In this work, we study the trade-off between the reliability and the investment cost of an unmanned aerial system (UAS) consisting of a set of unmanned aerial vehicles (UAVs) carrying radio access nodes, called portable access points (PAPs)), deployed to serve a set of ground nodes (GNs). Using the proposed algorithm, a given geographical region is equivalently represented as a set of circular regions, where each circle represents the coverage region of a PAP. Then, the steady-state availability of the UAS is analytically derived by modelling it as a continuous time birth-death Markov decision process (MDP). Numerical evaluations show that the investment cost to guarantee a given steady-state availability to a set of GNs can be reduced by considering the traffic demand and distribution of GNs.

One of the most fundamental problems in machine learning is to compare examples: Given a pair of objects we want to return a value which indicates degree of (dis)similarity. Similarity is often task specific, and pre-defined distances can perform poorly, leading to work in metric learning. However, being able to learn a similarity-sensitive distance function also presupposes access to a rich, discriminative representation for the objects at hand. In this dissertation we present contributions towards both ends. In the first part of the thesis, assuming good representations for the data, we present a formulation for metric learning that makes a more direct attempt to optimize for the k-NN accuracy as compared to prior work. We also present extensions of this formulation to metric learning for kNN regression, asymmetric similarity learning and discriminative learning of Hamming distance. In the second part, we consider a situation where we are on a limited computational budget i.e. optimizing over a space of possible metrics would be infeasible, but access to a label aware distance metric is still desirable. We present a simple, and computationally inexpensive approach for estimating a well motivated metric that relies only on gradient estimates, discussing theoretical and experimental results. In the final part, we address representational issues, considering group equivariant convolutional neural networks (GCNNs). Equivariance to symmetry transformations is explicitly encoded in GCNNs; a classical CNN being the simplest example. In particular, we present a SO(3)-equivariant neural network architecture for spherical data, that operates entirely in Fourier space, while also providing a formalism for the design of fully Fourier neural networks that are equivariant to the action of any continuous compact group.

We consider communication in a fully cooperative multi-agent system, where the agents have partial observation of the environment and must act jointly to maximize the overall reward. We have a discrete-time queueing network where agents route packets to queues based only on the partial information of the current queue lengths. The queues have limited buffer capacity, so packet drops happen when they are sent to a full queue. In this work, we implemented a communication channel for the agents to share their information in order to reduce the packet drop rate. For efficient information sharing we use an attention-based communication model, called ATVC, to select informative messages from other agents. The agents then infer the state of queues using a combination of the variational auto-encoder, VAE, and product-of-experts, PoE, model. Ultimately, the agents learn what they need to communicate and with whom, instead of communicating all the time with everyone. We also show empirically that ATVC is able to infer the true state of the queues and leads to a policy which outperforms existing baselines.

The proliferation of artificial intelligence is increasingly dependent on model understanding. Understanding demands both an interpretation - a human reasoning about a model's behavior - and an explanation - a symbolic representation of the functioning of the model. Notwithstanding the imperative of transparency for safety, trust, and acceptance, the opacity of state-of-the-art reinforcement learning algorithms conceals the rudiments of their learned strategies. We have developed a policy regularization method that asserts the global intrinsic affinities of learned strategies. These affinities provide a means of reasoning about a policy's behavior, thus making it inherently interpretable. We have demonstrated our method in personalized prosperity management where individuals' spending behavior in time dictate their investment strategies, i.e. distinct spending personalities may have dissimilar associations with different investment classes. We now explain our model by reproducing the underlying prototypical policies with discretized Markov models. These global surrogates are symbolic representations of the prototypical policies.

Reinforcement learning (RL) often requires decomposing a problem into subtasks and composing learned behaviors on these tasks. Compositionality in RL has the potential to create modular subtask units that interface with other system capabilities. However, generating compositional models requires the characterization of minimal assumptions for the robustness of the compositional feature. We develop a framework for a \emph{compositional theory} of RL using a categorical point of view. Given the categorical representation of compositionality, we investigate sufficient conditions under which learning-by-parts results in the same optimal policy as learning on the whole. In particular, our approach introduces a category $\mathsf{MDP}$, whose objects are Markov decision processes (MDPs) acting as models of tasks. We show that $\mathsf{MDP}$ admits natural compositional operations, such as certain fiber products and pushouts. These operations make explicit compositional phenomena in RL and unify existing constructions, such as puncturing hazardous states in composite MDPs and incorporating state-action symmetry. We also model sequential task completion by introducing the language of zig-zag diagrams that is an immediate application of the pushout operation in $\mathsf{MDP}$.

