With the rapid increase of Internet of Things-devices(IoT), demand for new machine learning algorithms and modelshas risen. The focus of this project is implementing a federatedlearning (FL) algorithm to detect anomalies in measurementsmade by a water monitoring IoT-sensor. The FL algorithm trainsacross a collection of decentralized IoT-devices, each using thelocal data acquired from the specific sensor. The local machinelearning models are then uploaded to a mutual server andaggregated into a global model. The global model is sent back tothe sensors and is used as a template when training starts againlocally. In this project, we only have had access to one physicalsensor. This has forced us to virtually simulate sensors. Thesimulation was done by splitting the data gathered by the onlyexisting sensor. To deal with the long, sequential data gatheredby the sensor, a long short-term memory (LSTM) network wasused. This is a special type of artificial neural network (ANN)capable of learning long-term dependencies. After analyzing theobtained results it became clear that FL has the potential toproduce good results, provided that more physical sensors aredeployed. / I samband med den snabba ökningen avInternet of Things-enheter (IoT) har efterfrågan på nya algoritmeroch modeller för maskininlärning ökat. Detta projektfokuserar på att implementera en federated learning (FL) algoritmför att detektera avvikelser i mätdata från en sensorsom övervakar vattenkvaliteten. FL algoritmen tränar en samlingdecentraliserade IoT-enheter, var och en med hjälp av lokaldata från sensorn i fråga. De lokala maskininlärningsmodellernaladdas upp till en gemensam server och sammanställs till englobal modell. Den globala modellen skickas sedan tillbaka tillsensorerna och används som mall när den lokala träningen börjarigen. I det här projektet hade vi endast tillgång till en fysisksensor. Vi har därför varit tvungna att simulera sensorer. Dettagjordes genom att dela upp datamängden som samlats in frånden fysiska sensorn. För att hantera den långa sekventiella dataanvänds ett long short-term memory (LSTM) nätverk. Detta ären speciell typ av artificiellt neuronnät (ANN) som är kapabeltatt minnas mönster under en längre tid. Efter att ha analyseratresultaten blev det tydligt att FL har potentialen att produceragoda resultat, givet att fler fysiska sensorer implementeras. / Kandidatexjobb i elektroteknik 2021, KTH, Stockholm

Recent years have witnessed remarkable advancements in machine learning, but with these advances come concerns about data privacy. Machine learning inherently involves learning functions from data, and this process can potentially lead to information leakage through various attacks on the learned model. Additionally, the presence of malicious actors who may poison input data to manipulate the model has become a growing concern. Consequently, the ability to unlearn specific data samples on demand has become critically important. Federated Learning (FL) has emerged as a powerful approach to address these challenges. In FL, multiple participants or clients collaborate to train a single global machine learning model without sharing their training data. However, the issue of machine unlearning is particularly pertinent in FL, especially in scenarios where clients are not fully trustworthy. This paper delves into the investigation of the efficacy of solving machine unlearning problems within the FL framework. The central research question this work tackles is: How can we effectively unlearn the entire dataset from one or multiple clients once an FL training is completed, while maintaining privacy and without access to the data? To address this challenge, we introduce the concept of ”contribution,” which quantifies how much each client contributes to the training of the global FL model. In our implementation, we employ an Encoder-Decoder model on the server’s end to disentangle these contributions as the FL process progresses. Notably, our approach is unique in that there is no existing work that utilizes a similar concept nor similar models. Our findings, supported by extensive experiments on datasets MNIST and FashionMNIST, demonstrate that our proposed approach successfully solves the unlearning task in FL. Remarkably, it achieves results comparable to retraining from scratch without requiring the participation of the specific client whose data needs to be unlearned. Moreover, additional ablation studies indicate the sensitivity of the proposed model to specific structural hyperparameters. / Här har de senaste åren bevittnat enastående framsteg inom maskininlärning, men med dessa framsteg kommer bekymmer om dataskydd. Maskininlärning innebär i grunden att lära sig funktioner från data, och denna process kan potentiellt leda till läckage av