July 19, 2018 In this thesis we present an investigation of multi-task and transfer learning using the recently introduced task of semantic tagging. First we employ a number of natural language processing tasks as auxiliaries for semantic tag- ging. Secondly, going in the other direction, we employ seman- tic tagging as an auxiliary task for three di erent NLP tasks: Part-of-Speech Tagging, Universal Dependency parsing, and Natural Language Inference. We compare full neural network sharing, partial neural network sharing, and what we term the learning what to share setting where neg- ative transfer between tasks is less likely. Fi- nally, we investigate multi-lingual learning framed as a special case of multi-task learning. Our ndings show considerable improvements for most experiments, demonstrating a variety of cases where multi-task and transfer learning methods are bene cial. 1 References 2

No description available.

No description available.

No description available.

Schüler/innen zu befähigen, die allgegenwärtige Rechentechnik in ihrem Umweld zu erkennen und zu bewerten ist ein international proklamiertes Ziel sekundärer Informatikbildung. Zu diesem Zweck müssen sie von ihrem schulischen Wissen auch tatsächlich im Alltag Gebrauch machen. Ausgehend von Theorien zu Lerntransfer und existierender Forschung zu Schülervorstellungen, untersucht diese Dissertation die Denk- und Lernprozesse von Schüler/innen über konkrete informatische Geräte. Die erste Studie untersucht, welche Arten von Technik Schüler/innen allgemein unterscheiden. Ich stelle eine Grounded Theory zu einer entsprechenden Taxonomie vor. Diese legt nahe, dass Rechentechnik keine vordergründige Kategorie für sie darstellt, was entsprechenden Transfer erschweren würde. Die zweite Studie untersucht, wie Schüler/innen Rechen- von Nicht-Rechentechnik unterscheiden. Ich stelle eine Grounded Theory entsprechender Denkprozesse vor. Diese zeigt, dass etliche Schüler/innen Rechentechnik unsachgemäß anhand inhärenter Fähigkeitsgrenzen unterscheiden, was ebenfalls Transfer behindern würde. Die dritte Studie untersucht daraufhin Lernprozesse im Kontext einer Intervention, die die oben genannten Punkte adressieren soll. Sie zeigt, dass einige Schüler/innen Probleme damit haben, Rechentechnik als gleichzeitig ökonomisch und leistungsfähig zu verstehen, was wiederum seine Verbreitung und Auswirkungen einschränkt. Die Analyse legt zudem erste Richtlinien für das Design entsprechender Interventionen nahe. Die Studien werden anschließend integriert diskutiert. Insbesondere stelle ich Lernziele und Aktivitäten vor, welche eine Teilantwort meiner ursprünglichen Leitfrage bilden: was müssen Schüler/innen lernen, um Rechentechnik im Alltag adäquat zu erkennen und zu bewerten? Ich diskutiere Implikationen für die Praxis sowie potentielle weiterführende Forschung, vor allem im Bezug zu einer Informatikbildung, die sich als Säule moderner Allgemeinbildung versteht. / Enabling students to recognize and evaluate the ubiquitous computing technologies in their lives is an internationally proclaimed goal of a secondary informatics education. To that end, they need to actually engage with their school-learned knowledge in the context of everyday situations. Based on theories of knowledge transfer and prior research on student conceptions, this thesis investigates students' related thinking and learning processes. The first study investigates what kinds of technology students generally distinguish. I propose a grounded theory for a related taxonomy. It suggests that computing technology is, in fact, not a very salient kind of technology for many, which poses a challenge for related transfer. The second study investigates how students even distinguish computing from non-computing technology. I propose a grounded theory of their related reasoning processes. It shows that students may inappropriately distinguish computing devices on the basis of inherent capability limitations, which would also be detrimental to transfer. The third study investigates students' learning processes in the context of an intervention designed to address these issues. It revealed that several students apparently had difficulty to conceive of computing technology as simultaneously economical and powerful, thus limiting its potential ubiquity and impact. The analysis also indicates some initial guidelines for the design of related interventions. The three studies are then integrated and discussed. In particular, I propose a set of learning objectives and activities as a partial answer to my original guiding question: what is it that students need to learn in order to adequately recognize and evaluate computing technologies in their lives? I discuss implications for practice and potential avenues for future research, especially with respect to a general informatics education that regards itself as part of a contemporary general education.