This research work is about recent development made in speech recognition. In this research work, analysis of isolated digit recognition in the presence of different bit rates and at different noise levels has been performed. This research work has been carried using audacity and HTK toolkit. Hidden Markov Model (HMM) is the recognition model which was used to perform this experiment. The feature extraction techniques used are Mel Frequency Cepstrum coefficient (MFCC), Linear Predictive Coding (LPC), perceptual linear predictive (PLP), mel spectrum (MELSPEC), filter bank (FBANK). There were three types of different noise levels which have been considered for testing of data. These include random noise, fan noise and random noise in real time environment. This was done to analyse the best environment which can used for real time applications. Further, five different types of commonly used bit rates at different sampling rates were considered to find out the most optimum bit rate.

Non-intrusive load monitoring (NILM) aims at energy consumption and appliance state information retrieval from aggregated consumption measurements, with the help of signal processing and machine learning algorithms. Representation learning with deep neural networks is successfully applied to several related disciplines. The main advantage of representation learning lies in replacing an expert-driven, hand-crafted feature extraction with hierarchical learning from many representations in raw data format. In this paper, we show how the NILM processing-chain can be improved, reduced in complexity and alternatively designed with recent deep learning algorithms. On the basis of an event-based appliance recognition approach, we evaluate seven different classification models: a classical machine learning approach that is based on a hand-crafted feature extraction, three different deep neural network architectures for automated feature extraction on raw waveform data, as well as three baseline approaches for raw data processing. We evaluate all approaches on two large-scale energy consumption datasets with more than 50,000 events of 44 appliances. We show that with the use of deep learning, we are able to reach and surpass the performance of the state-of-the-art classical machine learning approach for appliance recognition with an F-Score of 0.75 and 0.86 compared to 0.69 and 0.87 of the classical approach.

We investigate online Markov Decision Processes (MDPs) with adversarially changing loss functions and known transitions. We choose dynamic regret as the performance measure, defined as the performance difference between the learner and any sequence of feasible changing policies. The measure is strictly stronger than the standard static regret that benchmarks the learner's performance with a fixed compared policy. We consider three foundational models of online MDPs, including episodic loop-free Stochastic Shortest Path (SSP), episodic SSP, and infinite-horizon MDPs. For these three models, we propose novel online ensemble algorithms and establish their dynamic regret guarantees respectively, in which the results for episodic (loop-free) SSP are provably minimax optimal in terms of time horizon and certain non-stationarity measure. Furthermore, when the online environments encountered by the learner are predictable, we design improved algorithms and achieve better dynamic regret bounds for the episodic (loop-free) SSP; and moreover, we demonstrate impossibility results for the infinite-horizon MDPs.

Imbalanced learning is a fundamental challenge in data mining, where there is a disproportionate ratio of training samples in each class. Over-sampling is an effective technique to tackle imbalanced learning through generating synthetic samples for the minority class. While numerous over-sampling algorithms have been proposed, they heavily rely on heuristics, which could be sub-optimal since we may need different sampling strategies for different datasets and base classifiers, and they cannot directly optimize the performance metric. Motivated by this, we investigate developing a learning-based over-sampling algorithm to optimize the classification performance, which is a challenging task because of the huge and hierarchical decision space. At the high level, we need to decide how many synthetic samples to generate. At the low level, we need to determine where the synthetic samples should be located, which depends on the high-level decision since the optimal locations of the samples may differ for different numbers of samples. To address the challenges, we propose AutoSMOTE, an automated over-sampling algorithm that can jointly optimize different levels of decisions. Motivated by the success of SMOTE~\cite{chawla2002smote} and its extensions, we formulate the generation process as a Markov decision process (MDP) consisting of three levels of policies to generate synthetic samples within the SMOTE search space. Then we leverage deep hierarchical reinforcement learning to optimize the performance metric on the validation data. Extensive experiments on six real-world datasets demonstrate that AutoSMOTE significantly outperforms the state-of-the-art resampling algorithms. The code is at https://github.com/daochenzha/autosmote

Although deep reinforcement learning (RL) has recently enjoyed many successes, its methods are still data inefficient, which makes solving numerous problems prohibitively expensive in terms of data. We aim to remedy this by taking advantage of the rich supervisory signal in unlabeled data for learning state representations. This thesis introduces three different representation learning algorithms that have access to different subsets of the data sources that traditional RL algorithms use: (i) GRICA is inspired by independent component analysis (ICA) and trains a deep neural network to output statistically independent features of the input. GrICA does so by minimizing the mutual information between each feature and the other features. Additionally, GrICA only requires an unsorted collection of environment states. (ii) Latent Representation Prediction (LARP) requires more context: in addition to requiring a state as an input, it also needs the previous state and an action that connects them. This method learns state representations by predicting the representation of the environment's next state given a current state and action. The predictor is used with a graph search algorithm. (iii) RewPred learns a state representation by training a deep neural network to learn a smoothed version of the reward function. The representation is used for preprocessing inputs to deep RL, while the reward predictor is used for reward shaping. This method needs only state-reward pairs from the environment for learning the representation. We discover that every method has their strengths and weaknesses, and conclude from our experiments that including unsupervised representation learning in RL problem-solving pipelines can speed up learning.