Ultimatum salary bargaining with real effort

We report experimental results on ultimatum salary bargaining with a real task performed by employee subjects. Compared to the baseline treatment with a hypothetical task, the introduction of a real task raises offers, accepted wages and rejection rates.

Affirmative action and effort choice: An experimental investigation

We study the effect of affirmative action on effort in an experiment conducted in high schools in socioeconomically disadvantaged areas in Queensland, Australia. All participating schools have a large representation of indigenous Australians, a population group that is frequently targeted by affirmative action. Our participants perform a simple real-effort task in a competitive setting. Those ranked in the top third receive a high piece-rate payment and all the others receive a low payment. We introduce affirmative action by providing the lowest (bottom third) performers with a positive handicap increasing their chances to achieve the high payment target. Our findings show that the policy increases effort of those that it aims to favour, without discouraging effort of those who are indirectly penalized by affirmative action.

Multi-task learning using shared and task specific information

Multi-task learning solves multiple related learning problems simultaneously by sharing some common structure for improved generalization performance of each task. We propose a novel approach to multi-task learning which captures task similarity through a shared basis vector set. The variability across tasks is captured through task specific basis vector set. We use sparse support vector machine (SVM) algorithm to select the basis vector sets for the tasks. The approach results in a sparse model where the prediction is done using very few examples. The effectiveness of our approach is demonstrated through experiments on synthetic and real multi-task datasets.

Tournament outcomes and prosocial behaviour

Does participation in a tournament influence prosocial behaviour in subsequent interactions? We designed an experiment to collect data on charitable donations made by participants out of their earnings from a real-effort tournament. We varied the earnings associated with ranks across our treatments thereby allowing us to observe donations by participants who end up at different ranks but have the same earnings. Prior to finding out how well they performed, participants were also asked to report their expected rank. Controlling for differences in effort and earnings, participants who were ranked first donated significantly more than others, supporting the view that positive affect from winning may increase generosity. However, we find that this effect diminishes when the difference between realised and expected ranks are controlled for, lending support to the idea that positive surprise from winning also increases generosity.

“Success breeds success” or “Pride goes before a fall”? ☆: Teams and individuals in multi-contest tournaments

We study the impact of progress feedback on players' performance in multi-contest team tournaments, in which team members' efforts are not directly substitutable. In particular, we employ a real-effort laboratory experiment to understand, in a best-of-three tournament, how players' strategic mindsets change when they compete on a team compared to when they compete individually. Our data corroborate the theoretical predictions for teams: Neither a lead nor a lag in the first component contest affects a team's performance in the subsequent contests. In individual tournaments, however, contrary to the theoretical prediction, we observe that leaders perform worse—but laggards perform better—after learning the outcome of the first contest. Our findings offer the first empirical evidence from a controlled laboratory of the impact of progress feedback between team and individual tournaments, and contribute new insights on team incentives.

Applying systematic design: the flight refuelling probe project

This report is a product of close industry-academia collaboration between British Aerospace and the Cambridge Engineering Design Centre (EDC). British Aerospace designs and integrates some of the most complex systems in the world, and its expertise in this field has enabled the company to become the United Kingdom's largest exporter. However, to stay at the forefront of the highly competitive aerospace industry it is necessary to seek new ways to work more effectively and more efficiently. The Cambridge EDC has played a part in supporting these needs by providing access to the methods and tools that it has developed for improving the process of designing mechanical systems. The EDC has gained an international reputation for the quality of its work in this subject. Thus, the collaboration is between two organisations each of whom are leaders in their respective fields. The central aim of the project has been to demonstrate how a systematic design process can be applied to a real design task identified by industry. The task selected was the design of a flight refuelling probe which would enable a combat aircraft to refuel from a "flying tanker". However, the systematic approach, methods and tools described in this report are applicable to most engineering design tasks. The findings presented in this report provide a sound basis for comparing the recommended systematic design process with industrial practice. The results of this comparison would enable the company to define ways in which its existing design process can be improved. This research project has a high degree of industrial relevance. The value of the work may be judged in terms of the opportunities it opens up for positive changes to the company's engineering operations. Several members of the EDC have contributed to the project. These include Dr Lucienne Blessing, Dr Stuart Burgess, Dr Amaresh Chakrabarti, Major Mark Nowack, Aylmer Johnson and Dr Paul Weaver. At British Aerospace special thanks must go to Alan Dean and David Halliday for their interest and the support they have given. The project has been managed by Dr Nigel Upton of British Aerospace during a 3 year secondment to the EDC.