Integrating state-of-the-art sensory and display technologies with 3D computer graphics, extended reality (XR) affords capabilities to create enhanced human experiences by merging virtual elements with the real world. To better understand how Sound and Music Computing (SMC) can benefit from the capabilities of XR, this thesis presents novel research on the de- sign of musical experiences in extended reality (MusE-XR). Integrating XR with research on computer assisted musical instrument tutoring (CAMIT) as well as New Interfaces for Musical Expression (NIME), I explore the MusE-XR design space to contribute to a better understanding of the capabilities of XR for SMC. The first area of focus in this thesis is the application of XR technologies to CAMIT enabling extended reality enhanced musical instrument learning (XREMIL). A common approach in CAMIT is the automatic assessment of musical performance. Generally, these systems focus on the aural quality of the performance, but emerging XR related sensory technologies afford the development of systems to assess playing technique. Employing these technologies, the first contribution in this thesis is a CAMIT system for the automatic assessment of pianist hand posture using depth data. Hand posture assessment is performed through an applied computer vision (CV) and machine learning (ML) pipeline to classify a pianist’s hands captured by a depth camera into one of three posture classes. Assessment results from the system are intended to be integrated into a CAMIT interface to deliver feedback to students regarding their hand posture. One method to present the feedback is through real-time visual feedback (RTVF) displayed on a standard 2D computer display, but this method is limited by a need for the student to constantly shift focus between the instrument and the display. XR affords new methods to potentially address this limitation through capabilities to directly augment a musical instrument with RTVF by overlaying 3D virtual objects on the instrument. Due to limited research evaluating effectiveness of this approach, it is unclear how the added cognitive demands of RTVF in virtual environments (VEs) affect the learning process. To fill this gap, the second major contribution of this thesis is the first known user study evaluating the effectiveness of XREMIL. Results of the study show that an XR environment with RTVF improves participant performance during training, but may lead to decreased improvement after the training. On the other hand,interviews with participants indicate that the XR environment increased their confidence leading them to feel more engaged during training. In addition to enhancing CAMIT, the second area of focus in this thesis is the application of XR to NIME enabling virtual environments for musical expression (VEME). Development of VEME requires a workflow that integrates XR development tools with existing sound design tools. This presents numerous technical challenges, especially to novice XR developers. To simplify this process and facilitate VEME development, the third major contribution of this thesis is an open source toolkit, called OSC-XR. OSC-XR makes VEME development more accessible by providing developers with readily available Open Sound Control (OSC) virtual controllers. I present three new VEMEs, developed with OSC-XR, to identify affordances and guidelines for VEME design. The insights gained through these studies exploring the application of XR to musical learning and performance, lead to new affordances and guidelines for the design of effective and engaging MusE-XR. / Graduate

