Calendar
For information about attending a Salisbury Cafe Scientifique event, see the Attending an Event section; there is also more general information in the Frequently Asked Questions section and help on making the most of this calendar in the Calendar Help section. If you fancy a night of science outside but close to Salisbury, there is also this filtered list of nearby events.
-
Categories
-
Tags
3D printing acoustics active colloids AHVLA AI air pollution alcohol alien Amish ancestor archaeological science archaeology arithmetic Artificial Intelligence attention Attention Deficit Hyperactivity Disorder bacteria badger Balkans ballistics Basingstoke bio-safety biofuel biological science Biology biomedical science bird Bournemouth University brain breeding Bright Club British Trust for Ornithology bubbles bumblebee bus butterfly cabaret calculation Cambridge campaign cancer car cargo ship cartography chartered institute Checkhov chemical pathology chemistry Churches Conservation Trust climate change cloud code code breaking Cognitive Neuroscience collective behaviour comedy community computer computer science computing conservation conspiracy theory cooking Coombe Bissett cosmic ray Council for British Archaeology Cranfield University crowds cryptography crystal sponge cuckoo dairy data deep learning design diet digital collections digital micrograph disability disease distraction DNA doctor Doctor Who dog dorchester Dr Who DSTL earth science ebola ecology Eden Project EGS Energy Einstein elements energy engineering entertainment entomology environment epigenetics ethics evolutionary linguistics exploration explosives faeces fame family Festival of British Archaeology flat screen flock flooding fluids fluoride food food science foresters fossil fracking fuel functional inorganic chemistry Galapagos garden gardening gene genetics genmed genomics Genomics England geology geomatics geothermal GKN GNSS gravity green green energy gut health healthcare Henry VIII Heritage Lottery Fund history HIV homo erectus honey bee horticulture hospital human development human evolution hybrid power immersive immune system immunology Individual differences infinity insect Institute of Continuing Education International Year of Light internet IoP itinerary journalism kidney land management lcd light light microscopy liquid crystals liquid nitrogen liver machine machine learning magnetism map Mary Rose materials materials chemistry mathematics maths meadow media medicine memory metal-organic frameworks meteorology microbe microbiology microbiome microplastic microscope migration milk mind Mind wandering miniaturisation molecules mosquito moth multimedia munitions museum nanomaterials National Oceanographic Centre National Physics Laboratory National Poo Museum natural history Natural History Museum nature reserve navigation neutrino news NHS nightingale NOC numerical nutrition ocean ocean science Ohio oncology online optical computing optoelectronics pandemic particle accelerator pathology permaculture personal information PHE PHLS physical geography physics physiology pineal gland place plague plant Plantlife plastic plate tectonics policy poo privacy prosthetic limb psychology pub quantum quantum computing quantum mechanics recognition recycling resuscitation rheology risk rocket route Royal Holloway Royal Navy RTI sail power Sakhalin Salisbury Plain science Science Fiction scifi Secret Garden security self replicating sense about science Sherborne ship shoal Siberia silk moth skylark solar system Somerset Archaeological and Natural History Society Somerset County Council soundscape Southampton Solent University space space debris space rocket speed record sport stand up statistics supramolecular chemistry survey sustainability swarm swift synchrotron technology television tetricus Thomas Hardye School time travel TLS transplant TST Tudor UCL University College London University Hospital Southampton University of Bath University of Bristol University of Cambridge University of Exeter University of Hertfordshire University of Kent University of London University of Manchester University of Plymouth University of Portsmouth University of Reading University of Southampton University of Winchester vaccine variety vehicle video VLA warship water wave particle duality weather Wellcome Trust Wessex Archaeology Wessex Water wild flower wildlife Wiltshire Wildlife Trust work related illness World War II World Wide Web wreck zoonotic disease
Gaussian Processes are a tool of choice for modelling function with uncertainties. However, inference is only tractable analytically for the classical case of regression with Gaussian noise since all other likelihoods are not conjugate with the Gaussian prior.
In this talk I will show how one can transform a large class of likelihoods into conditional conjugate distributions by augmenting them with latent variables. These augmented models have the advantage that, while the posterior inference is still not fully analytic, the full conditionals are! Consequently, one can work easily (and efficiently!) with algorithms like Gibbs sampling or Coordinate Ascent VI (CAVI) and outperform existing inference methods.
Reference:
Galy-Fajou, Théo, Florian Wenzel, and Manfred Opper. “Automated Augmented Conjugate Inference for Non-conjugate Gaussian Process Models.” International Conference on Artificial Intelligence and Statistics. PMLR, 2020. https://arxiv.org/abs/2002.11451
Theme: *Adaptive Brain Computations*
Pain is a major clinical problem affecting 1 in 5 people in the world. There are unresolved questions that urgently require answers to treat pain effectively, a crucial one being how the feeling of pain arises from brain activity. Computational models of pain consider how the brain processes noxious information and allow mapping neural circuits and networks to cognition and behaviour. To date, they have generally have assumed two largely independent processes: perceptual and/or predictive inference, typically modelled as an approximate Bayesian process, and action control, typically modelled as a reinforcement learning process. However, inference and control are intertwined in complex ways, challenging the clarity of this distinction. I will discuss how they may comprise a parallel hierarchical architecture that combines pain inference, information-seeking, and adaptive value-based control. Finally, I will discuss whether and how these learning processes might contribute to chronic pain.
Flavia Mancini is a neuroscientist leading the NoxLab within the Computational and Biological Learning (CBL) research unit at the Department of Engineering, University of Cambridge. Flavia combines behavioural, computational, and neuroimaging tools to understand pain perception and behaviour in humans. She trained at the University of Milan and University College London. Currently, she is a MRC Career Development Award fellow.
Register in advance for this meeting:
https://us02web.zoom.us/meeting/register/tZwtc-ytrDkiEtIAmonfG6-XNR06o_4F9g0i
After registering, you will receive a confirmation email containing information about joining the meeting.
I shall describe the role of exponentially small terms in some PDEs and related differential difference equations, and ask questions about possible commonalities with QFTs. While I am mainly interested in dissipative (parabolic) cases, I shall – for reasons that are presumably obvious – focus on properties shared by the corresponding time-reversible (hyperbolic) models.
The tumour suppressor protein p53 orchestrates the response to cell damage and thus plays a central role in preventing cancer. p53 is tightly regulated by post-translational modifications (PTMs). The precise mechanisms through which p53 PTMs operate are difficult to elucidate due to the complexity of p53 signalling and challenges associated with preparing site-specifically modified p53 for in vitro studies. To address these issues, we have developed a protein semi-synthesis strategy to access p53 in defined PTM states. Using such ‘designer’ phospho-p53 variants we have probed the mechanism of p53 activation through phosphorylation in vitro. Moreover, we found that a spontaneous protein backbone modification, the rearrangement of an asparagine to an isoaspartate residue, reconfigures p53’s binding partner specificity, which suggests that p53 could act as a molecular time bomb. Given the importance of PTMs in p53 signalling, we believe that our chemistry-driven approach will contribute greatly to a mechanistic understanding of how this protein makes cellular life and death decisions.
PAC-Bayes is a frequentist framework for obtaining generalisation error bounds. It has been used to derive learning algorithms, provide explanations for generalisation in deep learning, and form connections between Bayesian and frequentist inference. This reading group will cover a broad introduction to PAC bounds, the proof ideas in PAC-Bayes, and a discussion of some recent applications.
Suggested reading:
# Computing Nonvacuous Generalization Bounds for Deep (Stochastic) Neural Networks with Many More Parameters than Training Data: https://arxiv.org/abs/1703.11008
# PAC-Bayesian Theory Meets Bayesian Inference: https://arxiv.org/abs/1605.08636
# Learning under Model Misspecification: Applications to Variational and Ensemble Methods: https://arxiv.org/abs/1912.08335
One of the large challenges in sea ice science is how sea ice properties (e.g., albedo, thermal conductivity) on small scales influence properties and processes on larger scales (e.g., the floe or ridge scale and basin-scale sea ice behaviour). To make progress one has to understand the variability in small-scale properties which can span several orders of magnitude (e.g. permeability). For many properties a strong dependence on temperature and salinity has been found, yet the detailed physical processes leading to this variability have remained unclear: On the one hand are the bulk fractions of sea ice constituents (gas, ice, brine and solid salts) often insufficiently known or measured. On the other hand, there is the lack in observations of 3D sea ice micro-structure to which the physical properties are related.