On-line policy optimisation of spoken dialogue systems via live interaction with human subjects

Statistical dialogue models have required a large number of dialogues to optimise the dialogue policy, relying on the use of a simulated user. This results in a mismatch between training and live conditions, and significant development costs for the simulator thereby mitigating many of the claimed benefits of such models. Recent work on Gaussian process reinforcement learning, has shown that learning can be substantially accelerated. This paper reports on an experiment to learn a policy for a real-world task directly from human interaction using rewards provided by users. It shows that a usable policy can be learnt in just a few hundred dialogues without needing a user simulator and, using a learning strategy that reduces the risk of taking bad actions. The paper also investigates adaptation behaviour when the system continues learning for several thousand dialogues and highlights the need for robustness to noisy rewards. © 2011 IEEE.

Prospective, Randomized Assessment of Transfer of Training (ToT) and Transfer Effectiveness Ratio (TER) of Virtual Reality Simulation Training for Laparoscopic Skill Acquisition

OBJECTIVES:: We assessed the effectiveness of ToT from VR laparoscopic simulation training in 2 studies. In a second study, we also assessed the TER. ToT is a detectable performance improvement between equivalent groups, and TER is the observed percentage performance differences between 2 matched groups carrying out the same task but with 1 group pretrained on VR simulation. Concordance between simulated and in-vivo procedure performance was also assessed. DESIGN:: Prospective, randomized, and blinded. PARTICIPANTS:: In Study 1, experienced laparoscopic surgeons (n = 195) and in Study 2 laparoscopic novices (n = 30) were randomized to either train on VR simulation before completing an equivalent real-world task or complete the real-world task only. RESULTS:: Experienced laparoscopic surgeons and novices who trained on the simulator performed significantly better than their controls, thus demonstrating ToT. Their performance showed a TER between 7% and 42% from the virtual to the real tasks. Simulation training impacted most on procedural error reduction in both studies (32- 42%). The correlation observed between the VR and real-world task performance was r > 0·96 (Study 2). CONCLUSIONS:: VR simulation training offers a powerful and effective platform for training safer skills.

