MRC Laboratory of Molecular Biology
CUED - Cambridge University Engineering DepartmentNot a member yet
45551 research outputs found
Sort by
Meta-learning stationary stochastic process prediction with convolutional neural processes
No full textStationary stochastic processes (SPs) are a key component of many probabilistic models, such as those for off-the-grid spatio-temporal data. They enable the statistical symmetry of underlying physical phenomena to be leveraged, thereby aiding generalization. Prediction in such models can be viewed as a translation equivariant map from observed data sets to predictive SPs, emphasizing the intimate relationship between stationarity and equivariance. Building on this, we propose the Convolutional Neural Process (ConvNP), which endows Neural Processes (NPs) with translation equivariance and extends convolutional conditional NPs to allow for dependencies in the predictive distribution. The latter enables ConvNPs to be deployed in settings which require coherent samples, such as Thompson sampling or conditional image completion. Moreover, we propose a new maximum-likelihood objective to replace the standard ELBO objective in NPs, which conceptually simplifies the framework and empirically improves performance. We demonstrate the strong performance and generalization capabilities of ConvNPs on 1D regression, image completion, and various tasks with real-world spatio-temporal data
High-precision solid catalysts for investigation of carbon nanotube synthesis and structure
No full textThe direct growth of single-walled carbon nanotubes (SWCNTs) with narrow chiral distribution remains elusive despite substantial benefits in properties and applications. Nanoparticle catalysts are vital for SWCNT and more generally nanomaterial synthesis, but understanding their effect is limited. Solid catalysts show promise in achieving chirality-controlled growth, but poor size control and synthesis efficiency hampers advancement. Here, we demonstrate the first synthesis of refractory metal nanoparticles (W, Mo, and Re) with near-monodisperse sizes. High concentrations (N = 105 to 107 cm−3) of nanoparticles (diameter 1 to 5 nm) are produced and reduced in a single process, enabling SWCNT synthesis with controlled chiral angles of 19° ± 5°, demonstrating abundance >93%. These results confirm the interface thermodynamics and kinetic growth theory mechanism, which has been extended here to include temporal dependence of fast-growing chiralities. The solid catalysts are further shown effective via floating catalyst growth, offering efficient production possibilities
Response of cells on a dense array of micro-posts
No full textWe have analysed the response of cells on a bed of micro-posts idealized as a Winkler foundation using a homeostatic mechanics framework. The framework enables quantitative estimates of the stochastic response of cells along with the coupled analysis of cell spreading, contractility and mechano-sensitivity. In particular the model is shown to accurately predict that: (i) the extent of cell spreading, actin polymerisation as well as the traction forces that cells exert increase with increasing stiffness of the foundation; (ii) the traction forces that cells exert are primarily concentrated along the cell periphery; and (iii) while the total tractions increase with increasing cell area the average tractions are reasonably independent of cell area, i.e. for a given substrate stiffness, the average tractions that are normalized by cell area do not vary strongly with cell size. These results thus suggest that the increased foundation stiffness causes both the cell area and the average tractions that the cells exert to increase through higher levels of stress-fibre polymerization rather than the enhanced total tractions being directly linked through causation to the larger cell areas. A defining feature of the model is that its predictions are statistical in the form of probability distributions of observables such as the traction forces and cell area. In contrast, most existing models present solutions to specific boundary value problems where the cell morphology is imposed a priori. In particular, in line with observations we predict that the diversity of cell shapes, sizes and measured traction forces increase with increasing foundation stiffness. The homeostatic mechanics framework thus suggests that the diversity of observations in in vitro experiments is inherent to the homeostatic equilibrium of cells rather than being a result of experimental errors
Guest Editorial Special Issue on Papers From the IEEE FLEPS Conference 2019
No full textGuest editorial for IEEE FLEPS Conference 201
Growth rate of lithium filaments in ceramic electrolytes
No full text© 2020 Lithium-ion batteries with single ion-conductor ceramic electrolytes short-circuit when subjected to charging currents above a critical current density. Here, we analyse the rate at which a lithium (Li) filament (sometimes referred to as a dendrite) will grow from the cathode towards the anode during charging of such batteries. The filament is modelled as a climbing edge dislocation with its growth occurring by Li+ flux from the electrolyte into the filament tip at constant chemical potential. The growth rate is set by a balance between the reduction of free-energy at the filament tip and energy dissipation associated with the resistance to the flux of Li+ through the filament tip. For charging currents above the critical current density, the filament growth rate increases with decreasing filament tip resistance. Imperfections, such as voids in the Li cathode along the electrolyte/cathode interface, decrease the critical current density but filament growth rates are also lower in these cases as filament growth rates scale with the charging currents. The predictions of the model are in excellent quantitative agreement with measurements and confirm that above the critical current density a filament can traverse the electrolyte in minutes or less. This suggests that initiation of filament growth is the critical step to prevent short-circuiting of the battery
Packet Speed and Cost in Mobile Wireless Delay-Tolerant Networks
No full textA mobile wireless delay-tolerant network (DTN) model is proposed and analyzed, in which infinitely many nodes are initially placed on according to a uniform Poisson point process (PPP) and subsequently travel, independently of each other, along trajectories comprised of line segments, changing travel direction at time instances that form a Poisson process, each time selecting a new travel direction from an arbitrary distribution; all nodes maintain constant speed. A single information packet is traveling towards a given direction using both wireless transmissions and sojourns on node buffers, according to a member of a class of utility-based routing rules. For this model, we compute the long-term averages of the speed with which the packet travels towards its destination and the rate with which the wireless transmission cost accumulates. Because of the complexity of the problem, we employ two intuitive, simplifying approximations; simulations show that the approximation error is typically small. Our results provide intuition on the fundamental trade-off that exists in mobile wireless DTNs between the packet speed and the packet delivery cost. The framework developed here is both general and versatile, and can be used as a starting point for further investigation
Identification and validation of a driver steering control model incorporating human sensory dynamics
No full textMost existing models of driver steering control do not consider the driver's sensory dynamics, despite many aspects of human sensory perception having been researched extensively. The authors recently reported the development of a driver model that incorporates sensory transfer functions, noise and delays. The present paper reports the experimental identification and validation of this model. An experiment was carried out with five test subjects in a driving simulator, aiming to replicate a real-world driving scenario with no motion scaling. The results of this experiment are used to identify parameter values for the driver model, and the model is found to describe the results of the experiment well. Predicted steering angles match the linear component of measured results with an average ‘variance accounted for’ of 98% using separate parameter sets for each trial, and 93% with a single fixed parameter set. The identified parameter values are compared with results from the literature and are found to be physically plausible, supporting the hypothesis that driver steering control can be predicted using models of human perception and control mechanisms
Designing Adaptation in Cars: An Exploratory Survey on Drivers’ Usage of ADAS and Car Adaptations
No full textCurrent production cycle cars offer a wide range of driver assistance features spanning from Advanced Driver Assistance Systems to more established systems such as wing mirrors. All these features allow an increasing amount of adaptation enabling the driver to tailor all them to his or her requirements. However, drivers’ usage of and attitude towards these features as well as their possible adaptations are largely unexplored and, as a consequence, not well understood. We present an exploratory survey on this topic and apply an inclusive design approach in order to accommodate the whole range of diversity in our population. The results indicate a low usage rate of driver assistance features as well as their possible adaptations. However, results suggest a high appreciation for a potential smart adaptation of driver assistance features
Origin of Defect Tolerance in InAs/GaAs Quantum Dot Lasers Grown on Silicon
No full text© 1983-2012 IEEE. High-performance III-V quantum-dot lasers monolithically grown on Si substrates have been demonstrated as a promising solution to realize Si-based laser sources with very low threshold current density, high output power, and long lifetime, even with relatively high density of defects due to the material dissimilarities between III-Vs and Si. On the other hand, although conventional III-V quantum-well lasers grown on Si have been demonstrated after great efforts worldwide for more than 40 years, their practicality is still a great challenge because of their very high threshold current density and very short lifetime. However, the physical mechanisms behind the superior performance of silicon-based III-V quantum-dot lasers remain unclear. In this paper, we directly compare the performance of a quantum-well and a quantum-dot laser monolithically grown on on-axis Si (001) substrates, both experimentally and theoretically, under the impact of the same density of threading dislocations. A quantum-dot laser grown on a Si substrate with a high operating temperature (105 °C) has been demonstrated with a low threshold current density of 173 A/cm2 and a high single facet output power >100 mW at room temperature, while there is no lasing operation for the quantum-well device at room temperature even at high injection levels. By using a rate equation travelling-wave model, the quantum-dot laser's superior performance compared with its quantum well-based counterpart on Si is theoretically explained in terms of the unique properties of quantum dots, i.e., efficient carrier capture and high thermal energy barriers preventing the carriers from migrating into defect states
Multi agent system for machine learning under uncertainty in cyber physical manufacturing system
No full textRecent advancement in predictive machine learning has led to its application in various use cases in manufacturing. Most research focused on maximising predictive accuracy without addressing the uncertainty associated with it. While accuracy is important, focusing primarily on it poses an overfitting danger, exposing manufacturers to risk, ultimately hindering the adoption of these techniques. In this paper, we determine the sources of uncertainty in machine learning and establish the success criteria of a machine learning system to function well under uncertainty in a cyber-physical manufacturing system (CPMS) scenario. Then, we propose a multi-agent system architecture which leverages probabilistic machine learning as a means of achieving such criteria. We propose possible scenarios for which our architecture is useful and discuss future work. Experimentally, we implement Bayesian Neural Networks for multi-tasks classification on a public dataset for the real-time condition monitoring of a hydraulic system and demonstrate the usefulness of the system by evaluating the probability of a prediction being accurate given its uncertainty. We deploy these models using our proposed agent-based framework and integrate web visualisation to demonstrate its real-time feasibility