information genom olika attacker mot den inlärda modellen. Dessutom har närvaron av illvilliga aktörer som kan förgifta indata för att manipulera modellen blivit en växande oro. Följaktligen har förmågan att avlära specifika datasatser på begäran blivit av avgörande betydelse. Federerad inlärning (FL) har framträtt som en kraftfull metod för att ta itu med dessa utmaningar. I FL samarbetar flera deltagare eller klienter för att träna en enda global maskininlärningsmodell utan att dela sina träningsdata. Emellertid är problemet med maskinavlärande särskilt relevant inom FL, särskilt i situationer där klienterna inte är fullt pålitliga. Denna artikel fördjupar sig i undersökningen av effektiviteten av att lösa problem med maskinavlärande inom FL-ramverket. Den centrala forskningsfråga som detta arbete behandlar är: Hur kan vi effektivt avlära hela datasamlingen från en eller flera klienter när FL-utbildningen är klar, samtidigt som vi bevarar integritet och inte har tillgång till datan? För att ta itu med denna utmaning introducerar vi begreppet ”bidrag,” som kvantifierar hur mycket varje klient bidrar till träningen av den globala FLmodellen. I vår implementering använder vi en Encoder-Decoder-modell på serverns sida för att reda ut dessa bidrag när FL-processen fortskrider. Det är värt att notera att vår metod är unik eftersom det inte finns något befintligt arbete som använder ett liknande koncept eller liknande modeller. Våra resultat, som stöds av omfattande experiment på dataseten MNIST och FashionMNIST, visar att vår föreslagna metod framgångsrikt löser avlärandeuppgiften i FL. Anmärkningsvärt uppnår den resultat som är jämförbara med att träna om från grunden utan att kräva deltagandet av den specifika klient vars data behöver avläras. Dessutom indikerar ytterligare avläggningsstudier känsligheten hos den föreslagna modellen för specifika strukturella hyperparametrar.

<p dir="ltr">The original definition of Nash Equilibrium applied to normal form games, but the notion has now been extended to various other forms of games including leader-follower games (Stackelberg games), extensive form games, stochastic games, games of incomplete information, cooperative games, and so on. We focus on general-sum stochastic Stackelberg games in this work. An example where such games would be natural to consider is in security games where a defender wishes to protect some targets through deployment of limited resources and an attacker wishes to strategically attack the targets to benefit themselves. The hierarchical order of play arises naturally since the defender typically acts first and deploys a strategy, while the attacker observes the strategy ofthe defender before attacking. Another example where this framework fits is in testing during epidemics, where the leader (the government) sets testing policies and the follower (the citizens) decide at every time step whether to get tested. The government wishes to minimize the number of infected people in the population while the follower wishes to minimize the cost of getting sick and testing. This thesis presents a learning algorithm for players to converge to their stationary policies in a general sum stochastic sequential Stackelberg game. The algorithm is a two time scale implicit policy gradient algorithm that provably converges to stationary points of the optimization problems of the two players. Our analysis allows us to move beyond the assumptions of zero-sum or static Stackelberg games made in the existing literature for learning algorithms to converge.</p><p dir="ltr"><br></p>

The objective of this thesis was to investigate how the addition of a privacy-preserving mechanism to a federated learning model was affecting the performance of the model for an image classification task. Further, it was to get knowledge on how the outlook to use federated learning in the biotech industry is and what possible threats and attacks that could obstruct the utilization of federated learning among competitors. In the project four different aggregation algorithms for federated learning were examined. The methods were weighted fedAvg, unweighted FedAvg, weighted FedProx and unweighted FedProx. The experiment was using tensorflow federated to simulate the different methods. They were evaluated using accuracy, loss, recall, precision and F1 score. The result of this study shows that the performance of the deep neural network model is decreasing as differential privacy is introduced to the process. Out of the four aggregation algorithms used, weighted fedProx was the one that performed the best despite the added noise. It was also concluded that federated learning has potential to be used in the biotechnology industry among competitors, but that there are still security threats and attacks to avoid.