Deep learning of representations and its application to computer vision

L’objectif de cette thèse par articles est de présenter modestement quelques étapes du parcours qui mènera (on espère) à une solution générale du problème de l’intelligence artificielle. Cette thèse contient quatre articles qui présentent chacun une différente nouvelle méthode d’inférence perceptive en utilisant l’apprentissage machine et, plus particulièrement, les réseaux neuronaux profonds. Chacun de ces documents met en évidence l’utilité de sa méthode proposée dans le cadre d’une tâche de vision par ordinateur. Ces méthodes sont applicables dans un contexte plus général, et dans certains cas elles on tété appliquées ailleurs, mais ceci ne sera pas abordé dans le contexte de cette de thèse. Dans le premier article, nous présentons deux nouveaux algorithmes d’inférence variationelle pour le modèle génératif d’images appelé codage parcimonieux “spike- and-slab” (CPSS). Ces méthodes d’inférence plus rapides nous permettent d’utiliser des modèles CPSS de tailles beaucoup plus grandes qu’auparavant. Nous démontrons qu’elles sont meilleures pour extraire des détecteur de caractéristiques quand très peu d’exemples étiquetés sont disponibles pour l’entraînement. Partant d’un modèle CPSS, nous construisons ensuite une architecture profonde, la machine de Boltzmann profonde partiellement dirigée (MBP-PD). Ce modèle a été conçu de manière à simplifier d’entraînement des machines de Boltzmann profondes qui nécessitent normalement une phase de pré-entraînement glouton pour chaque couche. Ce problème est réglé dans une certaine mesure, mais le coût d’inférence dans le nouveau modèle est relativement trop élevé pour permettre de l’utiliser de manière pratique. Dans le deuxième article, nous revenons au problème d’entraînement joint de machines de Boltzmann profondes. Cette fois, au lieu de changer de famille de modèles, nous introduisons un nouveau critère d’entraînement qui donne naissance aux machines de Boltzmann profondes à multiples prédictions (MBP-MP). Les MBP-MP sont entraînables en une seule étape et ont un meilleur taux de succès en classification que les MBP classiques. Elles s’entraînent aussi avec des méthodes variationelles standard au lieu de nécessiter un classificateur discriminant pour obtenir un bon taux de succès en classification. Par contre, un des inconvénients de tels modèles est leur incapacité de générer deséchantillons, mais ceci n’est pas trop grave puisque la performance de classification des machines de Boltzmann profondes n’est plus une priorité étant donné les dernières avancées en apprentissage supervisé. Malgré cela, les MBP-MP demeurent intéressantes parce qu’elles sont capable d’accomplir certaines tâches que des modèles purement supervisés ne peuvent pas faire, telles que celle de classifier des données incomplètes ou encore celle de combler intelligemment l’information manquante dans ces données incomplètes. Le travail présenté dans cette thèse s’est déroulé au milieu d’une période de transformations importantes du domaine de l’apprentissage à réseaux neuronaux profonds qui a été déclenchée par la découverte de l’algorithme de “dropout” par Geoffrey Hinton. Dropout rend possible un entraînement purement supervisé d’architectures de propagation unidirectionnel sans être exposé au danger de sur- entraînement. Le troisième article présenté dans cette thèse introduit une nouvelle fonction d’activation spécialement con ̧cue pour aller avec l’algorithme de Dropout. Cette fonction d’activation, appelée maxout, permet l’utilisation de aggrégation multi-canal dans un contexte d’apprentissage purement supervisé. Nous démontrons comment plusieurs tâches de reconnaissance d’objets sont mieux accomplies par l’utilisation de maxout. Pour terminer, sont présentons un vrai cas d’utilisation dans l’industrie pour la transcription d’adresses de maisons à plusieurs chiffres. En combinant maxout avec une nouvelle sorte de couche de sortie pour des réseaux neuronaux de convolution, nous démontrons qu’il est possible d’atteindre un taux de succès comparable à celui des humains sur un ensemble de données coriace constitué de photos prises par les voitures de Google. Ce système a été déployé avec succès chez Google pour lire environ cent million d’adresses de maisons.

Gradient-based reinforcement learning techniques for underwater robotics behavior learning

