Institute for Artificial Intelligence

Mentors: Student: Kaivalya Rawal

<p>Markov Logic Networks (MLNs) are a generalisation of first-order logic and probabilistic graphical models, thus forming a powerful mechanism of uncertain inference. PracMLN is an open-source toolkit that can be used to perform statistical relational learning (SRL) and probabilistic logic inference based on the Markov Logic representation. It was developed at the Institute for Artificial Intelligence at the University of Bremen.</p> <p>However, the software is written entirely in Python and can thus be slow at times. This project aims to rewrite some computationally intensive portions of PracMLN in Cython, and thus improve performance issues. Intelligent use of Cython's static typing functionality, which is unavailable in Python, will hopefully enable significant speedup, along with more extensive use of other similarly optimised libraries (such as NumPy).</p> <p>Additionally, this work will be documented (<a href="https://kaivalyar.github.io/gsoc18-pracmln/" target="_blank">https://kaivalyar.github.io/gsoc18-pracmln/</a>) in order to make further optimisation easier. Hopefully this will provide a clear, reproducible, live demonstration of the potential speedups that can be brought to PracMLN, along with a detailed guide for future contributors.</p>

Mentors: Student: Shixin Li

<p>This project targets at enabling the current RoboSherlock to work in form of parallel pipeline, which also includes merging results of different camera configurations. In this proposal, I present my plan, methods, timeline and some related experience.</p>

Mentors: Student: Shingo Kitagawa

<p>In GSoC, I propose to introduce deep learning annotators in RoboSherlock for more various task. Because many deep learning frameworks and repositories are written in Python, I first improve RoboSherlock system to be able to run annotators written in Python. Then, I implement annotators using deep learning such as object detection, semantic segmentation, instance segmentation and so on into RoboSherlock system. As the final result of the project, the real robot execute a manipulation task with the deep-learning annotators. The main contributions of my proposal will be as belows:</p> <ul> <li>Improve Robosherlock to execute annotators written in Python</li> <li>Introduce several useful deep-learning annotators in Robosherlock</li> <li>Implement the new annotators in daily-life task with real robot</li> </ul>

Mentors: Student: Mircea-Tudor Chifor

<p>My proposed Project consists of developing a GUI with Unreal Engine for Eye Tracking and any adjacent work involving the aforementioned tehnology.</p>

Mentors: Student: An T. Le

<p>The project proposes flexible pipeline handling mechanism for RoboSherlock framework. The current state of RoboSherlock implementation does not allow to process multiple Analysis Engine pipelines concurrently, as well as the abilities to branch pipeline into multiple subprocesses or merge the results from multiple pipelines into one Common Analysis Structure (CAS). Therefore, this extension project will develop paralleled pipelines handler API and enhance the processing effectiveness of RoboSherlock. The proposal describes the goal, high-level architecture plans of the module and finally milestones of the project.</p>