Search

The feedback that users provide through their choices (e.g., clicks, purchases) is one of the most common types of data readily available for training search and recommendation algorithms. However, myopically training systems based on choice data may only improve short-term engagement, but not the long-term sustainability of the platform and the long-term benefits to its users, content providers, and other stakeholders. In this paper, we thus develop a new framework in which decision makers (e.g., platform operators, regulators, users) can express long-term goals for the behavior of the platform (e.g., fairness, revenue distribution, legal requirements). These goals take the form of exposure or impact targets that go well beyond individual sessions, and we provide new control-based algorithms to achieve these goals. In particular, the controllers are designed to achieve the stated long-term goals with minimum impact on short-term engagement. Beyond the principled theoretical derivation of the controllers, we evaluate the algorithms on both synthetic and real-world data. While all controllers perform well, we find that they provide interesting trade-offs in efficiency, robustness, and the ability to plan ahead.

Research Paper for the course PSYC 2017. This paper discusses the correlation between humor and personality traits.

Additional file 2. SPIRIT 2013 checklist.

Additional file 1. Tabulated summary of feasibility outcomes.

Building systems that are good for society in the face of complex societal effects requires a dynamic approach. Recent approaches to machine learning (ML) documentation have demonstrated the promise of discursive frameworks for deliberation about these complexities. However, these developments have been grounded in a static ML paradigm, leaving the role of feedback and post-deployment performance unexamined. Meanwhile, recent work in reinforcement learning has shown that the effects of feedback and optimization objectives on system behavior can be wide-ranging and unpredictable. In this paper we sketch a framework for documenting deployed and iteratively updated learning systems, which we call Reward Reports. Taking inspiration from various contributions to the technical literature on reinforcement learning, we outline Reward Reports as living documents that track updates to design choices and assumptions behind what a particular automated system is optimizing for. They are intended to track dynamic phenomena arising from system deployment, rather than merely static properties of models or data. After presenting the elements of a Reward Report, we discuss a concrete example: Meta's BlenderBot 3 chatbot. Several others for game-playing (DeepMind's MuZero), content recommendation (MovieLens), and traffic control (Project Flow) are included in the appendix.

In this work, we consider how preference models in interactive recommendation systems determine the availability of content and users' opportunities for discovery. We propose an evaluation procedure based on stochastic reachability to quantify the maximum probability of recommending a target piece of content to an user for a set of allowable strategic modifications. This framework allows us to compute an upper bound on the likelihood of recommendation with minimal assumptions about user behavior. Stochastic reachability can be used to detect biases in the availability of content and diagnose limitations in the opportunities for discovery granted to users. We show that this metric can be computed efficiently as a convex program for a variety of practical settings, and further argue that reachability is not inherently at odds with accuracy. We demonstrate evaluations of recommendation algorithms trained on large datasets of explicit and implicit ratings. Our results illustrate how preference models, selection rules, and user interventions impact reachability and how these effects can be distributed unevenly., Comment: to appear ICML 2021

Modern nonlinear control theory seeks to develop feedback controllers that endow systems with properties such as safety and stability. The guarantees ensured by these controllers often rely on accurate estimates of the system state for determining control actions. In practice, measurement model uncertainty can lead to error in state estimates that degrades these guarantees. In this paper, we seek to unify techniques from control theory and machine learning to synthesize controllers that achieve safety in the presence of measurement model uncertainty. We define the notion of a Measurement-Robust Control Barrier Function (MR-CBF) as a tool for determining safe control inputs when facing measurement model uncertainty. Furthermore, MR-CBFs are used to inform sampling methodologies for learning-based perception systems and quantify tolerable error in the resulting learned models. We demonstrate the efficacy of MR-CBFs in achieving safety with measurement model uncertainty on a simulated Segway system.

In order to certify performance and safety, feedback control requires precise characterization of sensor errors. In this paper, we provide guarantees on such feedback systems when sensors are characterized by solving a supervised learning problem. We show a uniform error bound on nonparametric kernel regression under a dynamically-achievable dense sampling scheme. This allows for a finite-time convergence rate on the sub-optimality of using the regressor in closed-loop for waypoint tracking. We demonstrate our results in simulation with simplified unmanned aerial vehicle and autonomous driving examples., Comment: to appear at L4DC 2021

RAM Evidence Briefing paper. (PDF 829 kb)

Motivated by vision-based control of autonomous vehicles, we consider the problem of controlling a known linear dynamical system for which partial state information, such as vehicle position, is extracted from complex and nonlinear data, such as a camera image. Our approach is to use a learned perception map that predicts some linear function of the state and to design a corresponding safe set and robust controller for the closed loop system with this sensing scheme. We show that under suitable smoothness assumptions on both the perception map and the generative model relating state to complex and nonlinear data, parameters of the safe set can be learned via appropriately dense sampling of the state space. We then prove that the resulting perception-control loop has favorable generalization properties. We illustrate the usefulness of our approach on a synthetic example and on the self-driving car simulation platform CARLA., Comment: This revision includes reframing the local generalization problem, with relaxed the assumptions so that the robust problem depends on a local slope bound rather than a Lipschitz constant, and provide a method for learning the slope bound from data. We also include additional experiments with a CNN perception module

Patient-related measures. (PDF 446 kb)

Exemplar of GP participant information sheet and consent form. (PDF 127 kb)