While it is difficult for base stations to estimate the millimeter wave (mmWave) channels and find the optimal mmWave beam for user equipments (UEs) quickly, the sub-6 GHz channels which are usually easier to obtain and more robust to blockages could be used to reduce the time before initial access and enhance the reliability of mmWave communication. Considering that the channel information is collected by a massive number of radio base stations and would be sensitive to privacy and security, Federated Learning (FL) is a match for this use case. In practice, the channel vectors are usually subject to Non-Independently Distributed (non-IID) distributions due to the greatly varying wireless communication environments between different radio base stations and their UEs. To achieve satisfying performance for all radio base stations instead of only the majority of them, a useful solution is designing personalized methods for each radio base station. In this thesis, we implement two personalized FL methods including 1) Finetuning FL Model on Private Dataset of Each Client and 2) Adaptive Expert Models for FL to predict the optimal mmWave beamforming vector directly from the non-IID sub-6 GHz channel vectors generated from DeepMIMO. According to our experimental results, Finetuning FL Model on Private Dataset of Each Client achieves higher average mmWave downlink spectral efficiency than the global FL. Besides, in terms of the average Top-1 and Top-3 classification accuracies, its performance improvement over the global FL model even exceeds the improvement of the global FL over the pure local models. / Även om det är svårt för en basstation att uppskatta en kanal för millimetervåg (mmWave) och snabbt hitta den bästa mmWave-strålen för en användarutrustning (UE), kan den dra fördel av kanaler under 6 GHz, som i allmänhet är mer lättillgängliga och mer motståndskraftig mot blockering, för att minska tid för första besök och förbättra tillförlitligheten hos mmWave-kommunikation. Med tanke på att kanalinformation samlas in av ett stort antal radiobasstationer och är känslig för integritet och säkerhet är federated learning (FL) väl lämpat för detta användningsfall. I praktiken, eftersom den trådlösa kommunikationsmiljön varierar mycket mellan olika radiobasstationer och deras UE, följer kanalvektorer vanligtvis en icke-oberoende distribution (icke-IID). För att uppnå tillfredsställande prestanda för alla radiobasstationer, inte bara de flesta radiobasstationer, är en användbar lösning att utforma ett individuellt tillvägagångssätt för varje radiobasstation. I detta dokument implementerar vi två personliga FL-metoder, inklusive 1) finjustering av FL-modellen på varje klients privata datauppsättning och 2) en adaptiv expertmodell av FL för att direkt generera icke-IID sub-6 GHz kanalvektorer förutsäga optimal mmWave beamforming vektorer. Enligt våra experimentella resultat uppnår finjustering av FL-modellen på varje klients privata datauppsättning högre genomsnittlig mmWave-nedlänksspektral effektivitet än global FL. Dessutom överträffar dess prestandaförbättring jämfört med den globala FL-modellen till och med den för den globala FL jämfört med den rent lokala modellen vad gäller genomsnittlig klassificeringsnoggrannhet i topp-1 och topp-3.

