restructures student learning paradigms.

Closed-loop bioelectronic systems that adapt stimulation to real-time physiological feedback hold transformative potential for treating neurological and cardiac disorders and are emerging as key components of future ultrasonic brain-machine interfaces (uBMIs). Realizing this requires the simultaneous achievement of millimeter‑scale deep-tissue targeting, artifact-free physiological feedback, and robust wireless power and data transfer, which remain elusive with current methods. Here, we present an integrated ultrasonic platform engineered to overcome these fundamental limitations. We propose a physics-constrained metasurface design framework to enable high-resolution multifocal ultrasound energy delivery through highly aberrating biological barriers such as the skull and ribs, achieving improved experimental targeting accuracy (e.g., ±6.5% intensity uniformity across multiple foci). We demonstrate the platform's adaptive stimulation capabilities through two distinct paradigms: attention-based ultrasound stimulation and cardiac-synchronized ultrasound stimulation. Furthermore, we introduce a novel dual-channel acoustic link that enables continuous wireless power and wireless data streaming through the skull with a single acoustic metasurface, demonstrating robustness even with a 400-fold power differential. This integrated ultrasonic framework, providing seamless integration of precise spatial targeting through biological barriers, adaptive physiological feedback, and untethered operation, contributes to the development of next-generation uBMIs and closed-loop bioelectronic therapies.

Reward provides a feedback signal that modulates behaviour through several mechanisms, including invigorating performance and learning of action-outcome associations to guide future choices. After stroke, the ability to utilise reward feedback can be impaired, which may limit the benefits of rehabilitation approaches that use reinforcement. One possibility is that stroke causes a global impairment of reward processing, leading to both reduced invigoration and diminished learning from feedback. Alternatively, reward processing may be selectively disrupted, such that either invigoration or the ability to update beliefs from reward feedback is disproportionately affected. To test these competing hypotheses, we recruited forty chronic stroke survivors and thirty age-matched healthy controls to complete a probabilistic reversal learning task with both their strong (non-paretic/dominant) and weak (paretic/non-dominant) limb. On each trial, participants reached to one of two targets associated with different reward probabilities that changed unpredictably over time, requiring continued monitoring of outcomes and adaptation of choice behaviour. Stroke survivors showed reduced reward-based learning compared to controls, expressed as lower overall choice accuracy and a greater tendency to switch responses after rewarded trials (i.e., lower win-stay rates), particularly when using the weak upper limb. Control analyses confirmed that these selective impairments were not explained by general motor impairment or cognitive deficits. To identify the putative computations underlying these behavioural differences in reward-based learning we used an established model of hierarchical Bayesian inference, the Hierarchical Gaussian Filter (HGF). The HGF characterises learning dynamics as trial-by-trial updating of an agent's beliefs about action-outcome probabilities and their change over time (environmental volatility). Compared to healthy controls, stroke survivors were slower to update their beliefs about action-reward contingencies, an effect most pronounced for the weak upper limb, whereas updating beliefs about environmental volatility remained intact. Reward-based invigoration was also preserved: stronger trial-by-trial predictions about action-reward contingencies were associated with faster movement times, with comparable slopes of this association across groups, indicating that motivational drive was maintained in patients despite overall slower performance. This behavioural dissociation between preserved motivational invigoration but impaired probabilistic reward-based learning highlights a key translational opportunity: to leverage intact motivational pathways to enhance rehabilitation intensity and compliance, and to develop adaptive feedback strategies that compensate for impaired reward learning. Harnessing these complementary approaches could strengthen recovery outcomes and support greater long-term independence after stroke.

The activated sludge process serves as the core barrier in pharmaceutical wastewater treatment, yet its stability is inherently challenged by the extreme complexity of influent composition and the unpredictability of toxic shocks, particularly under contract development and manufacturing organization (CDMO) operations. Current biotoxicity assessment methods face inherent trade-offs among timeliness, specificity, and matrix robustness, resulting in fragmented, reactive management that lacks predictive capacity. In response, this review critically synthesizes evidence on toxicity pathways and monitoring technologies, systematically evaluating their mechanistic basis and engineering applicability. Building on these findings, we propose a conceptual perception–cognition–response architecture that structures decision-making across three adaptive tiers: (i) a perception layer that tolerates false positives for rapid anomaly detection; (ii) a cognition layer that requires effect-based biological verification; and (iii) a response layer that authorizes resilience-oriented interventions. Rather than a linear pipeline, the three tiers form an adaptive feedback cycle that dynamically aligns monitoring intensity, verification depth, and response authority with real-time risk gradients and site-specific constraints. By explicitly linking biological mechanisms to assessment limitations and tiered decision rules, this review provides a hypothesis-generating roadmap that orients biotoxicity management from episodic, composition-based assessment toward adaptive, effect-driven control. The proposed framework is intended to guide future pilot validation, multi-sensor integration, and context-specific calibration, offering a unified narrative for advancing proactive biotoxicity control in complex pharmaceutical wastewater systems.

