560 research outputs found
Decentralised learning for Intelligent Control Systems
No full textThis manuscript represents a collection of the most important research activities
carried out by the candidate during his three years of PhD studies.
This work leverages on the concept of Intelligent Control Systems, defined as
a framework where control methods attempt to emulate important characteristics
of human intelligence to generate control actions. Being referred to as a point of
contact between the scientific fields of control theory and artificial intelligence, they
aim to combine the mathematical rigor of the former with the representativeness of
the latter in order to exploit the potential of both of them.
While classical control theory model-based approaches commonly used to examine
the characteristics of a given system in terms of its stability, safety and optimality,
may fail to include environmental uncertainties and are subject to modelling errors,
data-driven controller design techniques aim to capture such stochasticities and
nonlinearities. This idea is behind the development of the first work which develops
neural-based control solution which envisages the use of deep neural networks within
the model predictive control framework with the aim to derive the optimal control law
in a distributed fashion by means of a cascading combination of one-step predictors.
The second work focuses on the learning processes of data-driven methodologies,
with particular attention to neural networks, whose approximation capabilities make
them of one of the most important tools in the approximation of system dynamics.
The research activity, carried out by the candidate in the scope of the POR FESR
FedMedAI project, develops a decentralised framework based on consensus-theory
aimed at allowing the training of a neural network over decentralised scenarios,
namely on data belonging to multiple actors who communicate with each other and
collaborate for the learning of a data-driven model aimed at solving an approximation
task. The specifications of this framework were defined during the course of the
project through interactions with the Italian Istituto Superiore di Sanità, allowing
the realization of a platform aimed at enhancing a privacy-preserving collaboration
among clinical institutions, without any exchange of clinical data.
The investigation of the mechanisms underpinning the interaction between different
actors is examined within the third work in the context of multi-agent systems.
Since communication is one of the tools used by agents to collaborate, a learningbased
strategy allowing agents to limit their communication while still achieving
their objective is proposed leveraging on the multi-agent reinforcement learning framework. The proposed approach allows to cope with real-world scenarios where
communication-related costs cannot be neglected.
The fourth work discusses multi-agent scenarios where each agent attempts to
accomplish its own objective independently of other agents’ cooperation. Numerous
settings find use for these non-cooperative scenarios, one of which being telecommunications.
In this context, the convergence properties of a class of load-balancing
strategies towards a set of approximate non-cooperative equilibria are examined. The
candidate also explores non-cooperative approaches in the domains of mobile edge
computing and automotive, whereby decentralised policy broadcasting mechanisms
and decision-making processes based on reinforcement learning are proposed.
All the studies incorporated into this work addresses various issues and challenges
that may arise when intelligent control systems are employed in multi-agent context.
In particular, control systems of this type find application in the control of complex
systems, such as health-related ones, in which the interaction with the human being
constitutes the most critical aspect. With respect to this issue, the high-level architecture
of the PON CADUCEO, POR FESR FedMedAI and Allenamente project is
described.
Every study under consideration is predicated on the use of various control theory
arguments and data-driven approaches, whose choice and combination is justified
and validated over different scenarios
Model Predictive Control for Collision-free Spacecraft Formation with Artificial Potential Functions
No full textA collision-free formation control strategy for flying in formation is presented. A linear control law is developed by means of Model Predictive Control (MPC) via the dual-mode paradigm [1]. Collision avoidance is dealt with by using Artificial Potential Functions (APFs) to keep a desired safe distance from the obstacles. The main innovation in the proposed approach is that each spacecraft independently performs the collision avoidance manoeuvres and, as a consequence, the APFs-based collision avoidance control is in charge also of the collision avoidance between two spacecraft. The optimality of the solution is discussed and numerical simulations show the effectiveness of the proposed method
Distributed MARL with Limited Sensing for Robot Navigation Problems
No full textThis paper proposes a Multi-Agent Reinforcement Learning (MARL) algorithm for the multi-robot navigation problem. Most of the proposals in the literature requires some form of information sharing and communications among agents to coordinate their action in order to complete the overall task. The proposed paper, named Limited Sensing MARL (LS-MARL), assumes that each robot decisions rely on local information and is provided with sensor, which can be switched on for the localization of the robots within a given range. Besides the navigation task, each agent aims at limiting the use of the sensor as much as possible (i.e., to be as independent as possible) for energy saving or safety reasons. The algorithm is evaluated by simulations and favourably compares to the one proposed in (Yu et al. (2015)), that assumes a similar setup in which the neighbouring agents share their positioning information
