Learning the learning rate for prediction with expert advice

Most standard algorithms for prediction with expert advice depend on a parameter called the learning rate. This learning rate needs to be large enough to fit the data well, but small enough to prevent overfitting. For the exponential weights algorithm, a sequence of prior work has established theoretical guarantees for higher and higher data-dependent tunings of the learning rate, which allow for increasingly aggressive learning. But in practice such theoretical tunings often still perform worse (as measured by their regret) than ad hoc tuning with an even higher learning rate. To close the gap between theory and practice we introduce an approach to learn the learning rate. Up to a factor that is at most (poly)logarithmic in the number of experts and the inverse of the learning rate, our method performs as well as if we would know the empirically best learning rate from a large range that includes both conservative small values and values that are much higher than those for which formal guarantees were previously available. Our method employs a grid of learning rates, yet runs in linear time regardless of the size of the grid.

Prior and posterior linear pooling for combining expert opinions: Uses and impact on Bayesian Networks

The use of expert knowledge to quantify a Bayesian Network (BN) is necessary when data is not available. This however raises questions regarding how opinions from multiple experts can be used in a BN. Linear pooling is a popular method for combining probability assessments from multiple experts. In particular, Prior Linear Pooling (PrLP), which pools opinions then places them into the BN is a common method. This paper firstly proposes an alternative pooling method, Posterior Linear Pooling (PoLP). This method constructs a BN for each expert, then pools the resulting probabilities at the nodes of interest. Secondly, it investigates the advantages and disadvantages of using these pooling methods to combine the opinions of multiple experts. Finally, the methods are applied to an existing BN, the Wayfinding Bayesian Network Model, to investigate the behaviour of different groups of people and how these different methods may be able to capture such differences. The paper focusses on 6 nodes Human Factors, Environmental Factors, Wayfinding, Communication, Visual Elements of Communication and Navigation Pathway, and three subgroups Gender (female, male),Travel Experience (experienced, inexperienced), and Travel Purpose (business, personal) and finds that different behaviors can indeed be captured by the different methods.

Submission to KPMG review of the Federal Government Initiative to “analyse dementia program and services funded by the Department of Social Services"

The comments I make are based on my nearly twenty years involvement in the dementia cause at both a national and international level. In preparation, I read two papers namely the Ministerial Dementia Forum – Option Paper produced by KPMG Management Consultants (2014) and Analysis of Dementia Programmes and Services Funded by the Department of Social Services: Conversation Starter prepared by KPMG as a preparation document for those attending a workshop in Brisbane on April 22nd 2015. Dementia is a complex “syndrome” and as is often said, “when you meet one person with dementia, you have met one” meaning that no two persons with dementia are the same. Even in dementia care, Australia is a “lucky country” and there is much to be said for the quality and diversity of dementia care available for people living with dementia. Despite this, I agree with the many views expressed in the material I read that there is scope for improvement, especially in the way that services are coordinated. In saying that, I do not purport to have all the solutions nor claim to have the knowledge required to comment on all the programs covered by this review. If I appear to be a “biased” advocate for Alzheimer’s Australia across the States and Territories, it is because I have seen constant evidence of ordinary people doing extraordinary things with inadequate resources. Dementia care is not cheap and if those funding dementia services are primarily only interested in economic outcomes and benefits, the real purpose of this consultation will be defeated. In addition, nowhere in the material I have read is there any recognition that in many instances program funding is a complex mix of government (at all levels) and private funding. This makes reviewing those programs more complex and less able to be coordinated at a Departmental level. It goes without saying therefore that the Federal Government is not” the only player in this game”. Of all those participating in this review, Alzheimer’s Australia is best placed to comment on programs as it is more connected to people living with dementia and has probably the best record of consulting with them. It would appear however that their role has been reduced to that of a “bit player”. Without wanting to be critical, the Forum Report which deals with the comments made at a gathering of 70 individuals and organisations, only three (3) or 4.28% were actual carers of people living with dementia. Even if it is argued that a number of organisations present represented consumers, the percentage goes up only marginally to 8.57% which is hardly an endorsement of the forum being “consumer driven”. The predominance of those present were service providers, each with their own agenda and each seeking advantage for their “business”. The final point I want to make before commenting on more specific, program related issues, is that many programs being reviewed have a much longer history than is reflected in the material I have read. Their growth and development was pioneered by Alzheimer’s Australia organisations across the country often with no government funding. Attempts to bring about better coordination of programs were often at the behest of Alzheimer’s Australia but in the main were ignored. The opportunity to now put this right is long overdue.

Adapting to concurrent expert evidence in medical litigation

In medical negligence litigation expert evidence has long played a dominant role. The trend towards the use of concurrent expert evidence is now well underway. However, for the lawyers and the doctors involved, the pathway is not yet familiar. Disputes have frequently arisen in the context of pre-hearing expert conclaves, given the adversarial nature of litigation and perhaps fuelled by fears of a less transparent process at this increasingly important stage. This article explains the concurrent expert evidence framework and examines areas of common dispute both in the conclaves and at trial, with a view to providing assistance to legal practitioners working in this area and the medical practitioners called upon to provide expert evidence in such litigation.

