Unsupervised reinforcement learning via state entropy maximization

Reinforcement Learning (RL) provides a powerful framework to address sequential decision-making problems in which the transition dynamics is unknown or too complex to be represented. The RL approach is based on speculating what is the best decision to make given sample estimates obtained from previous interactions, a recipe that led to several breakthroughs in various domains, ranging from game playing to robotics. Despite their success, current RL methods hardly generalize from one task to another, and achieving the kind of generalization obtained through unsupervised pre-training in non-sequential problems seems unthinkable. Unsupervised RL has recently emerged as a way to improve generalization of RL methods. Just as its non-sequential counterpart, the unsupervised RL framework comprises two phases: An unsupervised pre-training phase, in which the agent interacts with the environment without external feedback, and a supervised fine-tuning phase, in which the agent aims to efficiently solve a task in the same environment by exploiting the knowledge acquired during pre-training. In this thesis, we study unsupervised RL via state entropy maximization, in which the agent makes use of the unsupervised interactions to pre-train a policy that maximizes the entropy of its induced state distribution. First, we provide a theoretical characterization of the learning problem by considering a convex RL formulation that subsumes state entropy maximization. Our analysis shows that maximizing the state entropy in finite trials is inherently harder than RL. Then, we study the state entropy maximization problem from an optimization perspective. Especially, we show that the primal formulation of the corresponding optimization problem can be (approximately) addressed through tractable linear programs. Finally, we provide the first practical methodologies for state entropy maximization in complex domains, both when the pre-training takes place in a single environment as well as multiple environments.

Thermal and air quality modeling of an electric vehicle cabin with low-cost sensors and reinforcement learning

The design process of any electric vehicle system has to be oriented towards the best energy efficiency, together with the constraint of maintaining comfort in the vehicle cabin. Main aim of this study is to research the best thermal management solution in terms of HVAC efficiency without compromising occupant’s comfort and internal air quality. An Arduino controlled Low Cost System of Sensors was developed and compared against reference instrumentation (average R-squared of 0.92) and then used to characterise the vehicle cabin in real parking and driving conditions trials. Data on the energy use of the HVAC was retrieved from the car On-Board Diagnostic port. Energy savings using recirculation can reach 30 %, but pollutants concentration in the cabin builds up in this operating mode. Moreover, the temperature profile appeared strongly nonuniform with air temperature differences up to 10° C. Optimisation methods often require a high number of runs to find the optimal configuration of the system. Fast models proved to be beneficial for these task, while CFD-1D model are usually slower despite the higher level of detail provided. In this work, the collected dataset was used to train a fast ML model of both cabin and HVAC using linear regression. Average scaled RMSE over all trials is 0.4 %, while computation time is 0.0077 ms for each second of simulated time on a laptop computer. Finally, a reinforcement learning environment was built in OpenAI and Stable-Baselines3 using the built-in Proximal Policy Optimisation algorithm to update the policy and seek for the best compromise between comfort, air quality and energy reward terms. The learning curves show an oscillating behaviour overall, with only 2 experiments behaving as expected even if too slow. This result leaves large room for improvement, ranging from the reward function engineering to the expansion of the ML model.

Directly training spiking neural networks for cyber-physical systems: from supervised to reinforcement learning

Spiking Neural Networks (SNNs) are bio-inspired Artificial Neural Networks (ANNs) utilizing discrete spiking signals, akin to neuron communication in the brain, making them ideal for real-time and energy-efficient Cyber-Physical Systems (CPSs). This thesis explores their potential in Structural Health Monitoring (SHM), leveraging low-cost MEMS accelerometers for early damage detection in motorway bridges. The study focuses on Long Short-Term SNNs (LSNNs), although their complex learning processes pose challenges. Comparing LSNNs with other ANN models and training algorithms for SHM, findings indicate LSNNs' effectiveness in damage identification, comparable to ANNs trained using traditional methods. Additionally, an optimized embedded LSNN implementation demonstrates a 54% reduction in execution time, but with longer pre-processing due to spike-based encoding. Furthermore, SNNs are applied in UAV obstacle avoidance, trained directly using a Reinforcement Learning (RL) algorithm with event-based input from a Dynamic Vision Sensor (DVS). Performance evaluation against Convolutional Neural Networks (CNNs) highlights SNNs' superior energy efficiency, showing a 6x decrease in energy consumption. The study also investigates embedded SNN implementations' latency and throughput in real-world deployments, emphasizing their potential for energy-efficient monitoring systems. This research contributes to advancing SHM and UAV obstacle avoidance through SNNs' efficient information processing and decision-making capabilities within CPS domains.

