Event details

Event details

Event details

Uncertainty and variability is intrinsic to a plethora of biological processes that we want to understand, model and predict. In cardiac modelling, sources of uncertainty stem from the experimental error in the measurements from our protocols, lack of knowledge about the underlying mechanisms leading to structural error in our models, variability due to differences in cell and ion channel states due to cells being in different settings and gene expression patterns, and variability due to the inherent stochasticity of some of these processes exhibited at multiple time and spatial scales. To accommodate mathematical/phenomenological models in safety-critical clinical practice and drug development, it is therefore of utmost importance to quantify and propagate these uncertainties to model predictions. Bayesian statistics plays a major role in carrying out uncertainty quantification effectively. However, cardiac models pose a unique set of challenges for Bayesian statistical methods. In this talk I would present Bayesian statistical and modern machine learning approaches towards “forward” (from inputs to model predictions) and “inverse” (from experimental data to model structure) uncertainty quantification in cellular cardiac electropysiological models. Specifically, I would present approaches to overcome the computational and statistical challenges associated with uncertainty quantification in mechanistic models, described by differential equations, and highlight some of the open challenges. Furthermore, I would discuss the potential of modern machine learning techniques such as black-box variational inference and probabilistic programming towards solving the uncertainty quantification problem efficiently. Following are the references accompanying this talk: 1) Sanmitra Ghosh, David Gavaghan, Gary Mirams, “Gaussian process emulation for discontinuous response surfaces with applications for cardiac electrophysiology models”, https://arxiv.org/abs/1805.10020v1 2) Sanmitra Ghosh “Probabilistic Programming for Mechanistic Models (P2M2) tutorial repository”, https://github.com/sanmitraghosh/P2M2

Event details

One of the key challenges of visual perception is to extract abstract models of 3D objects and object categories from visual measurements, which are affected by complex nuisance factors such as viewpoint, occlusion, motion, and deformations. In this talk, I will introduce two approaches that, given a large number of images of an object and no other supervision, can factorize image deformations and appearance. I will demonstrate the applicability of this method to articulated objects and deformable objects such as human faces and body by learning embeddings from random synthetic transformations or optical flow correspondences, all without any manual supervision. The talk will cover three recent recent papers: [1] Thewlis, J., Bilen, H., & Vedaldi, A. (2017). Unsupervised learning of object landmarks by factorized spatial embeddings. In International Conference on Computer Vision (ICCV). [2] Thewlis, J., Bilen, H., & Vedaldi, A. (2017). Unsupervised learning of object landmarks by factorized spatial embeddings. In Neural Information Processing Systems (NIPS). [3] Jakab, T., Gupta, A., Bilen, H., & Vedaldi, A. (2018). Conditional Image Generation for Learning the Structure of Visual Objects. (NIPS).

Event details

Event details

Abstract: The increasing adoption of unmanned systems raises the challenge of the prevention of unauthorised access to them for nancial gain or malicious intent. The problem is exacerbated for maritime systems which are intentionally operated at considerable physical distance from the data or asset owner. The success of a naval mission is subject to the ful lment of a set of operational requirements before and during each voyage. As these requirements depend essentially on the maritime system components and the mission pro le, the e ects of failures can be very signi cant if they are not anticipated. In this paper, we use systems-theoretic process analysis (STPA) to develop a systematic mechanism to analyse the security functionalities of a fully autonomous ship. STPA is a hazard analysis technique capable of identifying potential hazardous design flaws, including software and system design errors and unsafe interactions among multiple system components. As part of the process analysis, we identified potential threats, vulnerabilities and attacks in an autonomous ship. The analyses can be used as a springboard to drive an autonomous ship system architecture and to designing a more e ective and secure system.

Event details

Event details

This talk will describe the packaging technique for flexible circuits used in smart textiles (e-textiles)

Event details

Title: Textile based Wearable Gas sensors Abstract: Contamination of air due to various chemical molecules in smart cities has both short and long-term effects on human health. Wearable sensors allow an individual to monitor the air quality in real time at any location enabling portability. It is important to make the sensor invisible and allow comfort to the user. This is achieved by fixing the device into textiles. This research is about incorporating gas sensors into textiles.