Dans l’apprentissage fédéré, un modèle global est appris en agrégeant les mises à jour du modèle calculées à partir d’un ensemble de nœuds clients, un défi clé dans ce domaine est l’hétérogénéité des données entre les clients qui dégrade les performances du modèle. Les algorithmes d’apprentissage fédéré standard effectuent plusieurs étapes de gradient avant de synchroniser le modèle, ce qui peut amener les clients à minimiser exagérément leur propre objectif local et à s’écarter de la solution globale. Nous démontrons que dans un tel contexte, les modèles de clients individuels subissent un oubli catastrophique par rapport aux données d’autres clients et nous proposons une approche simple mais efficace qui modifie l’objectif d’entropie croisée sur une base par client en repondérant le softmax de les logits avant de calculer la perte. Cette approche protège les classes en dehors de l’ensemble d’étiquettes d’un client d’un changement de représentation brutal. Grâce à une évaluation empirique approfondie, nous démontrons que notre approche peut atténuer ce problème, en apportant une amélioration continue aux algorithmes d’apprentissage fédéré standard. Cette approche est particulièrement avantageux dans les contextes d’apprentissage fédéré difficiles les plus étroitement alignés sur les scénarios du monde réel où l’hétérogénéité des données est élevée et la participation des clients à chaque cycle est faible. Nous étudions également les effets de l’utilisation de la normalisation par lots et de la normalisation de groupe avec notre méthode et constatons que la normalisation par lots, qui était auparavant considérée comme préjudiciable à l’apprentissage fédéré, fonctionne exceptionnellement bien avec notre softmax repondéré, remettant en question certaines hypothèses antérieures sur la normalisation dans un système fédéré / In Federated Learning, a global model is learned by aggregating model updates computed from a set of client nodes, a key challenge in this domain is data heterogeneity across clients which degrades model performance. Standard federated learning algorithms perform multiple gradient steps before synchronizing the model which can lead to clients overly minimizing their own local objective and diverging from the global solution. We demonstrate that in such a setting, individual client models experience a catastrophic forgetting with respect to data from other clients and we propose a simple yet efficient approach that modifies the cross-entropy objective on a per-client basis by re-weighting the softmax of the logits prior to computing the loss. This approach shields classes outside a client’s label set from abrupt representation change. Through extensive empirical evaluation, we demonstrate our approach can alleviate this problem, providing consistent improvement to standard federated learning algorithms. It is particularly beneficial under the challenging federated learning settings most closely aligned with real world scenarios where data heterogeneity is high and client participation in each round is low. We also investigate the effects of using batch normalization and group normalization with our method and find that batch normalization which has previously been considered detrimental to federated learning performs particularly well with our re-weighted softmax, calling into question some prior assumptions about normalization in a federated setting

Machine Learning (ML) is the current trend that most industries opt for to improve their business and operations. ML has also been adopted in the financial markets, where well-funded financial institutions employ the latest ML algorithms to gain an advantage on the market. The darker side of ML is the potential emergence of complex algorithmic trading schemes that are abusive and manipulative. Because of this, it is inevitable that ML will be applied to financial market surveillance in order to detect these abusive and manipulative trading strategies. Ideally, an accurate ML detection model would be developed with data from many financial institutions or trading venues. However, such ML models require vast quantities of data, which poses a problem in market surveillance where data is sensitive or limited. Data sharing between companies or countries is typically accompanied by legal and privacy concerns. By training ML models on distributed datasets, Federated Learning (FL) overcomes these issues by eliminating the need to centralise sensitive data. This thesis aimed to address these ML related issues in market surveillance by implementing and evaluating a FL model. FL enables a group of independent data-holding clients with the same intention to build a shared ML model collaboratively without compromising private data. In this work, a ML model is initially deployed in a centralised data setting and trained to detect the manipulative trading scheme known as spoofing. The LSTM-Autoencoder was the model chosen method for this task. The same model is also implemented in a federated setting but with decentralised data, using the FL framework FEDn. Another FL framework, Flower, is also employed to evaluate the performance of FEDn. Experiments were conducted comparing the FL models to the conventional centralised learning model, as well as comparing the two frameworks to each other. The results showed that under certain circumstances, the FL models performed better than the centralised model in detecting spoofing. FEDn was equivalent to Flower in terms of detection performance. In addition, the results indicated that Flower was marginally faster than FEDn. It is assumed that variations in the experimental setup and stochasticity account for the performance disparity.