Trajectory Planning in UAV-Aided Wireless Networks Via Reinforcement Learning

This thesis is focused on the design of a flexible, dynamic and innovative telecommunication's system for future 6G applications on vehicular communications. The system is based on the development of drones acting as mobile base stations in an urban scenario to cope with the increasing traffic demand and avoid network's congestion conditions. In particular, the exploitation of Reinforcement Learning algorithms is used to let the drone learn autonomously how to behave in a scenario full of obstacles with the goal of tracking and serve the maximum number of moving vehicles, by at the same time, minimizing the energy consumed to perform its tasks. This project is an extraordinary opportunity to open the doors to a new way of applying and develop telecommunications in an urban scenario by mixing it to the rising world of the Artificial Intelligence.

Application of Deep Learning techniques in the search for BSM Higgs bosons in the $\mu\mu$ final state in CMS

The Standard Model (SM) of particle physics predicts the existence of a Higgs field responsible for the generation of particles' mass. However, some aspects of this theory remain unsolved, supposing the presence of new physics Beyond the Standard Model (BSM) with the production of new particles at a higher energy scale compared to the current experimental limits. The search for additional Higgs bosons is, in fact, predicted by theoretical extensions of the SM including the Minimal Supersymmetry Standard Model (MSSM). In the MSSM, the Higgs sector consists of two Higgs doublets, resulting in five physical Higgs particles: two charged bosons $H^{\pm}$, two neutral scalars $h$ and $H$, and one pseudoscalar $A$. The work presented in this thesis is dedicated to the search of neutral non-Standard Model Higgs bosons decaying to two muons in the model independent MSSM scenario. Proton-proton collision data recorded by the CMS experiment at the CERN LHC at a center-of-mass energy of 13 TeV are used, corresponding to an integrated luminosity of $35.9\ \text{fb}^{-1}$. Such search is sensitive to neutral Higgs bosons produced either via gluon fusion process or in association with a $\text{b}\bar{\text{b}}$ quark pair. The extensive usage of Machine and Deep Learning techniques is a fundamental element in the discrimination between signal and background simulated events. A new network structure called parameterised Neural Network (pNN) has been implemented, replacing a whole set of single neural networks trained at a specific mass hypothesis value with a single neural network able to generalise well and interpolate in the entire mass range considered. The results of the pNN signal/background discrimination are used to set a model independent 95\% confidence level expected upper limit on the production cross section times branching ratio, for a generic $\phi$ boson decaying into a muon pair in the 130 to 1000 GeV range.

Deep learning and embeddings for problems of computational biology

The development of Next Generation Sequencing promotes Biology in the Big Data era. The ever-increasing gap between proteins with known sequences and those with a complete functional annotation requires computational methods for automatic structure and functional annotation. My research has been focusing on proteins and led so far to the development of three novel tools, DeepREx, E-SNPs&GO and ISPRED-SEQ, based on Machine and Deep Learning approaches. DeepREx computes the solvent exposure of residues in a protein chain. This problem is relevant for the definition of structural constraints regarding the possible folding of the protein. DeepREx exploits Long Short-Term Memory layers to capture residue-level interactions between positions distant in the sequence, achieving state-of-the-art performances. With DeepRex, I conducted a large-scale analysis investigating the relationship between solvent exposure of a residue and its probability to be pathogenic upon mutation. E-SNPs&GO predicts the pathogenicity of a Single Residue Variation. Variations occurring on a protein sequence can have different effects, possibly leading to the onset of diseases. E-SNPs&GO exploits protein embeddings generated by two novel Protein Language Models (PLMs), as well as a new way of representing functional information coming from the Gene Ontology. The method achieves state-of-the-art performances and is extremely time-efficient when compared to traditional approaches. ISPRED-SEQ predicts the presence of Protein-Protein Interaction sites in a protein sequence. Knowing how a protein interacts with other molecules is crucial for accurate functional characterization. ISPRED-SEQ exploits a convolutional layer to parse local context after embedding the protein sequence with two novel PLMs, greatly surpassing the current state-of-the-art. All methods are published in international journals and are available as user-friendly web servers. They have been developed keeping in mind standard guidelines for FAIRness (FAIR: Findable, Accessible, Interoperable, Reusable) and are integrated into the public collection of tools provided by ELIXIR, the European infrastructure for Bioinformatics.