ABSTRACT Taiwan’s 2019 National Curriculum Guidelines highlight assessment for learning (AfL) as central to student-centered instruction. Dynamic Assessment (DA) emphasizes providing calibrated support to help learners move beyond independent performance, offering a viable pathway for enacting AfL by integrating assessment and instruction through mediation. Yet, DA is demanding in practice, requiring teachers to design graduated prompts, manage classroom constraints, and cultivate student engagement. To better understand these difficulties, this study explores pre-service and in-service English teachers’ perceptions of DA and their appropriation of DA principles during a praxis-oriented professional development program. Employing a mixed-methods design, the study gathered data from surveys, reflective journals, DA test designs, and transcripts of one-on-one mediation sessions. Forty-eight teachers participated (32 pre-service; 16 in-service), enabling comparisons across experience levels. Findings revealed both convergence and divergence between groups. Pre-service teachers approached DA as a psychological tool, relying on structured scales and scoring rubrics to reason about mediation design while expressing uncertainty over timing and effectiveness. In contrast, in-service teachers embodied DA in practice, employing flexible questioning and adaptive feedback, and extending DA’s application to differentiated instruction, remedial programs, and textbook design. Across groups, teachers underscored the necessity of mutual commitment, emphasizing that their students’ recognition of mediation’s value was crucial for engagement. The study captures the transformative potential of praxis-oriented PD in bridging curriculum reform and classroom practice, showing that supporting teachers involves more than equipping them with DA techniques. Implications are offered for designing DA-oriented PD that strengthens teacher agency and advances AfL in exam-driven contexts.

Enhancing active engagement in post-stroke rehabilitation is critical for promoting neuroplasticity. Although adaptive feedback can optimize arousal to improve engagement, most approaches rely solely on motor or neural indicators, overlooking the integration of task-specific physical performance with neural adaptation. The purpose of this study is to validate the effectiveness of enhancing prefrontal cortex (PFC) neural activity through a closed-loop adaptive feedback system. In this study, a neuro- and motor-feedback (NMF) system is proposed. It utilizes functional near infrared spectroscopy (fNIRS) and tracking error to continuously monitor real-time neural activity and motor performance during a visual-motor task, and realizes online adaptive regulation of task difficulty through fuzzy logic controller. 10 healthy participants were recruited for a 5-day training program, during which each participant completed 15 task trials at both fixed and adaptive difficulty levels, serving as the control group and the NMF group. Compared to the control group, the NMF group showed increased tracking errors as well as heightened neural activity in the PFC and the sensorimotor cortex (SMC), in both single-task trial and after 5 days of training. Moreover, the NMF group exhibited significantly increased strength of brain functional connections between the PFC and sensorimotor areas after training compared to the control group. Our findings suggest that the proposed NMF system can enable online neural activity regulation in visual-motor tasks and achieve enhanced integration between cognitive and sensorimotor areas, with the potential to improve the rehabilitation training outcomes.

This paper investigates Target-Attacker-Defender (TAD) differential games comprising an active target, multiple attackers, and defenders, including some anomalous agents within the defenders' team. Firstly, we introduce two types of anomalous defenders, characterized by coefficients in the performance metrics, termed “greedy” and “fearful.” Then, we explore two distinct scenarios: (i) when the attackers possess unlimited observation capabilities while visibility limitations constrain the defenders and target, and (ii) when all agents are subject to imperfect information, meaning they only possess limited visibility. The interactions among agents are modeled as nonzero-sum games to analyze their optimal decision-making processes. Due to the agents' visibility constraints, a corresponding visibility network emerges during their interactions. To address this problem, we employ inverse game theory to derive Nash equilibrium strategies with adaptive state feedback for agents. Furthermore, we analyze the impact of anomalous defenders on the victory conditions and capture time of the defenders' team. Finally, we validate the effectiveness of our results through numerical simulations.