Incentivizing innovation: A review of the Brazilian federal innovation support programs

Innovation is the transformation of knowledge of any kind into new products or services in the market. Its importance as a production factor is widely acknowledged. In the age of the knowledge-based economy innovation became critical for any company or even country to compete globally. Many countries are encouraging innovation through various mechanisms, and one of the most widely used is the provision of special incentives for innovation. This paper investigates incentive systems for the growth of technology companies as a strategy to promote knowledge-based economic development. As for the case investigations the study focuses on an emerging economy, Brazil. The research is based upon the available literature, best practices, government policy and review of incentive systems. The findings provide insights from the case study in a country context and some lessons learned for other countries using incentive systems to boost the innovation capabilities of their technology companies.

Global standard for the composition of infant formula: Recommendations of an ESPGHAN coordinated international expert group

The Codex Alimentarius Commission of the Food and Agriculture Organization of the United Nations (FAO) and the World Health Organization (WHO) develops food standards, guidelines and related texts for protecting consumer health and ensuring fair trade practices globally. The major part of the world's population lives in more than 160 countries that are members of the Codex Alimentarius. The Codex Standard on Infant Formula was adopted in 1981 based on scientific knowledge available in the 1970s and is currently being revised. As part of this process, the Codex Committee on Nutrition and Foods for Special Dietary Uses asked the ESPGHAN Committee on Nutrition to initiate a consultation process with the international scientific community to provide a proposal on nutrient levels in infant formulae, based on scientific analysis and taking into account existing scientific reports on the subject. ESPGHAN accepted the request and, in collaboration with its sister societies in the Federation of International Societies on Pediatric Gastroenterology, Hepatology and Nutrition, invited highly qualified experts in the area of infant nutrition to form an International Expert Group (IEG) to review the issues raised. The group arrived at recommendations on the compositional requirements for a global infant formula standard which are reported here.

Social media in selected Australian federal and state election campaigns, 2010-15

Campaigning in Australian election campaigns at local, state, and federal levels is fundamentally affected by the fact that voting is compulsory in Australia, with citizens who are found to have failed to cast their vote subject to fines. This means that - contrary to the situation in most other nations – elections are decided not by which candidate or party has managed to encourage the largest number of nominal supporters to make the effort to cast their vote, but by some 10-20% of genuine ‘swinging voters’ who change their party preferences from one election to the next. Political campaigning is thus aimed less at existing party supporters (so-called ‘rusted on’ voters whose continued support for the party is essentially taken for granted) than at this genuinely undecided middle of the electorate. Over the past decades, this has resulted in a comparatively timid, vague campaigning style from both major party blocs (the progressive Australian Labor Party [ALP] and the conservative Coalition of the Liberal and National Parties [L/NP]). Election commitments that run the risk of being seen as too partisan and ideological are avoided as they could scare away swinging voters, and recent elections have been fought as much (or more) on the basis of party leaders’ perceived personas as they have on stated policies, even though Australia uses a parliamentary system in which the Prime Minister and state Premiers are elected by their party room rather than directly by voters. At the same time, this perceived lack of distinctiveness in policies between the major parties has also enabled the emergence of new, smaller parties which (under Australia’s Westminster-derived political system) have no hope of gaining a parliamentary majority but could, in a close election, come to hold the balance of power and thus exert disproportionate influence on a government which relies on their support.

Assessing performance in critical care nursing practice using the Benner Novice to expert model: A pilot study

Background It is often believed that by ensuring the ongoing completion of competency documents and life-long learning in nursing practice guarantees quality patient care. This is probably true in most cases where it provides reassurances that the nursing team is maintaining a safe “generalised” level of practice. However, competency does not always promise quality performance. There are a number of studies that have reported differences in what practitioners know and what they actually do despite being deemed competent. Aim The aim of this study was to assess whether our current competency documentation is fit for purpose and to ascertain whether performance assessment needs to be a key component in determining competence. Method 15 nurses within a General ICU who had been on the unit <4 years agreed to participate in this project. Using participant observation and assessing performance against key indicators of the Benner Novice to Expert5 model the participants were supported and assessed over the course of a ‘normal’ nursing shift. Results The results were surprising both positively and negatively. First, the nurses felt more empowered in their clinical decision making skills; second, it identified individual learning needs and milestones in educational development. There were some key challenges identified which included 5 nurses over estimating their level of competence, practice was still very much focused on task acquisition and skill and surprisingly some nurses still felt dominated by the other health professionals within the unit. Conclusion We found that the capacity and capabilities of our nursing workforce needs continual ongoing support especially if we want to move our staff from capable task-doer to competent performers. Using the key novice to expert indicators identified the way forward for us in how we assess performance and competence in practice particularly where promotion to higher grades is based on existing documentation.

