Implicit online learning

Online learning algorithms have recently risen to prominence due to their strong theoretical guarantees and an increasing number of practical applications for large-scale data analysis problems. In this paper, we analyze a class of online learning algorithms based on fixed potentials and nonlinearized losses, which yields algorithms with implicit update rules. We show how to efficiently compute these updates, and we prove regret bounds for the algorithms. We apply our formulation to several special cases where our approach has benefits over existing online learning methods. In particular, we provide improved algorithms and bounds for the online metric learning problem, and show improved robustness for online linear prediction problems. Results over a variety of data sets demonstrate the advantages of our framework.

Matrix regularization techniques for online multitask learning

In this paper we examine the problem of prediction with expert advice in a setup where the learner is presented with a sequence of examples coming from different tasks. In order for the learner to be able to benefit from performing multiple tasks simultaneously, we make assumptions of task relatedness by constraining the comparator to use a lesser number of best experts than the number of tasks. We show how this corresponds naturally to learning under spectral or structural matrix constraints, and propose regularization techniques to enforce the constraints. The regularization techniques proposed here are interesting in their own right and multitask learning is just one application for the ideas. A theoretical analysis of one such regularizer is performed, and a regret bound that shows benefits of this setup is reported.

Adaptive online gradient descent

We study the rates of growth of the regret in online convex optimization. First, we show that a simple extension of the algorithm of Hazan et al eliminates the need for a priori knowledge of the lower bound on the second derivatives of the observed functions. We then provide an algorithm, Adaptive Online Gradient Descent, which interpolates between the results of Zinkevich for linear functions and of Hazan et al for strongly convex functions, achieving intermediate rates between [square root T] and [log T]. Furthermore, we show strong optimality of the algorithm. Finally, we provide an extension of our results to general norms.

REGAL : a regularization based algorithm for reinforcement learning in weakly communicating MDPs

We provide an algorithm that achieves the optimal regret rate in an unknown weakly communicating Markov Decision Process (MDP). The algorithm proceeds in episodes where, in each episode, it picks a policy using regularization based on the span of the optimal bias vector. For an MDP with S states and A actions whose optimal bias vector has span bounded by H, we show a regret bound of ~ O(HS p AT ). We also relate the span to various diameter-like quantities associated with the MDP, demonstrating how our results improve on previous regret bounds.

Walking when intoxicated : an investigation of the factors which influence individuals' drink walking intentions

Annually, in Australia, 10-15% of all road-related fatalities involve pedestrians. Of those pedestrians fatally injured, approximately 45% were walking while intoxicated or 'drink walking'. Drink walking is increasing in prevalence and younger persons may be especially prone to engage in this behaviour and, thus, are at heightened risk of being injured or killed. Presently, limited research is available regarding the factors which influence individuals to drink walk. This study explored young people's (17-25 years) intentions to drink walk, using an extended Theory of Planned Behaviour (TPB). Participants (N = 215), completed a self-report questionnaire which assessed the standard TPB constructs (attitude, subjective norm, perceived behavioural control) as well as the extended constructs of risk perception, anticipated regret, and past behaviour. It was hypothesised that the standard TPB constructs would significantly predict individuals' reported intentions to drink walk and that the additional constructs would predict intentions over and above the TPB constructs. The TPB variables significantly predicted 63.2% of the variance in individuals' reported intentions to drink walk, and the additional variables, combined, explained a further 6.1% of the variance. Of the additional constructs, anticipated regret and past behaviour, but not risk perception, were significant predictors of drink walking intentions. As one of the first studies to provide a theoretically-based investigation of factors influencing individuals' drink walking intentions, the current study's findings have potentially significant implications for understanding young people's decisions to drink walk and the design of future countermeasures to ultimately reduce this behaviour.