The integration of Artificial Intelligence in education has enabled personalized and intelligent learning experiences; however, most existing AI-based platforms depend on continuous internet connectivity and cloud infrastructure, limiting their use in remote and resource-constrained regions. This project proposes an Adaptive Offline AI Learning System using a Multimodal Large Language Model (MM-LLM) architecture to overcome these limitations. The system supports multimodal inputs such as text, images, audio, and videos to enhance learner understanding and engagement. A lightweight, locally deployed MM-LLM enables offline reasoning, content generation, and adaptive feedback. Experimental results indicate improved learning efficiency, accessibility, and personalization, making the system suitable for rural and inclusive digital education environments

The AI-Based Interview Coach is an innovative and intelligent web-based platform designed to assist job seekers and students in enhancing their interview performance through real-time, personalized, and adaptive feedback. In today's highly competitive job market, candidates often lack access to effective and affordable interview preparation tools. Traditional methods such as peer mock interviews, coaching institutes, and static question banks fail to provide objective, data-driven, and immediate feedback. This project bridges that gap by leveraging state-of-the-art Artificial Intelligence technologies. The system is powered by Natural Language Processing (NLP), Sentiment Analysis, Speech Recognition, and Machine Learning algorithms. These technologies work in concert to evaluate a candidate's response across multiple dimensions including clarity, relevance, grammar, fluency, confidence, and professional tone. Users have the flexibility to either type or speak their responses, making the platform accessible to a wider audience. The AI Interview Coach generates role-specific and adaptive interview questions tailored to various job profiles such as Software Engineers, Marketing Professionals, HR Executives, Data Scientists, and more. This ensures that each practice session is highly relevant to the user's target role. Upon completing a session, users receive a comprehensive performance report with a numerical score, detailed feedback, and specific suggestions for improvement. An optional Computer Vision module, leveraging webcam input, further enhances the coaching experience by analyzing non-verbal cues such as facial expressions, eye contact, and body posture. These non-verbal aspects of communication are often critical in real-world interviews and are typically overlooked in conventional preparation methods The system consists of two primary modules: the Student Module, which handles user registration, job role selection, interview simulation, and result tracking, and the Admin Module, which manages students, job roles, interview histories, and system feedback. The application is built using the Python Flask framework for backend operations and MySQL as the database management system. The front-end is developed using HTML, CSS, Bootstrap, and JavaScript, ensuring a responsive and user-friendly interface

Abstract To address the limitations of slow convergence and susceptibility to local optima using the traditional Osprey Optimization Algorithm (OOA) in unmanned aerial vehicle (UAV) path planning, a Multi-strategy Improved OOA (MOOA) was proposed. First, a guided learning strategy based on historical population information is constructed at the global level. By utilizing the population standard deviation to establish an adaptive feedback loop for each dimension, an optimization benchmark is determined through dynamic updates, thereby enhancing the convergence efficiency. Second, a composite strategy integrating hyperbolic cosine and tangent search is employed during the exploitation phase. The nonlinear characteristics of the hyperbolic cosine function are leveraged to achieve adaptive attenuation of the search step size, ensuring a smooth transition to a fine-grained search mode, while micro-perturbations based on tangents are introduced to circumvent local stagnation. Furthermore, a high-altitude soaring strategy based on Lévy flight, coordinated with a sine factor varying over time, was incorporated to provide supplementary global search capabilities in the late convergence stage, further elevating the quality of the final solution. Ablation studies on the CEC2017 benchmark suite elucidate the individual contributions of each strategy. Comparative analysis against nine state-of-the-art algorithms confirms the superiority of the MOOA in convergence accuracy and stability, while experiments on typical engineering design problems verify its robustness in handling complex physical constraints. Finally, MOOA was applied to 3D UAV path planning scenarios of varying complexity. The simulation results demonstrate its capability to effectively minimize path costs and significantly maximize planning success rates, validating the algorithm’s effectiveness and engineering practicality in complex real-world environments.