Perceptions on the causes of individual and fraudulent co-offending: Views of forensic accountants

Individual and/or co-offenders fraudulent activities can have a devastating effect on a company’s reputation and credibility. Enron, Xerox, WorldCom, HIH Insurance and One.Tel are examples where stakeholders incurred substantial financial losses as a result of fraud and led to a loss of confidence in corporate dealings by the public in general. There are numerous theoretical approaches that attempt to explain how and why fraudulent acts occur, drawing on the fields of sociology, organisational, management and economic literature, but there is limited empirical evidence published in accounting literature. This qualitative inductive study analyses perceptions and experiences of forensic accountants to gain insights into individual fraud and co-offending in order to determine whether the conceptual framework developed from literature accurately depicts the causes of fraud committed by individuals and groups in the twenty-first century. Findings from the study both support and extend the conceptual framework, demonstrating that strain and anomie can result in fraud, that deviant sub-groups recruit and coerce members by providing relief from strain, and that inadequate corporate governance mechanisms both contribute to fraud occurring, and provide the opportunity for fraudulent activities to be executed and often remain undetected. Additional factors emerging from this study (the ‘technoconomy’, addiction and IT measures) were also identified as contributors to fraud, particularly relevant to the twenty-first century, and consequently, a refined conceptual framework is presented in the discussion and conclusion to the paper.

Eliciting and encoding expert knowledge on variable selection into classical or Bayesian species distribution models

The quality of species distribution models (SDMs) relies to a large degree on the quality of the input data, from bioclimatic indices to environmental and habitat descriptors (Austin, 2002). Recent reviews of SDM techniques, have sought to optimize predictive performance e.g. Elith et al., 2006. In general SDMs employ one of three approaches to variable selection. The simplest approach relies on the expert to select the variables, as in environmental niche models Nix, 1986 or a generalized linear model without variable selection (Miller and Franklin, 2002). A second approach explicitly incorporates variable selection into model fitting, which allows examination of particular combinations of variables. Examples include generalized linear or additive models with variable selection (Hastie et al. 2002); or classification trees with complexity or model based pruning (Breiman et al., 1984, Zeileis, 2008). A third approach uses model averaging, to summarize the overall contribution of a variable, without considering particular combinations. Examples include neural networks, boosted or bagged regression trees and Maximum Entropy as compared in Elith et al. 2006. Typically, users of SDMs will either consider a small number of variable sets, via the first approach, or else supply all of the candidate variables (often numbering more than a hundred) to the second or third approaches. Bayesian SDMs exist, with several methods for eliciting and encoding priors on model parameters (see review in Low Choy et al. 2010). However few methods have been published for informative variable selection; one example is Bayesian trees (O’Leary 2008). Here we report an elicitation protocol that helps makes explicit a priori expert judgements on the quality of candidate variables. This protocol can be flexibly applied to any of the three approaches to variable selection, described above, Bayesian or otherwise. We demonstrate how this information can be obtained then used to guide variable selection in classical or machine learning SDMs, or to define priors within Bayesian SDMs.

Conclaves and concurrent expert evidence: A positive development in Australian legal practice?

Many Australian courts now prefer pre-hearing meetings of experts (conclaves) being convened to prepare joint reports to identify areas of agreement and disagreement, followed by concurrent expert evidence at trial. This contrasts to the traditional approach where experts did not meet before trial and did not give evidence together. Most judges, lawyers and expert witnesses favour this as a positive development in Australian legal practice, at least for civil disputes. This new approach impacts medical practitioners who are called upon to give expert evidence, or who are parties to disputes before the courts. Arguably, it is too soon to tell whether the relative lack of transparency at the conclave stage will give rise to difficulties in the coronial, disciplinary and criminal arenas.

Policing expert testimony in a death investigation: Medical opinion as legal fact

Within coronial investigations, pathologists are called upon to given evidence as to cause of death. This evidence is given great weight by the coroners; after all, scientific ‘truth’ is widely deemed to be far more reliable than legal ‘opinion’. The purpose of this paper is to examine the ontological and epistemological status of that evidence, from the perspectives of both the pathologists and the coroners. As part of an Australian Research Council Linkage Grant, interviews were conducted with seven pathologists and 10 coroners from within the Queensland coronial system. Contrary to expectations, and the work of philosophers of science, such as Feyerabend (1975), pathologists did not present their findings in terms of unequivocal facts or objective truths relating to causes of death. Rather, their evidence was largely presented as ‘educated opinion’ based upon ‘the weight of evidence’. It was actually the coroners who translated that opinion into ‘medical fact’ within the proceedings of their death investigations, arguably as a consequence of the administrative necessity to reach a clear-cut finding as to cause of death, and on the basis of their own understanding of the ontology of medical knowledge. These findings support Latour’s (2010) claim that law requires a fundamentally different epistemology to science, and that science is not entirely to blame for the extravagant truth-claims made on its behalf