Emotions, crime and justice [Onati International Series in Law and Society, Number 1]

The return of emotions to debates about crime and criminal justice has been a striking development of recent decades across many jurisdictions. This has been registered in the return of shame to justice procedures, a heightened focus on victims and their emotional needs, fear of crime as a major preoccupation of citizens and politicians, and highly emotionalised public discourses on crime and justice. But how can we best make sense of these developments? Do we need to create "emotionally intelligent" justice systems, or are we messing recklessly with the rational foundations of liberal criminal justice? This volume brings together leading criminologists and sociologists from across the world in a much needed conversation about how to re-calibrate reason and emotion in crime and justice today. The contributions range from the micro-analysis of emotions in violent encounters to the paradoxes and tensions that arise from the emotionalisation of criminal justice in the public sphere. They explore the emotional labour of workers in police and penal institutions, the justice experiences of victims and offenders, and the role of vengeance, forgiveness and regret in the aftermath of violence and conflict resolution. The result is a set of original essays which offer a fresh and timely perspective on problems of crime and justice in contemporary liberal democracies.

Two-stage power network reconfiguration strategy considering node importance and restored generation capacity

Network reconfiguration after complete blackout of a power system is an essential step for power system restoration. A new node importance evaluation method is presented based on the concept of regret, and maximisation of the average importance of a path is employed as the objective of finding the optimal restoration path. Then, a two-stage method is presented to optimise the network reconfiguration strategy. Specifically, the restoration sequence of generating units is first optimised so as to maximise the restored generation capacity, then the optimal restoration path is selected to restore the generating nodes concerned and the issues of selecting a serial or parallel restoration mode and the reconnecting failure of a transmission line are next considered. Both the restoration path selection and skeleton-network determination are implemented together in the proposed method, which overcomes the shortcoming of separate decision-making in the existing methods. Finally, the New England 10-unit 39-bus power system and the Guangzhou power system in South China are employed to demonstrate the basic features of the proposed method.

Identifying the determinants of concealed and obvious texting while driving : are they distinct behaviours?

Concealed texting (CT) while driving involves a conscious effort to hide one’s texting while obvious texting (OT) does not involve such efforts to conceal the behaviour. Young drivers are the most frequent users of mobile phones while driving which is associated with heightened crash risk. This study investigated the extent to which CT and OT may be discrete behaviours to ascertain whether countermeasures would need to utilise distinct approaches. An extended Theory of Planned Behaviour (TPB) including moral norm, mobile phone involvement, and anticipated regret guided the research. Participants (n = 171) were aged 17 to 25 years, owned a mobile phone, had a current driver’s licence, and resided in Queensland. A repeated measures MANOVA found significant differences between CT and OT on all standard and extended TPB constructs. Hierarchical multiple regression analyses showed the standard TPB constructs accounted for 68.7% and 54.6% of the variance in intentions to engage in CT and OT, respectively. The extended predictors contributed additional variance in intentions over and above the standard TPB constructs. Further, in the final regression model, differences emerged in the significant predictors of each type of texting. These findings provide initial evidence that CT and OT are distinct behaviours. This distinction is important to the extent that it may influence the nature of advertising countermeasures aimed at reducing/preventing young drivers’ engagement in these risky behaviours.

Concealing their communication : exploring psychosocial predictors of young drivers’ intentions and engagement in concealed texting

Making a conscious effort to hide the fact that you are texting while driving (i.e., concealed texting) is a deliberate and risky behaviour involving attention diverted away from the road. As the most frequent users of text messaging services and mobile phones while driving, young people appear at heightened risk of crashing from engaging in this behaviour. This study investigated the phenomenon of concealed texting while driving, and utilised an extended Theory of Planned Behaviour (TPB) including the additional predictors of moral norm, mobile phone involvement, and anticipated regret to predict young drivers’ intentions and subsequent behaviour. Participants (n = 171) were aged 17 to 25 years, owned a mobile phone, and had a current driver’s licence. Participants completed a questionnaire measuring their intention to conceal texting while driving, and a follow-up questionnaire a week later to report their behavioural engagement. The results of hierarchical multiple regression analyses showed overall support for the predictive utility of the TPB with the standard constructs accounting for 69% of variance in drivers’ intentions, and the extended predictors contributing an additional 6% of variance in intentions over and above the standard constructs. Attitude, subjective norm, PBC, moral norm, and mobile phone involvement emerged as significant predictors of intentions; and intention was the only significant predictor of drivers’ self-reported behaviour. These constructs can provide insight into key focal points for countermeasures including advertising and other public education strategies aimed at influencing young drivers to reconsider their engagement in this risky behaviour.