Advancements in wearable human-machine interfaces require haptic systems that are mechanically compliant, functionally versatile, and capable of delivering rich tactile and proprioceptive feedback under varying interaction conditions. However, existing haptic devices are often constrained by bulky structures, rigid components, insufficient actuation modalities, and the absence of self-sensing capabilities, which restrict their wearability, responsiveness, robustness, and seamless integration with the human body. This work presents multimodal, self-sensing, haptic interfaces enabled by programmable soft magnetic composites and phase change materials, which provide three distinct working modes (normal, rotational shear, and skin stretch) within a compact, skin-conformal form factor. The haptic interface leverages soft magnets with programmable magnetization profiles to enable multimodal actuation. Hybrid electromagnetic coils, incorporating both solenoid and planar configurations fabricated with stretchable gallium-based conductors, are designed to enhance stretchability and actuation output. Combined with a Kirigami-patterned elastomeric spring, the haptic actuator generates forces and displacements that exceed human tactile perception thresholds while maintaining performance under deformation. Furthermore, the inductance-based self-sensing mechanism provides real-time displacement monitoring for closed-loop control, ensuring consistent performance across varying actuation and interaction conditions. Ultimately, our soft multimodal haptic device can facilitate selective stimulation of multiple cutaneous mechanoreceptors and has been demonstrated to accurately encode both limb spatial position and joint motion for comprehensive proprioceptive feedback.

This article focuses on the leader-follower output consensus (LFOC) control problem for unknown discrete-time multiagent systems (DTMASs) with sensor uncertainties under a stochastic communication protocol (SCP). An adaptive compensator is designed to eliminate the effects of sensor uncertainties, generating an adaptive estimation of the model output, which is then integrated into the feedback control law. Second, the exchange of information among agents in the multiagent system (MAS) is organized using an SCP governed by a Markov chain. At each communication instant, this protocol randomly selects a single state component of each agent for transmission. A unified Markov chain is constructed through the merging approach to represent the consensus error dynamics of the MAS. For dynamic systems with unknown parameters, a novel distributed adaptive output feedback controller is proposed to track the generated virtual reference signal and achieve LFOC under specific matching conditions. It is evident from the simulation results that the proposed controller guarantees output consensus control of the leader-follower closed-loop signals.

This paper studies the adaptive fuzzy command-filtered backstepping output feedback control for discrete-time nonlinear multi-agent systems. The command filter is used to address the causality contradiction and the error compensation mechanism can remove the filter errors. Fuzzy logic systems are utilised to approximate the unknown nonlinearities of each agent, and the fuzzy state observer is designed to estimate the immeasurable states. Adaptive updating laws are incorporated into both the observer and controller to handle unknown parameters. By constructing weighted Lyapunov functions, the stability of the closed-loop system is rigorously analyzed, proving that all signals are uniformly ultimately bounded. An example of vehicle system is provided to demonstrate the effectiveness of the proposed control strategy.

Abnormal gait associated with neuromuscular and musculoskeletal disorders represents a growing clinical burden, particularly in aging populations. This study presents a modular, low-cost Smart Rehabilitation Walker (SRW) that integrates multimodal sensing and real-time haptic feedback to enable simultaneous gait monitoring and corrective intervention in both clinical and home environments. The system combines force-sensing resistors for bilateral load symmetry assessment, inertial measurement units for fall detection, and surface electromyography (sEMG) for neuromuscular activity monitoring within a closed-loop assistive feedback architecture. A 15-day pilot study involving ten individuals with rheumatoid arthritis and clinically observed neurological gait abnormalities demonstrated measurable improvements in gait biomechanics. The Force Symmetry Index (FSI), calculated using the Robinson symmetry metric, decreased from an average of 0.9691 to 0.2019, corresponding to a 79.26% average reduction in inter-limb load asymmetry. Concurrently, sEMG measurements showed a substantial increase in neuromuscular activation (ΔEMG = 4.28), with statistical analysis confirming a significant improvement across participants (paired t-test: t(9) = 13.58, p < 0.001). To model rehabilitation trajectories, a nonlinear predictive framework based on Gaussian Process Regression achieved high predictive accuracy (R2 ≈ 0.9, with a mean RMSE of 0.0385), while providing uncertainty-aware trend estimation. Validation using an independent amyotrophic lateral sclerosis gait dataset further demonstrated the transferability of the analytical pipeline. These results highlight the potential of sensor-enabled assistive walkers as scalable platforms for quantitative gait rehabilitation, adaptive feedback, and long-term mobility monitoring.