Showing 1–2 of 2 results for author: Schrans, F

Flint for Safer Smart Contracts

Authors: Franklin Schrans, Daniel Hails, Alexander Harkness, Sophia Drossopoulou, Susan Eisenbach

Abstract: The Ethereum blockchain platform supports the execution of decentralised applications or smart contracts. These typically hold and transfer digital currency to other parties on the platform; however, they have been subject to numerous attacks due to the unintentional introduction of bugs. Over a billion dollars worth of currency has been stolen since its release in July 2015. As smart contracts ca… ▽ More The Ethereum blockchain platform supports the execution of decentralised applications or smart contracts. These typically hold and transfer digital currency to other parties on the platform; however, they have been subject to numerous attacks due to the unintentional introduction of bugs. Over a billion dollars worth of currency has been stolen since its release in July 2015. As smart contracts cannot be updated after deployment, it is imperative that the programming language supports the development of robust contracts. We propose Flint, a new statically-typed programming language specifically designed for writing robust smart contracts. Flint's features enforce the writing of safe and predictable code. To encourage good practices, we introduce protection blocks. Protection blocks restrict who can run code and when (using typestate) it can be executed. To prevent vulnerabilities relating to the unintentional loss of currency, Flint Asset traits provide safe atomic operations, ensuring the state of contracts is always consistent. Writes to state are restricted, simplifying reasoning about smart contracts. △ Less

Submitted 13 April, 2019; originally announced April 2019.

Learning under Distributed Weak Supervision

Authors: Martin Rajchl, Matthew C. H. Lee, Franklin Schrans, Alice Davidson, Jonathan Passerat-Palmbach, Giacomo Tarroni, Amir Alansary, Ozan Oktay, Bernhard Kainz, Daniel Rueckert

Abstract: The availability of training data for supervision is a frequently encountered bottleneck of medical image analysis methods. While typically established by a clinical expert rater, the increase in acquired imaging data renders traditional pixel-wise segmentations less feasible. In this paper, we examine the use of a crowdsourcing platform for the distribution of super-pixel weak annotation tasks an… ▽ More The availability of training data for supervision is a frequently encountered bottleneck of medical image analysis methods. While typically established by a clinical expert rater, the increase in acquired imaging data renders traditional pixel-wise segmentations less feasible. In this paper, we examine the use of a crowdsourcing platform for the distribution of super-pixel weak annotation tasks and collect such annotations from a crowd of non-expert raters. The crowd annotations are subsequently used for training a fully convolutional neural network to address the problem of fetal brain segmentation in T2-weighted MR images. Using this approach we report encouraging results compared to highly targeted, fully supervised methods and potentially address a frequent problem impeding image analysis research. △ Less

Submitted 3 June, 2016; originally announced June 2016.