Applying machine learning: a multi-role perspective

Machine (and deep) learning technologies are more and more present in several fields. It is undeniable that many aspects of our society are empowered by such technologies: web searches, content filtering on social networks, recommendations on e-commerce websites, mobile applications, etc., in addition to academic research. Moreover, mobile devices and internet sites, e.g., social networks, support the collection and sharing of information in real time. The pervasive deployment of the aforementioned technological instruments, both hardware and software, has led to the production of huge amounts of data. Such data has become more and more unmanageable, posing challenges to conventional computing platforms, and paving the way to the development and widespread use of the machine and deep learning. Nevertheless, machine learning is not only a technology. Given a task, machine learning is a way of proceeding (a way of thinking), and as such can be approached from different perspectives (points of view). This, in particular, will be the focus of this research. The entire work concentrates on machine learning, starting from different sources of data, e.g., signals and images, applied to different domains, e.g., Sport Science and Social History, and analyzed from different perspectives: from a non-data scientist point of view through tools and platforms; setting a problem stage from scratch; implementing an effective application for classification tasks; improving user interface experience through Data Visualization and eXtended Reality. In essence, not only in a quantitative task, not only in a scientific environment, and not only from a data-scientist perspective, machine (and deep) learning can do the difference.

Strategic bidding methodology for electricity markets using adaptive learning

The very particular characteristics of electricity markets, require deep studies of the interactions between the involved players. MASCEM is a market simulator developed to allow studying electricity market negotiations. This paper presents a new proposal for the definition of MASCEM players’ strategies to negotiate in the market. The proposed methodology is implemented as a multiagent system, using reinforcement learning algorithms to provide players with the capabilities to perceive the changes in the environment, while adapting their bids formulation according to their needs, using a set of different techniques that are at their disposal. This paper also presents a methodology to define players’ models based on the historic of their past actions, interpreting how their choices are affected by past experience, and competition.

Robotic Sensing and Manipulation of Deformable Linear Objects with Learning-based methods

Nowadays robotic applications are widespread and most of the manipulation tasks are efficiently solved. However, Deformable-Objects (DOs) still represent a huge limitation for robots. The main difficulty in DOs manipulation is dealing with the shape and dynamics uncertainties, which prevents the use of model-based approaches (since they are excessively computationally complex) and makes sensory data difficult to interpret. This thesis reports the research activities aimed to address some applications in robotic manipulation and sensing of Deformable-Linear-Objects (DLOs), with particular focus to electric wires. In all the works, a significant effort was made in the study of an effective strategy for analyzing sensory signals with various machine learning algorithms. In the former part of the document, the main focus concerns the wire terminals, i.e. detection, grasping, and insertion. First, a pipeline that integrates vision and tactile sensing is developed, then further improvements are proposed for each module. A novel procedure is proposed to gather and label massive amounts of training images for object detection with minimal human intervention. Together with this strategy, we extend a generic object detector based on Convolutional-Neural-Networks for orientation prediction. The insertion task is also extended by developing a closed-loop control capable to guide the insertion of a longer and curved segment of wire through a hole, where the contact forces are estimated by means of a Recurrent-Neural-Network. In the latter part of the thesis, the interest shifts to the DLO shape. Robotic reshaping of a DLO is addressed by means of a sequence of pick-and-place primitives, while a decision making process driven by visual data learns the optimal grasping locations exploiting Deep Q-learning and finds the best releasing point. The success of the solution leverages on a reliable interpretation of the DLO shape. For this reason, further developments are made on the visual segmentation.

Robotic Manipulation of DLOs for Wiring Harness Assembly: a Machine Learning Approach

In recent times, a significant research effort has been focused on how deformable linear objects (DLOs) can be manipulated for real world applications such as assembly of wiring harnesses for the automotive and aerospace sector. This represents an open topic because of the difficulties in modelling accurately the behaviour of these objects and simulate a task involving their manipulation, considering a variety of different scenarios. These problems have led to the development of data-driven techniques in which machine learning techniques are exploited to obtain reliable solutions. However, this approach makes the solution difficult to be extended, since the learning must be replicated almost from scratch as the scenario changes. It follows that some model-based methodology must be introduced to generalize the results and reduce the training effort accordingly. The objective of this thesis is to develop a solution for the DLOs manipulation to assemble a wiring harness for the automotive sector based on adaptation of a base trajectory set by means of reinforcement learning methods. The idea is to create a trajectory planning software capable of solving the proposed task, reducing where possible the learning time, which is done in real time, but at the same time presenting suitable performance and reliability. The solution has been implemented on a collaborative 7-DOFs Panda robot at the Laboratory of Automation and Robotics of the University of Bologna. Experimental results are reported showing how the robot is capable of optimizing the manipulation of the DLOs gaining experience along the task repetition, but showing at the same time a high success rate from the very beginning of the learning phase.