Federated learning poses a statistical challenge when training on highly heterogeneous sequence data. For example, time-series telecom data collected over long intervals regularly shows mixed fluctuations and patterns. These distinct distributions are an inconvenience when a node not only plans to contribute to the creation of the global model but also plans to apply it on its local dataset. In this scenario, adopting a one-fits-all approach might be inadequate, even when using state-of-the-art machine learning techniques for time series forecasting, such as Long Short-Term Memory (LSTM) networks, which have proven to be able to capture many idiosyncrasies and generalise to new patterns. In this work, we show that by clustering the clients using these patterns and selectively aggregating their updates in different global models can improve local performance with minimal overhead, as we demonstrate through experiments using realworld time series datasets and a basic LSTM model. / Federated Learning utgör en statistisk utmaning vid träning med starkt heterogen sekvensdata. Till exempel så uppvisar tidsseriedata inom telekomdomänen blandade variationer och mönster över längre tidsintervall. Dessa distinkta fördelningar utgör en utmaning när en nod inte bara ska bidra till skapandet av en global modell utan även ämnar applicera denna modell på sin lokala datamängd. Att i detta scenario införa en global modell som ska passa alla kan visa sig vara otillräckligt, även om vi använder oss av de mest framgångsrika modellerna inom maskininlärning för tidsserieprognoser, Long Short-Term Memory (LSTM) nätverk, vilka visat sig kunna fånga komplexa mönster och generalisera väl till nya mönster. I detta arbete visar vi att genom att klustra klienterna med hjälp av dessa mönster och selektivt aggregera deras uppdateringar i olika globala modeller kan vi uppnå förbättringar av den lokal prestandan med minimala kostnader, vilket vi demonstrerar genom experiment med riktigt tidsseriedata och en grundläggande LSTM-modell.

Time Series data has become ubiquitous thanks to affordable edge devices and sensors. Much of this data is valuable for decision making. In order to use these data for the forecasting task, the conventional centralized approach has shown deficiencies regarding large data communication and data privacy issues. Furthermore, Neural Network models cannot make use of the extra information from the time series, thus they usually fail to provide time series specific results. Both issues expose a challenge to large-scale Time Series Forecasting with Neural Network models. All these limitations lead to our research question:Can we realize decentralized time series forecasting with a Federated Learning mechanism that is comparable to the conventional centralized setup in forecasting performance?In this work, we propose a Federated Series Forecasting framework, resolving the challenge by allowing users to keep the data locally, and learns a shared model by aggregating locally computed updates. Besides, we design a hybrid model to enable Neural Network models utilizing the extra information from the time series to achieve a time series specific learning. In particular, the proposed hybrid outperforms state-of-art baseline data-central models with NN5 and Ericsson KPI data. Meanwhile, the federated settings of purposed model yields comparable results to data-central settings on both NN5 and Ericsson KPI data. These results together answer the research question of this thesis. / Tidseriedata har blivit allmänt förekommande tack vare överkomliga kantenheter och sensorer. Mycket av denna data är värdefull för beslutsfattande. För att kunna använda datan för prognosuppgifter har den konventionella centraliserade metoden visat brister avseende storskalig datakommunikation och integritetsfrågor. Vidare har neurala nätverksmodeller inte klarat av att utnyttja den extra informationen från tidsserierna, vilket leder till misslyckanden med att ge specifikt tidsserierelaterade resultat. Båda frågorna exponerar en utmaning för storskalig tidsserieprognostisering med neurala nätverksmodeller. Alla dessa begränsningar leder till vår forskningsfråga:Kan vi realisera decentraliserad tidsserieprognostisering med en federerad lärningsmekanism som presterar jämförbart med konventionella centrala lösningar i prognostisering?I det här arbetet föreslår vi ett ramverk för federerad tidsserieprognos som löser utmaningen genom att låta användaren behålla data lokalt och lära sig en delad modell genom att aggregera lokalt beräknade uppdateringar. Dessutom utformar vi en hybrid modell för att möjliggöra neurala nätverksmodeller som kan utnyttja den extra informationen från tidsserierna för att uppnå inlärning av specifika tidsserier. Den föreslagna hybrida modellen presterar bättre än state-of-art centraliserade grundläggande modeller med NN5och Ericsson KPIdata. Samtidigt ger den federerade ansatsen jämförbara resultat med de datacentrala ansatserna för både NN5och Ericsson KPI-data. Dessa resultat svarar tillsammans på forskningsfrågan av denna avhandling.