In the present talk I will focus on the concept of “Digital Sea Ice physics” to improve our understanding of sea ice properties, their dependence on microstructure and growth conditions, and illustrate several applications to geophysical sea ice problems. The concept is adopted from rock science where it has been established as “Digital Rock Physics” (DRP) during the last decade. It is based on 3D X-ray tomographic imaging and digitizing of the sea ice pore space, followed by direct numerical computation of its effective physical properties. In this way the relationship between effective physical properties of sea ice and its bulk constituents (volume fractions of ice, air, brine and solid salts) is determined and related to micro-scale characteristics of the pore space, providing an improved understanding of the properties’ variability.
I will begin the talk with an overview of sea ice properties and microstructure and their variability, to illustrate related challenges and open questions in sea ice science, identifying the need of 3D microstructure information for many topics. I will then describe the work flow of “Digital Sea Ice Physics” from field sampling to physical property computations, as well as the challenges for the porous medium sea ice, when compared to rocks and snow. I will discuss several applications to obtain sea ice properties that are relevant for sea ice and climate modelling: (i) transport properties of sea ice and recent results on sea ice permeability and electrical conductivity; (ii) the microstructure at the sea ice ocean interface (with relevance for on ice-ocean heat, salt and momentum exchange as well as ice-ice friction) and (iii) the ice surface regime (with relevance for albedo and inter-facial processes between sea ice and snow). Digital Sea ice physics is a concept that has large future potential due to increasing computational power to handle large 3D images. The talk closes with an overview of climate-relevant sea ice properties for which the approach opens new paths to fundamental knowledge and understanding.
NB Live event CANCELLED. Due to a conflict of teaching commitments Prof Pharoah is no longer able to deliver his lecture live at the scheduled time. He has agreed to record the lecture and make it available. Please contact the Training Programme for further information
*Abstract:* The talk takes a multi-level approach to the question of how psychedelics work in the brain. Key themes include:
* the pharmacology of classic serotonergic psychedelics,
* what this tells us about the function and evolutionary purpose of the serotonin 2A receptor,
* the acute brain effects of psychedelics as determined by functional brain imaging,
* the entropic brain hypothesis,
* current evidence for psychedelic therapy,
* the new ‘REBUS’ hierarchical predictive processing model of the action of psychedelics, and
* how this maps on to the phenomenology of the acute psychedelic experience and therapeutic outcomes.
*Biography:* Dr Robin Carhart-Harris moved to Imperial College London in 2008 after obtaining a PhD in Psychopharmacology from the University of Bristol and an MA in Psychoanalysis from Brunel University. At Imperial, Robin has designed and completed human brain imaging studies with LSD, psilocybin, MDMA and DMT, a clinical trial of psilocybin for treatment-resistant depression, a double-blind randomised controlled trial comparing psilocybin with escitalopram for major depressive disorder and a multimodal imaging study in healthy volunteers receiving psilocybin for the first time. Robin founded the Centre for Psychedelic Research at Imperial College London in April 2019, the first of its kind in the world. For more detailed, please visit: https://www.imperial.ac.uk/people/r.carhart-harris
Normative theories and statistical inference provide complementary approaches for the study of biological systems. A normative theory postulates that organisms have adapted to efficiently solve essential tasks and proceeds to mathematically work out testable consequences of such optimality; parameters that maximize the hypothesized organismal function can be derived ab initio, without reference to experimental data. In contrast, statistical inference focuses on the efficient utilization of data to learn model parameters, without reference to any a priori notion of biological function. Traditionally, these two approaches were developed independently and applied separately. Here, we unify them in a coherent Bayesian framework that embeds a normative theory into a family of maximum-entropy “optimization priors.” This family defines a smooth interpolation between a data-rich inference regime and a data-limited prediction regime. I will illustrate how this framework can productively guide our thinking on several neuroscience and non-neuroscience examples.
Join Zoom Meeting
https://us02web.zoom.us/j/85212773275?pwd=cUJ2ZTNaazZzODVTTDFCZm9PWUdKQT09
Meeting ID: 852 1277 3275
Passcode: 640801
Join Zoom Meeting
https://us02web.zoom.us/j/85212773275?pwd=cUJ2ZTNaazZzODVTTDFCZm9PWUdKQT09
Meeting ID: 852 1277 3275
Passcode: 640801