A high quality and large scale training data-set is an important guarantee to teach an ideal classifier for image classification. Manually constructing a training data- set  with  appropriate  labels  is  an  expensive  and  time  consuming  task.    Active learning techniques have been used to improved the existing models by reducing the  number  of  required  annotations.    The  present  work  aims  to  investigate the  way  to  build  a  model  for  identifying  and  utilizing  potential  informative and  representativeness  unlabeled  samples.    To  this  end,  two  approaches  for deep image classification using active learning are proposed, implemented and evaluated.  The two versions of active leaning for deep image classification differ in  the  input  space  exploration  so  as  to  investigate  how  classifier  performance varies  when  automatic  labelization  on  the  high  confidence  unlabeled  samples is  performed.    Active  learning  heuristics  based  on  uncertainty  measurements on low confidence predicted samples,  a pseudo-labelization technique to boost active  learning  by  reducing  the  number  of  human  interactions  and  knowledge transferring  form  pre-trained  models,  are  proposed  and  combined  into  our methodology.  The experimental results on two benchmark image classification data-sets  verify  the  effectiveness  of  the  proposed  methodology.    In  addition, a  new  pool-based  active  learning  query  strategy  is  proposed.     Dealing  with retraining-based algorithms we define a ”forgetting event” to have occurred when an  individual  training  example  transitions  the  maximum  predicted  probability class over the course of retraining. We integrated the new approach with the semi- supervised learning method in order to tackle the above challenges and observedgood performance against existing methods. / En  högkvalitativ  och  storskalig  träningsdataset  är  en  viktig  garanti  för  att  bli en  idealisk  klassificerare  för  bildklassificering.     Att  manuellt  konstruera  en träningsdatasats  med  lämpliga  etiketter  är  en  dyr  och  tidskrävande  uppgift. Aktiv  inlärningstekniker  har  använts  för  att  förbättra  de  befintliga  modellerna genom att minska antalet nödvändiga annoteringar. Det nuvarande arbetet syftar till  att  undersöka  sättet  att  bygga  en  modell  för  att  identifiera  och  använda potentiella informativa och representativa omärkta prover.   För detta ändamål föreslås, genomförs och genomförs två metoder för djup bildklassificering med aktivt  lärande  utvärderas.      De  två  versionerna  av  aktivt  lärande  för  djup bildklassificering  skiljer  sig  åt  i  undersökningen  av  ingångsutrymmet  för  att undersöka hur klassificeringsprestanda varierar när automatisk märkning på de omärkta  proverna  med  hög  konfidens  utförs.   Aktiv  lärande  heuristik  baserad på  osäkerhetsmätningar  på  förutsagda  prover  med  låg  konfidens,  en  pseudo- märkningsteknik för att öka aktivt lärande genom att minska antalet mänskliga interaktioner  och  kunskapsöverföring  av  förutbildade  modeller,  föreslås  och kombineras   i   vår   metod.      Experimentella   resultat   på   två   riktmärken   för bildklassificering datauppsättningar verifierar effektiviteten hos den föreslagna metodiken.   Dessutom föreslås en ny poolbaserad aktiv inlärningsfrågestrategi. När  vi  använder  omskolningsbaserade  algoritmer  definierar  vi  en  ”glömmer händelse” som skulle ha inträffat när ett individuellt träningsexempel överskrider den maximala förutsagda sannolikhetsklassen under omskolningsprocessen.  Vi integrerade den nya metoden med den semi-övervakad inlärning för att hanteraovanstående utmaningar och observeras bra prestanda mot befintliga metoder.

Disertace se zabývá aplikaci neuronových sítí pro detekci objektů na multimodální data v robotice. Celkem cílí na tři oblasti: tvorbu datasetu, zpracování multimodálních dat a trénování neuronových sítí. Nejdůležitější části práce je návrh metody pro tvorbu rozsáhlých anotovaných datasetů bez časové náročného lidského zásahu. Metoda používá neuronové sítě trénované na RGB obrázcích. Užitím dat z několika snímačů pro vytvoření modelu okolí a mapuje anotace z RGB obrázků na jinou datovou doménu jako jsou termální obrázky, či mračna bodů. Pomoci této metody autor vytvořil dataset několika set tisíc anotovaných obrázků a použil je pro trénink neuronové sítě, která následně překonala modely trénované na menších, lidmi anotovaných datasetech. Dále se autor v práci zabývá robustností detekce objektů v několika datových doménách za různých povětrnostních podmínek. Práce také popisuje kompletní řetězec zpracování multimodálních dat, které autor vytvořil během svého doktorského studia. To Zahrnuje vývoj unikátního senzorického zařízení, které je vybavené řadou snímačů běžně užívaných v robotice. Dále autor popisuje proces tvorby rozsáhlého, veřejně dostupného datasetu Brno Urban Dataset. Na závěr autor popisuje software, který vznikl během jeho studia a jak je tento software užit při zpracování dat v rámci jeho práce (Atlas Fusion a Robotic Template Library).