Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Next revision		Previous revision
Next revisionBoth sides next revision
teaching:gsoc2018 [2018/01/16 09:11] – created ahaiduteaching:gsoc2018 [2018/01/21 20:29] – balintbe
Line 1: Line 1:
 +
 +{{ :teaching:gsoc:gsoc2016logo.jpg?400 |}}
 +
 ====== Google Summer of Code 2018 ====== ====== Google Summer of Code 2018 ======
 ~~NOTOC~~ ~~NOTOC~~
 +===== pracmln =====
 +
 +//pracmln// is a toolbox for statistical relational learning (SRL) and 
 +reasoning and as such also includes tools for standard probabilistic 
 +graphical models. //pracmln// is an implementation of Markov logic networks (MLN) 
 +that supports efficient learning and inference in relational domains. 
 +Its learning and inference engines are entirely written in the Python 
 +programming language. Markov logic networks generalize both first-order
 +logic and probabilistic graphical models and have proven successful
 +in many real-world applications such as natural-language understanding
 +and robot perception. This makes MLNs one of the most general and most
 +powerful representation formalisms for uncertain knowledge.
 +
 +//pracmln// was designed with the particular needs of technical systems in 
 +mind. Our methods are geared towards practical applicability and can 
 +easily be integrated into other applications. The availability of 
 +graphical tools makes both learning or inference sessions a 
 +user-friendly experience. Scripting support enables automation, and
 +a browser-based [[http://pracmln.open-ease.org/|web application]] is intended for teaching and getting
 +familiar with MLNs.
 +
 +//pracmln// is an open-source project hosted on [[https://github.com/danielnyga/pracmln|GitHub.]] It has its own
 +project page ([[http://www.pracmln.org]]) that provides extensive documentation 
 +and tutorials that facilitate getting started with MLNs. It is provided as a pip
 +package in the Python package index ([[https://pypi.python.org/pypi/pracmln|PyPI]]).
 +
 +
 +===== RoboSherlock -- Framework for Cognitive Perception =====
 +
 +RoboSherlock is a common framework for cognitive perception, based on the principle of unstructured information management (UIM). UIM has proven itself to be a powerful paradigm for scaling intelligent information and question answering systems towards real-world complexity (i.e. the Watson system from IBM). Complexity in UIM is handled by identifying (or hypothesizing) pieces of
 +structured information in unstructured documents, by applying ensembles of experts for annotating information pieces, and by testing and integrating these isolated annotations into a comprehensive interpretation of the document.
 +
 +RoboSherlock builds on top of the ROS ecosystem and is able to wrap almost any existing perception algorithm/framework, and allows easy and coherent combination of the results of these. The framework has a close integration with two of the most popular libraries used in robotic perception, namely OpneCV and PCL. More details about RoboSherlock can be found on the project [[http://robosherlock.org/|webpage]].
 +
  
 ===== Proposed Topics ===== ===== Proposed Topics =====
  
 In the following, we list our proposals for the Google Summer of Code topics that contribute to the aforementioned open-source projects. In the following, we list our proposals for the Google Summer of Code topics that contribute to the aforementioned open-source projects.
 +==== Topic 1: Markov logic networks in Python ====
 +
 +[[http://www.pracmln.org|{{  :teaching:gsoc:pracmln-gsoc-figure.png?200|}}]]
 +
 +
 +**Main Objective:** The current implementation of //pracmln// is entirely written in pure 
 +Python. The main objective of this project is to port the 
 +computationally heavy parts of the learning and inference algorithms to 
 +[[http://www.cython.org|Cython]], an extension to Python that allows static 
 +compilation of Python modules to C libraries. Cython allows to add 
 +static type declarations to Python, which can significantly speed up 
 +execution (up to a factor of 1000 compared to regular Python code).
 +
 +As most of the Python libraries for machine learning and scientific 
 +computing (e.g. scikit-learn, numpy) make use of static compilation,
 +it is expected that the practical applicability of Markov logic networks
 +will substanially be pushed to more demanding real-world scenarios.
 +
 +**Task Difficulty:** There is alreay a high-quality code base, however, 
 +solving this task properly requires understanding and experience in 
 +gradient-based optimization, machine learning and logics. 
 +
 +**Requirements:** Good programming skills in the Python programming
 +language (CPython/Cython), experience in Artificial Intelligence and Machine Learning
 +(ideally SRL technques and logic)
 +
 +**Expected Results:** The core components of pracmln, i.e. the learning
 +and inference modules are expected to be ported to the Cython language
 +exploiting the means of implementational optimization provided by
 +Cython.
 +
 +**Contact:** [[team/daniel_nyga|Daniel Nyga]]
 +
 +
 +==== Topic 1: Multi-modal Cluttered Scene Analysis in Knowledge Intensive Scenarios ====
 +
 +{{  :teaching:gsoc:topic1_rs.png?nolink&200|}}
 +
 +**Main Objective:** In this topic we will develop algorithms that en-
 +able robots in a human environment to recognize objects in diffi-
 +cult and challenging scenarios. To achieve this the participant will
 +develop annotators for RoboSherlock that are particularly aimed at
 +object-hypotheses generation and merging. Generating a hypotheses
 +essentially means to generate regions/clusters in our raw data that
 +form a single object or object-part. In particular this entails the de-
 +velopment of segmentation algorithms for visually challenging scenes
 +or object properties, as the likes of transparent objects, or cluttered,
 +occluded scenes. The addressed scenarios include stacked, occluded
 +objects placed on shelves, objects in drawers, refrigerators, dishwash-
 +ers, cupboards etc. In typical scenarios, these confined spaces also
 +bare an underlying structure, which will be exploited, and used as
 +background knowledge, to aid perception (e.g. stacked plates would
 +show up as parallel lines using an edge detection). Specifically we
 +would start from (but not necessarly limit ourselves to) the implemen-
 +tation of two state-of-the-art algorithms described in recent papers:
 +
 +[1] Aleksandrs Ecins, Cornelia Fermuller and Yiannis Aloimonos, Cluttered Scene Segmentation Using the Symmetry Constraint, International Conference on Robotics and Automation(ICRA) 2016
 +[2] Richtsfeld A., M ̈
 +orwald T., Prankl J., Zillich M. and Vincze
 +M. - Segmentation of Unknown Objects in Indoor Environments.
 +IEEE/RSJ International Conference on Intelligent Robots and Sys-
 +tems (IROS), 2012.
 +
 +**Task Difficulty:** The task is considered to be challenging, as it is still a hot research topic where general solutions do not exist.
 +  
 +**Requirements:** Good programming skills in C++ and basic knowledge of CMake and ROS. Experience with PCL, OpenCV is prefered.
 +
 +**Expected Results:** Currently the RoboSherlock framework lacks good perception algorithms that can generate object-hypotheses in challenging scenarios(clutter and/or occlusion). The expected results are several software components based on recent advances in cluttered scene analysis that are able to successfully recognized objects in the scenarios mentioned in the objectives, or a subset of these.
 +
 +Contact: [[team/ferenc_balint-benczedi|Ferenc Bálint-Benczédi]]
 +
 +
- Impressum - Datenschutz



Prof. Dr. hc. Michael Beetz PhD
Head of Institute

Contact via
Andrea Cowley
assistant to Prof. Beetz
ai-office@cs.uni-bremen.de

Discover our VRB for innovative and interactive research


Memberships and associations:


Social Media: