Variational autoencoders (VAEs) learn representations of data by jointly training a probabilistic encoder and decoder network. Typically these models encode all features of the data into a single ...variable. Here we are interested in learning disentangled representations that encode distinct aspects of the data into separate variables. We propose to learn such representations using model architectures that generalise from standard VAEs, employing a general graphical model structure in the encoder and decoder. This allows us to train partially-specified models that make relatively strong assumptions about a subset of interpretable variables and rely on the flexibility of neural networks to learn representations for the remaining variables. We further define a general objective for semi-supervised learning in this model class, which can be approximated using an importance sampling procedure. We evaluate our framework's ability to learn disentangled representations, both by qualitative exploration of its generative capacity, and quantitative evaluation of its discriminative ability on a variety of models and datasets.
The COVID-19 pandemic has highlighted the importance of in-silico epidemiological modelling in predicting the dynamics of infectious diseases to inform health policy and decision makers about ...suitable prevention and containment strategies. Work in this setting involves solving challenging inference and control problems in individual-based models of ever increasing complexity. Here we discuss recent breakthroughs in machine learning, specifically in simulation-based inference, and explore its potential as a novel venue for model calibration to support the design and evaluation of public health interventions. To further stimulate research, we are developing software interfaces that turn two cornerstone COVID-19 and malaria epidemiology models COVID-sim, (https://github.com/mrc-ide/covid-sim/) and OpenMalaria (https://github.com/SwissTPH/openmalaria) into probabilistic programs, enabling efficient interpretable Bayesian inference within those simulators.
We develop a framework for incorporating structured graphical models in the \emph{encoders} of variational autoencoders (VAEs) that allows us to induce interpretable representations through ...approximate variational inference. This allows us to both perform reasoning (e.g. classification) under the structural constraints of a given graphical model, and use deep generative models to deal with messy, high-dimensional domains where it is often difficult to model all the variation. Learning in this framework is carried out end-to-end with a variational objective, applying to both unsupervised and semi-supervised schemes.
Trans to cis photoisomerization of epidermal urocanic acid (UCA) has been proposed as a primary event in UV-induced immunosuppression. 8-Methoxypsoralen (8-MOP) in combination with UVA radiation ...(PUVA) is used for photochemotherapy of several immunologicaly based skin disorders, and PUVA is known to cause immunosuppression. We examined the photointeraction of 8-MOP and other skin photosensitizing drugs (amiodarone, ciprofloxacin, doxycycline and ketoprofen) with
trans-UCA in vitro. In oxic condition 8-MOP did not enhance the ability of radiation from standard PUVA fluorescent sources (0.2% radiation below 320 nm) to produce
cis-UCA. When oxygen was removed, some 8-MOP photosensitization could be detected. 8-MOP, ciprofloxacin and ketoprofen also enhanced inefficient UCA photoisomerization by radiation from a filtered tungsten halogen source (0.2% radiation below 320 nm). 8-MOP photosensitization is essential for radiation from PUVA tubes to have a therapeutic or immunosuppressive effect. We have shown in vitro that 8-MOP does not enhance
cis-UCA production by this source. Extrapolation of our data to the in vivo situation suggests that UCA may not be involved in either the therapeutic or the immunosuppressive effect of PUVA.
Estimating tensors for matching over multiple views Lasenby, Joan; Lasenby, Anthony N.
Philosophical transactions of the Royal Society of London. Series A: Mathematical, physical, and engineering sciences,
05/1998, Letnik:
356, Številka:
1740
Journal Article
Recenzirano
In this paper we give purely geometric derivations of the constraints between point and line correspondences in multiple views of a static field. The analysis is carried out using geometric algebra, ...a system which provides a useful tool in many computer vision applications. It is shown that with a straightforward geometric interpretation it is simple to derive the degrees of freedom of such tensors and to understand their structure. Given such information, minimal parametrizations of the tensors are possible. Such parametrizations may be useful for estimation of the tensors and subsequent matching of points.
Four silicas, two fumed silicas (one hydrophilic and one hydrophobic) and two precipitated silicas (one hydrophilic and one hydrophobic), were added as filler to solvent‐based polyurethane (PU) ...adhesive formulations. In general, the addition of silica increased the viscosity, the storage and loss moduli of the PU adhesives but only the hydrophilic fumed silica exhibited pseudoplasticity and thixotropy. The rheological properties imparted by adding filmed silicas to PU adhesive solutions were more noticeable than that of precipitated silicas. Interactions between the hydrophilic fumed silica, the polyurethane and/or the solvent seemed to be responsible for the improved rheological properties of filled PU adhesives.
Neuraminidase was injected into the cerebrospinal fluid of normal rats to investigate the assembly and fate of the desialylated Reissner's fiber glycoproteins. It was established that a single ...injection of neuraminidase cleaved the sialic acid residues of the Reissner's fiber glycoproteins that had been assembled before the injection, and of the molecules that were released over a period of at least 4 h after the injection. These desialylated glycoproteins underwent an abnormal assembly that led to the formation of spheres instead of a fiber. The number of these spheres increased during the 4-h period following the injection, indicating that neuraminidase did not prevent the secretion of the Reissner's fiber glycoproteins into the cerebrospinal fluid. The spheres remained attached to the surface of the subcommissural organ and became intermingled with infiltrating cells, many of which were immunocytochemically identified as macrophages. The latter were seen to contain immunoreactive Reissner's fiber material. It is concluded that the desialylated Reissner's fiber glycoproteins forming the spheres underwent an in situ degradation by macrophages, thus resembling the normal process undergone by the Reissner's fiber glycoproteins reaching the massa caudalis.
We report the results of an investigation of the effects of a pure superconductor on external gravitomagnetic and magnetic fields {bold B}{sub {ital g},0} and {bold B}{sub 0}, respectively. We find ...that the internal fields are given by {bold B}({ital z}){approx}{minus}({ital m}{sup 2}{mu}{sub {ital g}}/{ital q}{sup 2}{mu}){bold B}{sub 0}{minus}({ital m}/{ital q}){bold B}{sub {ital g},0}, {bold B}{sub {ital g}}({ital z}){approx}{bold B}{sub {ital g},0}+({mu}{sub {ital g}m}/{ital q}{mu}){bold B}{sub 0}, where {mu}{sub {ital g}} and {mu} are the gravitomagnetic and magnetic permeabilities of the superconductor. These results show that a small residual uniform magnetic field will pervade the superconductor and that the external fields mutually induce'' additional small internal perturbation fields. The sum of the fields {bold B}+({ital m}/{ital q}){bold B}{sub {ital g}} falls exponentially to zero over a characteristic distance {lambda}, which is consistent with previous findings that {bold B}+({ital m}/{ital q}){bold B}{sub {ital g}}=0 inside a pure superconductor.