<p>Graph representation learning (GRL) has been increasing used to model and understand data from a wide variety of complex systems spanning social, technological, bio-chemical and physical domains. GRL consists of two main components (1) a parametrized encoder that provides representations of graph data and (2) a learning process to train the encoder parameters. Designing flexible encoders that capture the underlying invariances and characteristics of graph data are crucial to the success of GRL. On the other hand, the learning process drives the quality of the encoder representations and developing principled learning mechanisms are vital for a number of growing applications in self-supervised, transfer and federated learning settings. To this end, we propose a suite of models and learning algorithms for GRL which form the two main thrusts of this dissertation.</p> <p><br></p> <p>In Thrust I, we propose two novel encoders which build upon on a widely popular GRL encoder class called graph neural networks (GNNs). First, we empirically study the prediction performance of current GNN based encoders when applied to graphs with heterogeneous node mixing patterns using our proposed notion of local assortativity. We find that GNN performance in node prediction tasks strongly correlates with our local assortativity metric---thereby introducing a limit. We propose to transform the input graph into a computation graph with proximity and structural information as distinct types of edges. We then propose a novel GNN based encoder that operates on this computation graph and adaptively chooses between structure and proximity information. Empirically, adopting our transformation and encoder framework leads to improved node classification performance compared to baselines in real-world graphs that exhibit diverse mixing.</p> <p>Secondly, we study the trade-off between expressivity and efficiency of GNNs when applied to temporal graphs for the task of link ranking. We develop an encoder that incorporates a labeling approach designed to allow for efficient inference over the candidate set jointly, while provably boosting expressivity. We also propose to optimize a list-wise loss for improved ranking. With extensive evaluation on real-world temporal graphs, we demonstrate its improved performance and efficiency compared to baselines.</p> <p><br></p> <p>In Thrust II, we propose two principled encoder learning mechanisms for challenging and realistic graph data settings. First, we consider a scenario where only limited or even no labelled data is available for GRL. Recent research has converged on graph contrastive learning (GCL), where GNNs are trained to maximize the correspondence between representations of the same graph in its different augmented forms. However, we find that GNNs trained by traditional GCL often risk capturing redundant graph features and thus may be brittle and provide sub-par performance in downstream tasks. We then propose a novel principle, termed adversarial-GCL (AD-GCL), which enables GNNs to avoid capturing redundant information during the training by optimizing adversarial graph augmentation strategies used in GCL. We pair AD-GCL with theoretical explanations and design a practical instantiation based on trainable edge-dropping graph augmentation. We experimentally validate AD-GCL by comparing with state-of-the-art GCL methods and achieve performance gains in semi-supervised, unsupervised and transfer learning settings using benchmark chemical and biological molecule datasets. </p> <p>Secondly, we consider a scenario where graph data is silo-ed across clients for GRL. We focus on two unique challenges encountered when applying distributed training to GRL: (i) client task heterogeneity and (ii) label scarcity. We propose a novel learning framework called federated self-supervised graph learning (FedSGL), which first utilizes a self-supervised objective to train GNNs in a federated fashion across clients and then, each client fine-tunes the obtained GNNs based on its local task and available labels. Our framework enables the federated GNN model to extract patterns from the common feature (attribute and graph topology) space without the need of labels or being biased by heterogeneous local tasks. Extensive empirical study of FedSGL on both node and graph classification tasks yields fruitful insights into how the level of feature / task heterogeneity, the adopted federated algorithm and the level of label scarcity affects the clients’ performance in their tasks.</p>