In the field of machine learning, domain-specific annotated data is an invaluable resource for training effective models. However, in the medical domain, this data often includes Personal Health Information (PHI), raising significant privacy concerns. The stringent regulations surrounding PHI limit the availability and sharing of medical datasets, which poses a substantial challenge for researchers and practitioners aiming to develop advanced machine learning models. In this paper, we introduce a novel method to "clone" datasets containing PHI. Our approach ensures that the cloned datasets retain the essential characteristics and utility of the original data without compromising patient privacy. By leveraging differential-privacy techniques and a novel fine-tuning task, our method produces datasets that are free from identifiable information while preserving the statistical properties necessary for model training. We conduct utility testing to evaluate the performance of machine learning models trained on the cloned datasets. The results demonstrate that our cloned datasets not only uphold privacy standards but also enhance model performance compared to those trained on traditional anonymized datasets. This work offers a viable solution for the ethical and effective utilization of sensitive medical data in machine learning, facilitating progress in medical research and the development of robust predictive models.

Federated Learning (FL) is a novel machine learning approach that allows the model trainer to access more data samples, by training the model across multiple decentralized data sources, while data access constraints are in place. Such trained models can achieve significantly higher performance beyond what can be done when trained on a single data source. As part of FL's promises, none of the training data is ever transmitted to any central location, ensuring that sensitive data remains local and private. These characteristics make FL perfectly suited for large-scale applications in healthcare, where a variety of compliance constraints restrict how data may be handled, processed, and stored. Despite the apparent benefits of federated learning, the heterogeneity in the local data distributions pose significant challenges, and such challenges are even more pronounced in the case of multilingual data providers. In this paper we present a federated learning system for training a large-scale multi-lingual model suitable for fine-tuning on downstream tasks such as medical entity tagging. Our work represents one of the first such production-scale systems, capable of training across multiple highly heterogeneous data providers, and achieving levels of accuracy that could not be otherwise achieved by using central training with public data. Finally, we show that the global model performance can be further improved by a training step performed locally.

In this paper, we establish a new evolution property of Lyapunov functions that satisfy the Razumikhin Stability Theorem conditions. The derived property is shown to be a valid alternative criterion to prove the asymptotic stability of a delay system without calculating the derivative of the Lyapunov function, and it is especially useful in solving stabilization problems for time delay systems when the delay is unknown. As an example application of this derived property, we investigate the stabilization of continuous-time linear systems with an unknown time-varying delay, by proposing a switched low gain feedback control technique that utilizes the derived property. Existing literature on stabilization of continuous time delay systems all require some knowledge, such as an upper bound, of the delay. In this work, by proposing a switched low gain feedback control law utilizing the derived property, we are able to achieve asymptotic stabilization of the system without any knowledge of the delay. Furthermore, the proposed switched low gain feedback law also significantly improves the closed-loop system performance in terms of overshoot and convergence speed, in comparison with the traditional fixed gain low gain feedback design. Simulation study verifies the theoretical results.

In this note, we investigate the stabilization of discrete-time linear systems with an unknown time-varying delay. Existing literature on the stabilization of time-delay systems require some knowledge, such as an upper bound, of the delay. In this paper, by proposing a switched low-gain feedback control law, we are able to achieve asymptotic stabilization of the system without any knowledge of the delay. In addition, the proposed switched low-gain feedback law also significantly improves the closed-loop system performance in terms of overshoot and convergence speed in comparison with the traditional fixed gain design. Simulation study illustrates the effectiveness of our theoretical results.

We revisit the problem of virtual leader tracking by a group of agents of second order dynamics when only a portion of the agents are informed of the information of the virtual leader, under a state-dependent interaction topology. By proposing a simpler controller structure, we obtain results on coordinated tracking control under a jointly connected topology, both in the presence and in the absence of the virtual leader velocity information. The simplification in control protocols also enables us to arrive at our second contribution on connectivity enhancing coordinated control. In comparison with the existing connectivity preserving control algorithms, our proposed connectivity enhancing mechanism is effective in maintaining both the initially existent topology edges and those newly formed topology edges that were not initially existent. Besides, our proposed algorithms are designed in such a way that the local Lipschitz condition is satisfied and thus technical issues on solution uniqueness of the closed-loop system are avoided. Our proposed technique applies to both consensus tracking and flocking.

This paper investigates the distributed adaptive control problem for synchronization of multi-agent systems where the dynamics of the agents are nonlinear, nonidentical, unknown and subject to external disturbances. In our recent work, we solved this problem for general higher order systems under two types of communication topologies, represented, respectively, by a fixed strongly connected directed graph and by a switching connected undirected graph. A common Lyapunov function technique is employed to establish the results. In this paper, we solve the problem for a more general communication topology which is represented by a switching strongly-connected directed graph. We construct a sequence of different Lyapunov functions and use them to establish our results. Simulation study verifies the effectiveness of our theoretical results.