Testing an extended theory of planned behavior to predict young people's intentions to join a bone marrow donor registry

An extended theory of planned behavior (TPB) was used to understand the factors, particularly control perceptions and affective reactions, given conflicting findings in previous research, informing younger people's intentions to join a bone marrow registry. Participants (N  = 174) completed attitude, subjective norm, perceived behavioral control (PBC), moral norm, anticipated regret, self-identity, and intention items for registering. The extended TPB (except PBC) explained 67.2% of variance in intention. Further testing is needed as to the volitional nature of registering. Moral norm, anticipated regret, and self-identity are likely intervention targets for increasing younger people's bone marrow registry participation.

Child sex scandal in two countries rocks Anglican church

A CHILD sex scandal involving victims in Australia and Britain has hit the top echelon of the Anglican Church, with allegations that some of its most senior clergymen failed to respond properly to complaints of horrific abuse. The former archbishop of York, now Lord (David) Hope of Thornes, yesterday expressed regret over failing to report to police allegations in 1999 and 2003 about a former Queensland Anglican school principal, who rose to become the head of education for the church in Britain. The late reverend Robert Waddington has been accused of beating and sexually abusing students during the 1960s at St Barnabas boarding school in Ravenshoe, north Queensland, and later, when he was in charge of the choir as dean of Manchester. A joint investigation by The Australian and The Times newspaper in London has revealed that church officials, including Lord Hope, failed to report the 1999 allegations of abuse made by a former Queensland student and similar claims made in 2003 by the family of a choirboy in Manchester. The alleged victims were never told of the existence of the other allegations.

Body and forgetting

Body and Forgetting is a powerful dance performance that brings together the work of choreographer Liz Roche and film maker Alan Gilsenan, with a live score by Denis Roche. Inspired by the writings of Milan Kundera, Liz Roche Company's remarkable dancers find their way through delicately woven circumstances of disappearance, loss, relationship and hope. Their attempts to hold fast to memories and objects of meaning is at the heart of this work. The live performers move in dialogue with filmed versions of their dancing selves. They re-write their histories, make better endings to their stories, say what they regret not having said. These filmed reflections or versions of themselves, by offering a mirror, ultimately bring the performers back to themselves, richer from the experience.

Prediction with limited advice and multiarmed bandits with paid observations

We study two problems of online learning under restricted information access. In the first problem, prediction with limited advice, we consider a game of prediction with expert advice, where on each round of the game we query the advice of a subset of M out of N experts. We present an algorithm that achieves O(√(N/M)TlnN ) regret on T rounds of this game. The second problem, the multiarmed bandit with paid observations, is a variant of the adversarial N-armed bandit game, where on round t of the game we can observe the reward of any number of arms, but each observation has a cost c. We present an algorithm that achieves O((cNlnN) 1/3 T2/3+√TlnN ) regret on T rounds of this game in the worst case. Furthermore, we present a number of refinements that treat arm- and time-dependent observation costs and achieve lower regret under benign conditions. We present lower bounds that show that, apart from the logarithmic factors, the worst-case regret bounds cannot be improved.

Open problem : adversarial multiarmed bandits with limited advice

Adversarial multiarmed bandits with expert advice is one of the fundamental problems in studying the exploration-exploitation trade-o. It is known that if we observe the advice of all experts on every round we can achieve O(√KTlnN) regret, where K is the number of arms, T is the number of game rounds, and N is the number of experts. It is also known that if we observe the advice of just one expert on every round, we can achieve regret of order O(√NT). Our open problem is what can be achieved by asking M experts on every round, where 1 < M < N.

Universal codes from switching strategies

We discuss algorithms for combining sequential prediction strategies, a task which can be viewed as a natural generalisation of the concept of universal coding. We describe a graphical language based on Hidden Markov Models for defining prediction strategies, and we provide both existing and new models as examples. The models include efficient, parameterless models for switching between the input strategies over time, including a model for the case where switches tend to occur in clusters, and finally a new model for the scenario where the prediction strategies have a known relationship, and where jumps are typically between strongly related ones. This last model is relevant for coding time series data where parameter drift is expected. As theoretical contributions we introduce an interpolation construction that is useful in the development and analysis of new algorithms, and we establish a new sophisticated lemma for analysing the individual sequence regret of parameterised models.