Darrerament, l'interès pel desenvolupament d'aplicacions amb robots submarins autònoms (AUV) ha crescut de forma considerable. Els AUVs són atractius gràcies al seu tamany i el fet que no necessiten un operador humà per pilotar-los. Tot i això, és impossible comparar, en termes d'eficiència i flexibilitat, l'habilitat d'un pilot humà amb les escasses capacitats operatives que ofereixen els AUVs actuals. L'utilització de AUVs per cobrir grans àrees implica resoldre problemes complexos, especialment si es desitja que el nostre robot reaccioni en temps real a canvis sobtats en les condicions de treball. Per aquestes raons, el desenvolupament de sistemes de control autònom amb l'objectiu de millorar aquestes capacitats ha esdevingut una prioritat. Aquesta tesi tracta sobre el problema de la presa de decisions utilizant AUVs. El treball presentat es centra en l'estudi, disseny i aplicació de comportaments per a AUVs utilitzant tècniques d'aprenentatge per reforç (RL). La contribució principal d'aquesta tesi consisteix en l'aplicació de diverses tècniques de RL per tal de millorar l'autonomia dels robots submarins, amb l'objectiu final de demostrar la viabilitat d'aquests algoritmes per aprendre tasques submarines autònomes en temps real. En RL, el robot intenta maximitzar un reforç escalar obtingut com a conseqüència de la seva interacció amb l'entorn. L'objectiu és trobar una política òptima que relaciona tots els estats possibles amb les accions a executar per a cada estat que maximitzen la suma de reforços totals. Així, aquesta tesi investiga principalment dues tipologies d'algoritmes basats en RL: mètodes basats en funcions de valor (VF) i mètodes basats en el gradient (PG). Els resultats experimentals finals mostren el robot submarí Ictineu en una tasca autònoma real de seguiment de cables submarins. Per portar-la a terme, s'ha dissenyat un algoritme anomenat mètode d'Actor i Crític (AC), fruit de la fusió de mètodes VF amb tècniques de PG.

Annual Report of Nebraska's Reading First

Thi report summarizes the results of the first year of the Reading First Initiative in Nebraska. Results show that teachers and schools have made a real effort to change. Student performance in the earlier grades has shown great promise for the following years, however, growth in fluency and comprehension in grades 2 and 3 were not as impressive and require additional attention. Overall, students make at least a year’s progress in most schools and most demographic groups. Schools can make much better use of the data they were collecting and need further direction in this area The assessment results were triangulated by observations in the classrooms, interviews, and teacher responses to professional development- teachers know how to teach PA and the alphabetic principle but were still struggling with: finding time and effective strategies for fluency training teaching comprehension strategies teaching self monitoring. Growing gaps for SPED, Ethnic minorities, and ELL students suggest an emphasis on the secondary and tertiary levels of intervention in the schools.

Visual scanning in sports actions: comparison between soccer goalkeepers and judo fighters

Visual search and oculomotor behaviour are believed to be very relevant for athlete performance, especially for sports requiring refined visuo-motor coordination skills. Modern coaches believe that a correct visuo-motor strategy may be part of advanced training programs. In this thesis two experiments are reported in which gaze behaviour of expert and novice athletes were investigated while they were doing a real sport specific task. The experiments concern two different sports: judo and soccer. In each experiment, number of fixations, fixation locations and mean fixation duration (ms) were considered. An observational analysis was done at the end of the paper to see perceptual differences between near and far space. Purpose: The aim of the judo study was to delineate differences in gaze behaviour characteristics between a population of athletes and one of non athletes. Aspects specifically investigated were: search rate, search order and viewing time across different conditions in a real-world task. The second study was aimed at identifying gaze behaviour in varsity soccer goalkeepers while facing a penalty kick executed with instep and inside foot. Then an attempt has been done to compare the gaze strategies of expert judoka and soccer goalkeepers in order to delineate possible differences related to the different conditions of reacting to events occurring in near (peripersonal) or far (extrapersonal) space. Judo Methods: A sample of 9 judoka (black belt) and 11 near judoka (white belt) were studied. Eye movements were recorded at 500Hz using a video based eye tracker (EyeLink II). Each subject participated in 40 sessions for about 40 minutes. Gaze behaviour was considered as average number of locations fixated per trial, the average number of fixations per trial, and mean fixation duration. Soccer Methods: Seven (n = 7) intermediate level male volunteered for the experiment. The kickers and goalkeepers, had at least varsity level soccer experience. The vision-in-action (VIA) system (Vickers 1996; Vickers 2007) was used to collect the coupled gaze and motor behaviours of the goalkeepers. This system integrated input from a mobile eye tracking system (Applied Sciences Laboratories) with an external video of the goalkeeper’s saving actions. The goalkeepers took 30 penalty kicks on a synthetic pitch in accordance with FIFA (2008) laws. Judo Results: Results indicate that experts group differed significantly from near expert for fixations duration, and number of fixations per trial. The expert judokas used a less exhaustive search strategy involving fewer fixations of longer duration than their novice counterparts and focused on central regions of the body. The results showed that in defence and attack situation expert group did a greater number of transitions with respect to their novice counterpart. Soccer Results: We found significant main effect for the number of locations fixated across outcome (goal/save) but not for foot contact (instep/inside). Participants spent more time fixating the areas in instep than inside kick and in goal than in save situation. Mean and standard error in search strategy as a result of foot contact and outcome indicate that the most gaze behaviour start and finish on ball interest areas. Conclusions: Expert goalkeepers tend to spend more time in inside-save than instep-save penalty, differences that was opposite in scored penalty kick. Judo results show that differences in visual behaviour related to the level of expertise appear mainly when the test presentation is continuous, last for a relatively long period of time and present a high level of uncertainty with regard to the chronology and the nature of events. Expert judoist performers “anchor” the fovea on central regions of the scene (lapel and face) while using peripheral vision to monitor opponents’ limb movements. The differences between judo and soccer gaze strategies are discussed on the light of physiological and neuropsychological differences between near and far space perception.