Additional trial information (consistent with SPIRIT 2013 checklist) not included in the main protocol document. (PDF 344 kb)

Behaviour change techniques employed in FREE in relation to general practitioners. (PDF 413 kb)

Completed PRISMA-P+ checklist. (DOCX 29 kb)

Teacher interview schedule. (DOCX 16 kb)

Exemplar of patient participant information sheets and consent form. (PDF 173 kb)

General practitioner-related measures. (PDF 418 kb)

Criteria for quality appraisal. (DOCX 15 kb)

Behaviour change techniques employed in FREE in relation to patients. (PDF 409 kb)

Fidelity to delivery (quality) checklist. (DOCX 16 kb)

Child focus group schedule. (DOCX 16 kb)

SPIRIT checklist. (PDF 277 kb)

Search strategy. (DOCX 15 kb)

Fidelity to delivery (content) checklist. (DOCX 13 kb)

Purpose: Upper limb disability following stroke may have multiple effects on the individual. Existing assessment instruments tend to focus on impairment and function and may miss other changes that are personally important. This study aimed to identify personally significant impacts of upper limb disability following stroke. Methods: Accounts by stroke survivors, in the form of web-based diaries (blogs) and stories, were sought using a blog search engine and in stroke-related web-sites. Thematic analysis using the World Health Organisation’s International Classification of Functioning Disability and Health (ICF) was used to identify personal impacts of upper limb disability following stroke. Results: Ninety-nine sources from at least four countries were analysed. Many impacts were classifiable using the ICF, but a number of additional themes emerged, including emotional, cognitive and behavioural changes. Blogs and other web-based accounts were easily accessible and rich sources of data, although using them raised several methodological issues, including potential sample bias. Conclusions: A range of impacts was identified, some of which (such as use of information technology and alienation from the upper limb) are not addressed in current assessment instruments. They should be considered in post-stroke assessments. Blogs may help in the development of more comprehensive assessments.Implications for RehabilitationA comprehensive assessment of the upper limb following stroke should include the impact of upper limb problems on social participation, as well as associated emotional, cognitive and behavioural changes.Using personalised assessment instruments alongside standardised measures may help ensure that these broader domains are considered in discussions between clinicians and patients.Rehabilitation researchers should investigate whether and how these domains could be addressed and operationalised in standard upper limb assessment instruments. A comprehensive assessment of the upper limb following stroke should include the impact of upper limb problems on social participation, as well as associated emotional, cognitive and behavioural changes. Using personalised assessment instruments alongside standardised measures may help ensure that these broader domains are considered in discussions between clinicians and patients. Rehabilitation researchers should investigate whether and how these domains could be addressed and operationalised in standard upper limb assessment instruments.

Background: Promoting physical activity (PA) via primary care exercise referral schemes (ERS) is common but there is no rigorous evidence for long term changes in PA (Pavey et al, 2011) among those with chronic conditions. From July 2015, for 15 months, the e-coachER trial began to recruit 1400 patients (in SW England, Birmingham and Glasgow) with one or more chronic conditions including diabetes, obesity, hypertension, osteoarthritis, or depression, who are eligible and about to attend an ERS. The two-arm parallel RCT is powered to determine if the addition of a web-based, interactive, theory-driven and evidence-based support system called e-coachER (hosted on the ‘LifeGuide’ platform) will result in at least 10% more patients who do 150 mins or more per week of accelerometer assessed moderate or vigorous physical activity (MVPA) at 12 months. Recruitment into the trial is within primary care, using both mail-merged patient invitations and opportunistic GP invitations (and exercise referrals). Within the trial, after participants are screened, provide consent and complete baseline assessments, they are randomised to receive usual ERS at each site or usual ERS plus a mailed Welcome Pack with registration details to access e-coachER on-line. Inclusion criteria for entering the trial are: (1) Aged 16-74 years; (2) with one or more of the following: obesity (BMI 30-35), hypertension (SBP 140-179 or DBP 90-109, type 2 diabetes, lower limb osteoarthritis, recent history of treatment for depression; (3) Participants who are in the two lowest (of four) groups using the GP Physical Activity Questionnaire; (4) have an e-mail address and access to the internet; (5) Eligible for an ERS. The intervention rationale, design and content are reported in another presentation. Aims: This presentation will provide initial findings from a 3 month internal pilot phase with a focus on trial recruitment and initial intervention engagement. We will present data on the achievement of our targets for recruiting 180 participants across the three sites and proportion who initially go online to register and use e-coachER. We also aim, based on detailed process evaluation work, to present a summary of factors influencing recruitment and intervention engagement, and any adaptations needed for the recruitment process. Methods: Data is being captured to inform a CONSORT chart with details on number of patient invitations, number and reasons for trial exclusion, and numbers and characteristics of those recruited. The LifeGuide system will provide data on numbers/proportion assigned to the intervention arm who register and start to use e-coachER. Qualitative data on recruitment processes and intervention engagement, as part of the process evaluation, will be summarised. Results: Due to a number of factors recruitment began 2-4 months later than planned across the 3 sites. Data will be presented on achieved recruitment and e-coachER engagement relative to our targets in the internal pilot trial. Qualitative data will be presented on the acceptability and feasibility of the trial methods and intervention. Conclusion: A summary of what we have learned about recruitment and intervention engagement will be presented and ideas for improvement, as appropriate. Supported with funding from the NIHR (HTA) Project: 13/25/20

status: published