Deep Deterministic Policy Gradient Control of Type 1 Diabetes
No full textType 1 diabetes is one of the major concerns in current medical studies. Traditional clinical practice involves non-autonomous manual injection of insulin in the blood, while current research in the field of autonomous regulation of blood glucose concentration mostly focuses on model-based control techniques. This paper introduces a novel Reinforcement Learning-based controller for autonomous glycemic regulation in the treatment of type 1 diabetes, building on the Deep Deterministic Policy Gradient algorithm. The proposed control method is validated through in-vitro simulations on the Bergman glucoregulatory model, proving that it successfully preserves healthy values of blood glucose concentration, while overcoming both standard clinical practice and classical model-based control techniques in terms of both control effort and computational efficiency for real-time applications
AdaFed: Performance-based Adaptive Federated Learning
No full textFederated Learning is a distributed and privacy-preserving machine learning technique that allows local clients to learn a model without sharing their own data by coordinating with a global server. In this work, we present the Adaptive Federated Learning (AdaFed) algorithm, which aims at improving the training performance of deep neural networks in Federated Learning settings by: (i) dynamically weighting the local models in the model averaging procedure; (ii) by adapting the loss function used by the federation at every communication round. We discuss the specialisation of AdaFed for both classification and regression tasks, providing several validation examples. Due to its adaptive design, the AdaFed algorithm showed a robust behaviour against unbalanced data distributions and adversarial clients
Hierarchical Federated Learning for Edge Intelligence through Average Consensus
No full textConsensus of multi-agent systems has recently been studied in the context of Federated Learning (FL), an emerging branch of distributed machine learning. The present paper proposes a two-level hierarchical algorithm for FL in the context of edge computing, developing a fully decentralized solution that relies on results obtained for discrete-time consensus of dynamical systems. The proposed architecture and algorithm are validated on a test case and compared to current solutions, which require a centralized server
Which kind of philosopher was Danilo Pejović?
No full textU ovom kratkom prilogu autorica pokušava opisati narav Danila Pejovića kao filozofa. Njegovo bitno obilježje bila je filozofijska i svetovna suverenost.In this short contribution the author tries to describe the nature of Danilo Pejović as a philosopher. His main characteristic was a philosophical and secular sovereignity
Deep Reinforcement Learning Control of Type-1 Diabetes with Cross-Patient Generalization
No full textType 1 diabetes is one of the major concerns in current medical studies, as the World Health Organisation plans to reduce mortality due to such disease by one third by 2030. Standard clinical practice involves self-administered injections of insulin, while current research in the field of automatic control of blood glucose concentration mostly focuses on model-based control techniques. This work presents an application of a Deep Reinforcement Learning-based controller for autonomous treatment of type 1 diabetes, building on the Deep Determin-istic Policy Gradient algorithm. Such control framework is applied for the first time on the Python implementation of the UVA/Padova simulator, named Simglucose. The proposed methodology is validated through in-vitro simulations on an inter-cluster cross-generalization group of virtual adult patients, showing that normoglycemia is successfully preserved while assuring cross-patient generalization and outperforming clinical practice, without the direct knowledge of the amount of ingested carbohydrates
Deep Reinforcement Learning Platooning Control of Non-Cooperative Autonomous Vehicles in a Mixed Traffic Environment
No full textEnsuring secure spacing between vehicles is vital for road safety, efficient traffic flow, and system stability in autonomous driving. While traditional cooperative platooning approach, relying on centralized coordination exploiting wireless network, faces practical implementation challenges due to communication constraints and diverse driving behaviors, this work introduces a scalable non-cooperative multi-agent platooning strategy based on Deep Reinforcement Learning, leveraging on decentralized decision-making principles. The agents’ aim is to adjust their velocities dynamically to ensure safe following distances and adapt to surrounding vehicle behavior, without the possibility of exchanging information over a wireless network. Extensive simulations validate the effectiveness and robustness of the proposed approach, making it suitable for real-world autonomous driving scenarios
An Integrated Music and Artificial Intelligence System in Support of Pediatric Neurorehabilitation
No full textThis study aims at the implementation of an Artificial Intelligence approach to the use of music for supporting the neurorehabilitation of children with brain injuries or neurological difficulties.The output of this study will be the implementation of an app for mobile devices with games to be played by pediatric patients, allowing time for their cognitive and motor abilities to recover while enjoying pleasant activities. In particular, a Neural Network Classification approach is proposed in order to automatically adapt the game difficulty to the current cognitive capabilities of the child
- …
