Kyle Marshall

Joshua Staker, Kyle Marshall, Robert Abel et al.

Chemical structure extraction from documents remains a hard problem due to both false positive identification of structures during segmentation and errors in the predicted structures. Current approaches rely on handcrafted rules and subroutines that perform reasonably well generally, but still routinely encounter situations where recognition rates are not yet satisfactory and systematic improvement is challenging. Complications impacting performance of current approaches include the diversity in visual styles used by various software to render structures, the frequent use of ad hoc annotations, and other challenges related to image quality, including resolution and noise. We here present end-to-end deep learning solutions for both segmenting molecular structures from documents and for predicting chemical structures from these segmented images. This deep learning-based approach does not require any handcrafted features, is learned directly from data, and is robust against variations in image quality and style. Using the deep-learning approach described herein we show that it is possible to perform well on both segmentation and prediction of low resolution images containing moderately sized molecules found in journal articles and patents.

Anna-Lena Horlemann-Trautmann, Kyle Marshall, Joachim Rosenthal

We present a new attack against cryptosystems based on the rank metric. Our attack allows us to cryptanalyze two variants of the GPT cryptosystem which were designed to resist the attack of Overbeck.

Kyle Marshall, Davide Schipani, Anna-Lena Trautmann et al.

Fuzzy vault is a scheme providing secure authentication based on fuzzy matching of sets. A major application is the use of biometric features for authentication, whereby unencrypted storage of these features is not an option because of security concerns. While there is still ongoing research around the practical implementation of such schemes, we propose and analyze here an alternative construction based on subspace codes. This offers some advantages in terms of security, as an eventual discovery of the key does not provide an obvious access to the features. Crucial for an efficient implementation are the computational complexity and the choice of good code parameters. The parameters depend on the particular application, e.g. the biometric feature to be stored and the rate one wants to allow for false acceptance. The developed theory is closely linked to constructions of subspace codes studied in the area of random network coding.

Felix Fontein, Kyle Marshall, Joachim Rosenthal et al.

Secure storage of noisy data for authentication purposes usually involves the use of error correcting codes. We propose a new model scenario involving burst errors and present for that several constructions.