Speech to text models tend to be trained and evaluated against a single target accent. This is especially true for English for which native speakers from the United States became the main benchmark. In this work, we are going to show how two simple methods: pre-trained embeddings and auxiliary classification losses can improve the performance of ASR systems. We are looking for upgrades as universal as possible and therefore we will explore their impact on several models architectures and several languages.

This thesis considers sequential decision problems, where the loss/reward incurred by selecting an action may not be inferred from observed feedback. A major part of this thesis focuses on the unsupervised sequential selection problem, where one can not infer the loss incurred for selecting an action from observed feedback. We also introduce a new setup named Censored Semi Bandits, where the loss incurred for selecting an action can be observed under certain conditions. Finally, we study the channel selection problem in the communication networks, where the reward for an action is only observed when no other player selects that action to play in the round. These problems find applications in many fields like healthcare, crowd-sourcing, security, adaptive resource allocation, among many others. This thesis aims to address the above-described sequential decision problems by exploiting specific structures these problems exhibit. We develop provably optimal algorithms for each of these setups with weak feedback and validate their empirical performance on different problem instances derived from synthetic and real datasets.

Network embedding (NE) approaches have emerged as a predominant technique to represent complex networks and have benefited numerous tasks. However, most NE approaches rely on a homophily assumption to learn embeddings with the guidance of supervisory signals, leaving the unsupervised heterophilous scenario relatively unexplored. This problem becomes especially relevant in fields where a scarcity of labels exists. Here, we formulate the unsupervised NE task as an r-ego network discrimination problem and develop the SELENE framework for learning on networks with homophily and heterophily. Specifically, we design a dual-channel feature embedding pipeline to discriminate r-ego networks using node attributes and structural information separately. We employ heterophily adapted self-supervised learning objective functions to optimise the framework to learn intrinsic node embeddings. We show that SELENE's components improve the quality of node embeddings, facilitating the discrimination of connected heterophilous nodes.

Patient triage at emergency departments (EDs) is necessary to prioritize care for patients with critical and time-sensitive conditions. Different tools are used for patient triage and one of the most common ones is the emergency severity index (ESI), which has a scale of five levels, where level 1 is the most urgent and level 5 is the least urgent. This paper proposes a framework for utilizing machine learning to develop an e-triage tool that can be used at EDs. A large retrospective dataset of ED patient visits is obtained from the electronic health record of a healthcare provider in the Midwest of the US for three years. However, the main challenge of using machine learning algorithms is that most of them have many parameters and without optimizing these parameters, developing a high-performance model is not possible. This paper proposes an approach to optimize the hyperparameters of machine learning. The metaheuristic optimization algorithms simulated annealing (SA) and adaptive simulated annealing (ASA) are proposed to optimize the parameters of extreme gradient boosting (XGB) and categorical boosting (CaB). The newly proposed algorithms are SA-XGB, ASA-XGB, SA-CaB, ASA-CaB. Grid search (GS), which is a traditional approach used for machine learning fine-tunning is also used to fine-tune the parameters of XGB and CaB, which are named GS-XGB and GS-CaB. The six algorithms are trained and tested using eight data groups obtained from the feature selection phase. The results show ASA-CaB outperformed all the proposed algorithms with accuracy, precision, recall, and f1 of 83.3%, 83.2%, 83.3%, 83.2%, respectively.

Analogical proportions (AP) are statements of the form "a is to b ascis to d". They compare the pairs of items(a,b) and(c, d) in terms of their differences and similarities. The explicit use of APs in analogical reasoning has contributed to a renewal of its applications, leading to many developments, especially in the last decade; see [30] for a survey. However, even if much has been already done both at the theoretical and at the practical levels, the very nature of APs may not yet be fully understood and their full potential explored. In the following, we survey recent works on APs along three directions: their role in classification tasks [4]; their use for providing explanations [20]; their relation with multi-valued dependencies [21]. This just intends to be an introductory paper, and the reader is referred to the above references for more details on each issue.