We revisit the problem of virtual leader tracking by a group of agents of second order dynamics when only a portion of the agents are informed of the information of the virtual leader. By proposing a simpler controller structure, we obtain results on coordinated consensus tracking control under a state dependent jointly connected topology. Such a network connectivity condition is weaker than the assumption that the interaction topology is connected all the time, which is an assumption made in most existing literature on this type of problems.

This paper revisits the distributed adaptive control problem for synchronization of multiagent systems where the dynamics of the agents are nonlinear, nonidentical, unknown, and subject to external disturbances. Two communication topologies, represented, respectively, by a fixed strongly-connected directed graph and by a switching connected undirected graph, are considered. Under both of these communication topologies, we use distributed neural networks to approximate the uncertain dynamics. Decentralized adaptive control protocols are then constructed to solve the cooperative tracker problem, the problem of synchronization of all follower agents to a leader agent. In particular, we show that, under the proposed decentralized control protocols, the synchronization errors are ultimately bounded, and their ultimate bounds can be reduced arbitrarily by choosing the control parameter appropriately. Simulation study verifies the effectiveness of our proposed protocols.

In 2005, Ren and Beard, under the network connectivity assumption that the union of the directed interaction graphs contains a spanning tree frequently enough, proposed a control protocol to solve the distributed consensus control problem for a multi-agent system with first order integrator agent dynamics. The consensus control protocol for second order systems proposed by the same authors, Ren and Beard, requires that the switching directed interaction graph has a spanning tree at every time instant to guarantee consensus. In this technical note, we propose a distributed consensus control protocol which extends the first order system results of Ren and Beard to agents represented by a general higher order controllable linear system, under the same network connectivity assumption that the union of the directed interaction graphs contains a spanning tree frequently enough. We show that the consensus can be achieved asymptotically by the group of agents under our proposed distributed control protocol. Simulation study illustrates the effectiveness of our proposed method.

This paper studies the distributed adaptive control problem for synchronization of multi-agent systems where the dynamics of the agents are nonlinear, nonidentical, unknown and subject to external disturbances. In our recent work, we solved this problem for general higher order systems under two types of communication topologies, represented, respectively, by a fixed strongly connected directed graph and by a switching connected undirected graph. The common Lyapunov function technique is employed to establish the results. In this paper, we solve the problem for a more general communication topology which is represented by a switching strongly-connected directed graph. We construct a sequence of different Lyapunov functions and use them to establish our results.

In this paper, we propose a distributed consensus control protocol which extends the first order system results of Ren and Beard (2005) to agents represented by a general higher order controllable linear system, under the same network connectivity assumption that the union of the directed interaction graphs contains a spanning tree frequently enough. We show that the consensus can be achieved asymptotically by the group of agents under our proposed distributed control protocol. Simulation study illustrates the effectiveness of our proposed method.

This paper revisits the distributed adaptive control problem for synchronization of multi-agent systems where the dynamics of the agents are nonlinear, nonidentical, unknown and subject to external disturbances. The communication topology under consideration is represented by a fixed strongly-connected directed graph. Distributed neural networks are used to approximate the uncertain dynamics and decentralized control protocols using local neighborhood information are proposed to solve the cooperative tracker problem, the problem of synchronization of all follower agents to a leader agent. In particular, we show that, under the proposed decentralized control protocols, the synchronization errors are ultimately bounded and their ultimate bounds can be reduced arbitrarily by choosing the control parameter appropriately.

Conflicts in freeway merge areas contribute significantly to freeway congestion. With advances in vehicle-to-vehicle and vehicle-to-infrastructure communications, a new concept such as the Connected Vehicle environment, where all the vehicles and infrastructure components can communicate with each other, emerges and allows for new approaches to better manage freeway merge conflicts. Given this background, a prototype advanced freeway merging control algorithm was developed in this paper. This algorithm estimates the anticipated lead and lag gap availability, and selects an appropriate strategy for controlling mainline leading and lagging vehicles and a ramp vehicle to create a sufficient gap for a smoother merge. Evaluation was conducted using VISSIM microscopic traffic simulations of an actual freeway network in Virginia. The results show that the proposed freeway merging control algorithm can significantly improve the network performance. In an ideal situation, the proposed algorithm generates a 6.3% increase in vehicles miles traveled, a 42.2% increase in average speeds, a 25.9% reduction in total travel times, and a 54.6% reduction in total delay times. In addition, fuel consumption is reduced by 12.0% and CO2 emission by 12.7%. Finally, a sensitivity analysis indicates that at least a 60% advisory compliance rate and a 50% market penetration rate are necessary to generate meaningful benefits from the proposed algorithm. All these results support that there is significant potential in improving freeway operations when a connected vehicle environment is established.

Distributed coordination control of complex multi-agent networks has received significant attention over the past decades, due to its widely recognized advantages and potentials in many applications such as large scale sensor networks, robotic networks, power grid, distributed computing clouds, social networks and biological networks.Within the control theory community, the main research task in an engineering multi-agent network system is to design distributed control algorithms, which only use local neighborhood information, to achieve some specified global objectives.