Adaptive learning in agents behaviour: A framework for electricity markets simulation

Electricity markets are complex environments, involving a large number of different entities, playing in a dynamic scene to obtain the best advantages and profits. MASCEM (Multi-Agent System for Competitive Electricity Markets) is a multi-agent electricity market simulator that models market players and simulates their operation in the market. Market players are entities with specific characteristics and objectives, making their decisions and interacting with other players. This paper presents a methodology to provide decision support to electricity market negotiating players. This model allows integrating different strategic approaches for electricity market negotiations, and choosing the most appropriate one at each time, for each different negotiation context. This methodology is integrated in ALBidS (Adaptive Learning strategic Bidding System) – a multiagent system that provides decision support to MASCEM's negotiating agents so that they can properly achieve their goals. ALBidS uses artificial intelligence methodologies and data analysis algorithms to provide effective adaptive learning capabilities to such negotiating entities. The main contribution is provided by a methodology that combines several distinct strategies to build actions proposals, so that the best can be chosen at each time, depending on the context and simulation circumstances. The choosing process includes reinforcement learning algorithms, a mechanism for negotiating contexts analysis, a mechanism for the management of the efficiency/effectiveness balance of the system, and a mechanism for competitor players' profiles definition.

Semi-online neural-Q_leaming for real-time robot learning

Reinforcement learning (RL) is a very suitable technique for robot learning, as it can learn in unknown environments and in real-time computation. The main difficulties in adapting classic RL algorithms to robotic systems are the generalization problem and the correct observation of the Markovian state. This paper attempts to solve the generalization problem by proposing the semi-online neural-Q_learning algorithm (SONQL). The algorithm uses the classic Q_learning technique with two modifications. First, a neural network (NN) approximates the Q_function allowing the use of continuous states and actions. Second, a database of the most representative learning samples accelerates and stabilizes the convergence. The term semi-online is referred to the fact that the algorithm uses the current but also past learning samples. However, the algorithm is able to learn in real-time while the robot is interacting with the environment. The paper shows simulated results with the "mountain-car" benchmark and, also, real results with an underwater robot in a target following behavior

Semi-online neural-Q_leaming for real-time robot learning

Reinforcement learning (RL) is a very suitable technique for robot learning, as it can learn in unknown environments and in real-time computation. The main difficulties in adapting classic RL algorithms to robotic systems are the generalization problem and the correct observation of the Markovian state. This paper attempts to solve the generalization problem by proposing the semi-online neural-Q_learning algorithm (SONQL). The algorithm uses the classic Q_learning technique with two modifications. First, a neural network (NN) approximates the Q_function allowing the use of continuous states and actions. Second, a database of the most representative learning samples accelerates and stabilizes the convergence. The term semi-online is referred to the fact that the algorithm uses the current but also past learning samples. However, the algorithm is able to learn in real-time while the robot is interacting with the environment. The paper shows simulated results with the "mountain-car" benchmark and, also, real results with an underwater robot in a target following behavior

Linear Bayes policy for learning in contextual-bandits

Machine and Statistical Learning techniques are used in almost all online advertisement systems. The problem of discovering which content is more demanded (e.g. receive more clicks) can be modeled as a multi-armed bandit problem. Contextual bandits (i.e., bandits with covariates, side information or associative reinforcement learning) associate, to each specific content, several features that define the “context” in which it appears (e.g. user, web page, time, region). This problem can be studied in the stochastic/statistical setting by means of the conditional probability paradigm using the Bayes’ theorem. However, for very large contextual information and/or real-time constraints, the exact calculation of the Bayes’ rule is computationally infeasible. In this article, we present a method that is able to handle large contextual information for learning in contextual-bandits problems. This method was tested in the Challenge on Yahoo! dataset at ICML2012’s Workshop “new Challenges for Exploration & Exploitation 3”, obtaining the second place. Its basic exploration policy is deterministic in the sense that for the same input data (as a time-series) the same results are obtained. We address the deterministic exploration vs. exploitation issue, explaining the way in which the proposed method deterministically finds an effective dynamic trade-off based solely in the input-data, in contrast to other methods that use a random number generator.

Learning processes and the neural analysis of conditioning

Classical and operant conditioning principles, such as the behavioral discrepancy-derived assumption that reinforcement always selects antecedent stimulus and response relations, have been studied at the neural level, mainly by observing the strengthening of neuronal responses or synaptic connections. A review of the literature on the neural basis of behavior provided extensive scientific data that indicate a synthesis between the two conditioning processes based mainly on stimulus control in learning tasks. The resulting analysis revealed the following aspects. Dopamine acts as a behavioral discrepancy signal in the midbrain pathway of positive reinforcement, leading toward the nucleus accumbens. Dopamine modulates both types of conditioning in the Aplysia mollusk and in mammals. In vivo and in vitro mollusk preparations show convergence of both types of conditioning in the same motor neuron. Frontal cortical neurons are involved in behavioral discrimination in reversal and extinction procedures, and these neurons preferentially deliver glutamate through conditioned stimulus or discriminative stimulus pathways. Discriminative neural responses can reliably precede operant movements and can also be common to stimuli that share complex symbolic relations. The present article discusses convergent and divergent points between conditioning paradigms at the neural level of analysis to advance our knowledge on reinforcement.