Auditory-visual speech integration by prelinguistic infants: Perception of an emergent consonant in the McGurk effect

The McGurk effect, in which auditory [ba] dubbed onto [go] lip movements is perceived as da or tha, was employed in a real-time task to investigate auditory-visual speech perception in prelingual infants. Experiments 1A and 1B established the validity of real-time dubbing for producing the effect. In Experiment 2, 4(1)/(2)-month-olds were tested in a habituation-test paradigm, in which 2 an auditory-visual stimulus was presented contingent upon visual fixation of a live face. The experimental group was habituated to a McGurk stimulus (auditory [ba] visual [ga]), and the control group to matching auditory-visual [ba]. Each group was then presented with three auditory-only test trials, [ba], [da], and [deltaa] (as in then). Visual-fixation durations in test trials showed that the experimental group treated the emergent percept in the McGurk effect, [da] or [deltaa], as familiar (even though they had not heard these sounds previously) and [ba] as novel. For control group infants [da] and [deltaa] were no more familiar than [ba]. These results are consistent with infants'perception of the McGurk effect, and support the conclusion that prelinguistic infants integrate auditory and visual speech information. (C) 2004 Wiley Periodicals, Inc.

Causal attributions for success or failure by passing and failing students in College Algebra

Success in mathematics has been identified as a predictor of baccalaureate degree completion. Within the coursework of college mathematics, College Algebra has been identified as a high-risk course due to its low success rates. ^ Research in the field of attribution theory and academic achievement suggests a relationship between a student's attributional style and achievement. Theorists and researchers contend that attributions influence individual reactions to success and failure. They also report that individuals use attributions to explain and justify their performance. Studies in mathematics education identify attribution theory as the theoretical orientation most suited to explain academic performance in mathematics. This study focused on the relationship among a high risk course, low success rates, and attribution by examining the difference in the attributions passing and failing students gave for their performance in College Algebra. ^ The methods for the study included a pilot administration of the Causal Dimension Scale (CDSII) which was used to conduct reliability and principal component analyses. Then, students (n = 410) self-reported their performance on an in-class test and attributed their performance along the dimensions of locus of causality, stability, personal controllability, and external controllability. They also provided open-ended attribution statements to explain the cause of their performance. The quantitative data compared the passing and failing groups and their attributions for performance on a test using One-Way ANOVA and Pearson chi square procedures. The open-ended attribution statements were coded in relation to ability, effort, task difficulty, and luck and compared using a Pearson chi square procedure. ^ The results of the quantitative data comparing passing and failing groups and their attributions along the dimensions measured by the CDSII indicated statistical significance in locus of causality, stability, and personal controllability. The results comparing the open-ended attribution statements indicated statistical significance in the categories of effort and task difficulty. ^