143174 research outputs found
Sort by
Update estimation and scheduling for over-the-air federated learning with energy harvesting devices
No full textWe study over-the-air federated learning for energy harvesting devices with heterogeneous data distribution over wireless fading multiple access channels. To address the impact of low energy arrivals and data heterogeneity on global learning, we propose different user scheduling strategies. Specifically, we develop two approaches: 1) entropy-based scheduling for known data distributions, and 2) least-squares-based user representation estimation for scheduling with unknown data distributions at the parameter server. Both methods aim to select a diverse set of users to participate in the learning process, mitigating bias and enhancing convergence behavior. Numerical and analytical results demonstrate improved learning performance
Long-term sustainable development goal interactions on the road to Paris
No full textThis PhD thesis examines the interplay between the Sustainable Development Goals (SDGs) and the Paris Agreement, two key global frameworks aimed at addressing urgent social, economic, and environmental challenges. Given the complex synergies and trade-offs between these agendas, the research explores the implications of various Paris-compliant climate mitigation strategies on the SDG Agenda through three studies with different indicator scopes and regional detail.
The first study, using a global model with detailed representation of atmosphere-related SDG metrics, focuses on the potential impacts of mitigation strategies relying on technological or nature-based solutions, and demand-side behavioural changes, on a selection of energy-environment SDGs. Findings highlight that while some strategies may lead to adverse effects on food and water resources, forest cover, those involving demand-side changes, could reduce GHG emissions while simultaneously improving air quality, lowering food prices, and reducing impacts on agriculture.
The second study, using a set of harmonised models, deep dives into how EU decarbonization pathways might affect SDG performances. Results indicate that ambitious climate actions could enhance sustainability, health, and agricultural productivity, particularly in countries currently lagging in SDG performance.
The third study combines modelling outputs with a species loss model to address the intersection of climate change and biodiversity loss, evaluating the effects of land-based mitigation strategies such as afforestation and bioenergy expansion or a shift toward plant-based diets in scenarios with increasing land protected areas. The analysis reveals that while bioenergy expansion may harm biodiversity, dietary shifts could substantially enhance biodiversity gains.
Overall, this PhD thesis provides critical insights into how the pathways chosen to meet climate goals can influence broader sustainability objectives, emphasizing the need for integrated and context-specific policy approaches. In particular, it highlights that demand-side changes and compensating policies emerge as critical to retaining the integrity of Paris and the SDGs when taken together.Open Acces
Online unsupervised adaptation of latent representation for myoelectric control during user-decoder co-adaptation
No full textMyoelectric control interfaces, which map electromyographic (EMG) signals into control commands for external devices, have applications in active prosthesis control. However, the statistical characteristics of EMG signals change over time (e.g., because of changes in the electrode location), which makes interfaces based on static mapping unstable. Thus the user-decoder co-adaptation is needed during online operations. Nevertheless, current online decoder adaptation approaches present several practical challenges, such as expensive data labeling and slow convergence. Thus we introduce an unsupervised decoder adaptation method that converges rapidly. We use an autoencoder to extract motor intent representation in the latent manifold space rather than the sensor space, and further introduce an online unsupervised adaptation scheme based on Moore-Penrose Inverse, a noniterative approach suited for fast network re-training, to track the evolving manifold. A validation experiment first showed that the convergence time of the proposed adaptation scheme was reduced to about 50% of that for state-of-the-art methods. Online experiments further evaluated cursor and prosthetic hand control by the proposed myocontrol interface, where perturbations were representatively introduced by shifting the electrodes. Results showed that our scheme reached comparable improvements in robustness as supervised counterparts. Moreover, in a cup relocation test with a prosthetic hand, the completion time in the post-adaptation phase with electrode shift was comparable to that in the baseline phase without shift. These results suggest that our method effectively improves the accessibility and reliability of decoder adaptation, which has the potential to reduce the translational gap of myoelectric control interfaces by effective co-adaptation during operation
Ease of access to general practice, patient experience and use of the NHS App: an ecological analysis of English data
No full textObjective
To assess the relationship between GP experience, phone and website access measures, and NHS App use.
Design and setting
An ecological study using practice-level data from the NHS in England between March 2020 and June 2022. Patient-reported experience and ease of access scores were from the General Practice Patient Survey and NHS App data from NHS Dashboard. Practice experience and access measures were grouped into five quintiles and Incident Rate Ratios (IRR) from negative binomial regressions compared NHS App usage across these groups. Models were adjusted for age, sex, deprivation, ethnicity and long-term healthcare needs.
Participants
Patients registered at 6,386 GP practices in England.
Main Outcomes
Weekly rates of NHS App functions used (registrations, logins, prescriptions ordered, medical record views and appointments booked) per 1,000 GP registered population.
Results
Fully adjusted models found lower NHS App use in practices with the highest patient experience. Registration rates were 3.5% lower in practices with the highest vs. lowest experience scores (IRR 0.96, p<0.001) and logins were 5.2% lower (IRR 0.95, p<0.001). Practices with better phone access had 27.0% higher prescription orders (IRR highest vs. lowest=1.27, p<0.001), and 57.8% higher appointment bookings (IRR highest vs. lowest= 1.58, p<0.001). Prescriptions were 7.7% higher in practices with the highest vs. lowest web access scores (IRR 1.08, p<0.001).