The main target of retrosynthesis is to recursively decompose desired molecules into available building blocks. Existing template-based retrosynthesis methods follow a template selection stereotype and suffer from limited training templates, which prevents them from discovering novel reactions. To overcome this limitation, we propose an innovative retrosynthesis prediction framework that can compose novel templates beyond training templates. As far as we know, this is the first method that uses machine learning to compose reaction templates for retrosynthesis prediction. Besides, we propose an effective reactant candidate scoring model that can capture atom-level transformations, which helps our method outperform previous methods on the USPTO-50K dataset. Experimental results show that our method can produce novel templates for 15 USPTO-50K test reactions that are not covered by training templates. We have released our source implementation.

Landmine blasts can be fatal and cause injuries including blindness, burns, damaged limbs, and shrapnel wounds. While many nations have stopped using and producing landmines, 59 countries and territories remain contaminated by mines or other explosives. In 2019, landmines and similar explosives caused at least 5,554 casualties, across 55 countries and regions, with civilians accounting for the majority (80%) and children representing nearly half of civilian casualties (43%). Over one million landmines were dropped in Afghanistan in the 1980s. About two million landmines have been planted on the Korean Peninsula since the Korean War ended in 1953.

Prior works on improving speech quality with visual input typically study each type of auditory distortion separately (e.g., separation, inpainting, video-to-speech) and present tailored algorithms. This paper proposes to unify these subjects and study Generalized Speech Enhancement, where the goal is not to reconstruct the exact reference clean signal, but to focus on improving certain aspects of speech. In particular, this paper concerns intelligibility, quality, and video synchronization. We cast the problem as audio-visual speech resynthesis, which is composed of two steps: pseudo audio-visual speech recognition (P-AVSR) and pseudo text-to-speech synthesis (P-TTS). P-AVSR and P-TTS are connected by discrete units derived from a self-supervised speech model. Moreover, we utilize self-supervised audio-visual speech model to initialize P-AVSR. The proposed model is coined ReVISE. ReVISE is the first high-quality model for in-the-wild video-to-speech synthesis and achieves superior performance on all LRS3 audio-visual enhancement tasks with a single model. To demonstrates its applicability in the real world, ReVISE is also evaluated on EasyCom, an audio-visual benchmark collected under challenging acoustic conditions with only 1.6 hours of training data. Similarly, ReVISE greatly suppresses noise and improves quality. Project page: https://wnhsu.github.io/ReVISE.

The celebrated proverb that "speech is silver, silence is golden" has a long multinational history and multiple specific meanings. In written texts punctuation can in fact be considered one of its manifestations. Indeed, the virtue of effectively speaking and writing involves - often decisively - the capacity to apply the properly placed breaks. In the present study, based on a large corpus of world-famous and representative literary texts in seven major Western languages, it is shown that the distribution of intervals between consecutive punctuation marks in almost all texts can universally be characterised by only two parameters of the discrete Weibull distribution which can be given an intuitive interpretation in terms of the so-called hazard function. The values of these two parameters tend to be language-specific, however, and even appear to navigate translations. The properties of the computed hazard functions indicate that among the studied languages, English turns out to be the least constrained by the necessity to place a consecutive punctuation mark to partition a sequence of words. This may suggest that when compared to other studied languages, English is more flexible, in the sense of allowing longer uninterrupted sequences of words. Spanish reveals similar tendency to only a bit lesser extent.

Music discovery services let users identify songs from short mobile recordings. These solutions are often based on Audio Fingerprinting, and rely more specifically on the extraction of spectral peaks in order to be robust to a number of distortions. Few works have been done to study the robustness of these algorithms to background noise captured in real environments. In particular, AFP systems still struggle when the signal to noise ratio is low, i.e when the background noise is strong. In this project, we tackle this problematic with Deep Learning. We test a new hybrid strategy which consists of inserting a denoising DL model in front of a peak-based AFP algorithm. We simulate noisy music recordings using a realistic data augmentation pipeline, and train a DL model to denoise them. The denoising model limits the impact of background noise on the AFP system's extracted peaks, improving its robustness to noise. We further propose a novel loss function to adapt the DL model to the considered AFP system, increasing its precision in terms of retrieved spectral peaks. To the best of our knowledge, this hybrid strategy has not been tested before.