Conclusion
NHS App use was lower in practices with the highest patient experience, but generally higher in practices with better phone and web access. Results highlight the need for coordinated action to improve access and patient satisfaction
Projection of cortical beta band oscillations to a motor neuron pool across the full range of recruitment
No full textCortical beta band oscillations (13–30 Hz) are associated with sensorimotor control, but their precise role remains unclear. Evidence suggests that for low-threshold motor neurons (MNs), these oscillations are conveyed to muscles via the fastest corticospinal fibers. However, their transmission to MNs of different sizes may vary due to differences in the relative strength of corticospinal and reticulospinal projections across the MN pool. Consequently, it remains uncertain whether corticospinal beta transmission follows similar pathways and maintains consistent strength across the entire MN pool. To investigate this, we examined beta activity in MNs innervating the tibialis anterior muscle across the full range of recruitment thresholds in a study involving 12 participants of both sexes. We characterized beta activity at both the cortical and motor unit (MU) levels, while participants performed contractions from mild to submaximal levels. Corticomuscular coherence remained unchanged across contraction forces after normalizing for the net MU spike rate, suggesting that beta oscillations are transmitted with similar strength to MNs, regardless of size. To further explore beta transmission, we estimated corticospinal delays using the cumulant density function, identifying peak correlations between cortical and muscular activity. Once compensated for variable peripheral axonal propagation delay across MNs, the corticospinal delay remained stable, and its value (∼14 ms) indicated projections through the fastest corticospinal fibers for all MNs. These findings demonstrate that corticospinal beta band transmission is determined by the fastest pathway connecting in the corticospinal tract, projecting similarly across the entire MN pool
AI deception: formalising and evaluating deception in AI agents
No full textThis thesis presents a theory of \emph{deception} applicable to AI agents. Tremendous progress has been made towards generally capable AI systems. Deception has been recognised as a core problem for the alignment of such systems, which may develop misaligned goals, due to deception, or deceive to subvert human control. In short, misaligned AI agents may use deceptive strategies to achieve a range of harmful goals. Although deception has been recognised as an important problem in discussions of alignment, a compelling theory of AI deception has yet been absent from the literature.
Deception involves one agent intentionally causing another to believe something false. Therefore, the first key contribution of this thesis is the development of philosophically grounded definitions of belief, intention, and deception. These definitions are formalised within (structural) causal games, a unifying framework for modelling causal and game-theoretic interactions between agents. Numerous formal results, and examples, demonstrate that these definitions are useful in the context of AI. Building on this foundation, the thesis introduces two algorithms for mitigating deception, demonstrated by the training of reinforcement learning agents that provably do not deceive.
Recent progress has been driven by language models (LMs) and language is a natural medium for lying and deception. The thesis illustrates how our theory applies to frontier LMs. Moreover, we find that LMs fine-tuned to be evaluated as truthful learn to deceive a systematically mistaken evaluator; they intentionally provide answers which they do not believe, in order to be evaluated as truthful when the evaluator makes mistakes. Additionally, LMs can learn to reaffirm falsehoods when asked follow-up questions --- even though they were not trained to do so. State-of-the-art LMs are typically fine-tuned on human feedback, which, these findings suggest, may inadvertently incentivise deception. This has consequences for widely-adopted applications which integrate frontier LMs, such as ChatGPT.Open Acces
Cost-effectiveness of bempedoic acid in high cardiovascular risk patients with statin intolerance: an analysis of the CLEAR outcomes trial
No full textBackground
In the CLEAR Outcomes study, 13,970 high cardiovascular risk patients with hypercholesterolemia and statin intolerance were randomized to treatment with bempedoic acid or standard of care (placebo). Bempedoic acid reduced the risk of major adverse cardiovascular events by 13%. However, the cost-effectiveness of bempedoic acid in this patient population is unknown.
Methods
Markov modeling estimated cost-effectiveness of bempedoic acid versus standard of care alone to reduce cardiovascular risk from a US third-party payer perspective. Baseline risk was estimated by applying individual patient characteristics from the trial to established risk equations. Treatment benefit was extrapolated over a lifetime horizon using hazard ratios for individual major adverse cardiovascular event (MACE) components from CLEAR Outcomes. Scenario analyses included on-treatment analysis, alternate bempedoic acid costs, and modeling effects of the fixed-dose combination with ezetimibe on low-density lipoprotein cholesterol (LDL-C) reduction and predicted MACE.
Results
Bempedoic acid was estimated to reduce lifetime MACE (1.58 versus 1.95 per patient) versus standard of care. At list price, bempedoic acid was associated with increased costs (+ 166,830 per QALY. The on-treatment analysis resulted in an ICER of 99,993 per QALY. Modeling the effects of the fixed-dose combination resulted in an ICER of 150,000 per QALY).
Trial Registration
ClinicalTrials.gov identifier: NCT02993406 (CLEAR Outcomes study)
Towards annual updating of forced warming to date and constrained climate projections
No full textIn the context of rapid human-caused climate change, regular updates of the state of knowledge of current and future climate are needed. New statistical methods using observational constraints underpinned estimates of present-day human-induced warming and projected future warming in the most recent IPCC report. As time goes by, and new updated observational records become available, how should estimates of the current and projected human-caused climate change be updated? Here, we use a perfect model framework and show that incorporating observations from every new year in observationally constrained projections improves their accuracy, without causing major year-to-year spurious variability on outcomes. The forced warming estimated for the current year also exhibits high enough stability to be considered as a robust indicator of the state of the climate system
