Loading...
Loading...
Loading...
Loading...
Loading...
Loading...
Loading...
Loading...
Loading...
Loading...
Loading...
Loading...
Loading...
Loading...
Check out the already participating projects!
From June 7 to 10, 2022, and with support of the State Chancellery of Schleswig-Holstein, the Waterkant Festival and Kiel.AI will bring together machine learning and AI enthusiasts to push their machine learning projects to the next level or to get expertise by joining others working on their projects - and all this in the unique atmosphere of the Waterkant Festival!
Register for the event HERE! Under the same link you also find more information on the participation.
The following projects are already participating in the event:
KI.EL Racing Challenge (Jake Petersen, opencampus.sh)
AI and Art (Vladimir Alexeev, OpenAI Embassador; Michael Haas, Freelance Digital Artist)
Detection of Important Genome Elements (Yiqing Wang and Dustin Hanke, University of Kiel)
Identifying Correct Arguments in Open Text Answers (Christian Mayer, University of Mannheim)
Marine Munitions: Identifying Geo-Locations in Historical Documents (Simon Van der Wulp and Juan Veliz, north.io)
Predicting Stock Price Dynamics (Jonas Mielck, University of Kiel and stackOcean GmbH)
Automated Feedback for Students' Argumentative Essays (Thorben Jansen and Nils-Jonathan Schaller, IPN Kiel)
Enabling Researchers to Automatically Code their Text Responses from Assessments (Fabian Zehner and Nico Andersen, DIPF Frankfurt)
Improving Lung Imaging with AI (Niklas Koser and Claus Glüer, University of Kiel)
Natural Language Query in Applied Contexts (Lennart Büttner, Sourceboat GmbH & Co. KG)
Visualizing Language Algorithms (Lena Krüger, Muthesius University of Fine Arts and Design)
Presentation Slides:
Link to the Hugging Face course:
A new form of art is born - be part of it!
Vladimir Alexeev, Digital Artist and OpenAI Ambassador Michael Haas, Freelance Artist
Instead of increasing efficiency through machine learning and doing predictive analytics, we let AI, for example, write poems in the style of Kurt Schwitters and perform them by AI-generated jazz musicians. We can use AI to create previously non-existent people, art movements, and dream sequences, or short films made entirely by AI (script, video, music, and actors).
Your memories and fantasies. :-)
We look forward to participants who are interested in combining artificial intelligence and creativity.
Help to identify mobile genetic elements, for instance, to predict the spread of antibiotic resistances.
Yiqing Wang, University of Kiel Dustin Hanke, University of Kiel
Transposable elements (TE) are frequent genomic sequences that can change their position ineffectively in the genome by copy & paste mechanisms. During this process they are often dragging essential or beneficial genes like antibiotic resistance genes to another location. Sometimes characteristics of TEs are missing or TEs are defragmented, which makes the identification challenging. In this project we are approaching to use the gene frequency landscape of genomic sequences to detect putative TE regions utilizing methods like LSTMs or transformer models.
We provide a protein family frequency dataset of 9226 genomic sequences originating from Klebsiella, Escherichia, and Salmonella.
We are happy about anyone to join, who is interested in understanding the fundamentals of transposable elements and wants to push our approach. Especially, we would be glad about participants having a strong technical background in applying LSTMs and/or Transformer models. Experience in adjusting, improving and building complex ML models would be beneficial.
Support the future of teaching and education.
Christian Mayer, University of Mannheim
In learning contexts it is crucial to provide fast feedback to the learner. Therefore, as a first step, we will categorize open text responses provided on complex problem solving tasks according to the quality and correctness of the arguments used. Later we want to use these to automatically provide appropriate feedback.
About 1,800 open text responses from students.
We are happy for anyone to join our project. You should have a strong interest in how to automate the scoring or classification of open-ended text responses. Basic programming knowledge would be good. You are also welcome to use our data for a fine-tuning approach to the classification.
Challenge the traditionally used VAR(p) and GARCH models with neural networks
Jonas Mielck, University of Kiel / stackOcean GmbH
Prior research has shown that it is possible to include sentiment data in, for example, GARCH or VAR(p) models and thereby increase model accuracy. This projects key question is how these sentiment augmented models perform in comparison to machine learning approaches like neural networks. Therefore, the traditional sentiment augmented models are seen as benchmarks and different ML approaches will be tried to see if the accuracy can be increased.
A dataset containing 1761 days of stock specific trading data extended by sentiment data calculated from news about the company in the years 2013 - 2020. The dataset has dimensions: 1761 rows & 19 columns.
I am happy about everyone who wants to work jointly on the project. You should have knowledge in at least one of the areas: time series statistics, data mining, sentiment analysis or machine learning models for time series data in general. I am also happy to simply exchange ideas on these topics.
Get insights into labeling text data and prompt design using large language models.
Thorben Jansen, IPN Kiel Nils-Jonathan Schaller, IPN Kiel
We are a research project investigating how automated feedback and text assessment can promote students' written argumentation. The project aims to develop and evaluate a digital learning tool for students and teachers. Students will receive automated feedback on their writing and teachers will get an overview of the strengths and weaknesses in their students' written argumentations. We developed argumentation tasks on controversial, socially relevant, real-world problems informed by science (socio-scientific issues), which are situated in the context of climate change. As the basis of our machine learning model, we ask 1500 secondary school students to complete the tasks. Two specially trained raters will rate each argument's content and structure in every text in the resulting corpus.
100 annotated texts and annotation guidelines
We have two goals that we would like to work on with you: first, creating annotation guidelines and training annotators. Second, the partial automation of annotations through zero-shot classifications by use of prompt design.
Train your AI driver to be the fastest on the track!
Jake Petersen, opencampus.sh
Together we want to build, code and race a Self-Driving Car equipped with an NVIDIA Jetson Nano.
We generate the training data by driving the car via a manual remote control.
We are looking forward to participants having very limited coding experience to having already experience in programming machine learning models to complete the track in the shortest amount of time.
Support the Clearance of Munitions Dumped in the Sea
Simon Van der Wulp, north.io Juan Veliz, north.io
German (and international) coastal waters are scattered with munition, from the two world wars. To deal with these munitions and prevent toxic chemical from entering the water marine ecosystem, we need to find, identify and remediate.
The first step deals with the past, mining information available from historical documents to learn where, what and how many munitions could be found, to make the search as effective as possible. This included documents describing munition dumping, sea-battles or mine laying activities that occurred during and in post-war periods.
Using scans of historical text documents, we want to retrieve the positions of ships loaded with munitions that were dumped after World War II. Currently a Neural Net is, in a first step, used to improve the image quality for the OCR scans and, in a second step, a further net is used to identify and extract relevant text information.
We have a training data set consisting of ~1100 scans; it includes the original scans, a corresponding set of scans with manually improved image quality and a corresponding set of labelled text element for each scan.
We are looking forward to participants who want to support us in improving our models and applying new models that have been published in recent times.
Exchange Your ideas on Text Classification
Fabian Zehner, DIPF Frankfurt Nico Andersem, DIPF Frankfurt
When test persons respond to test questions with natural language, the responses typically need to be scored by humans. We use natural language processing techniques and embeddings to represent word semantics for grouping (i.e., clustering) and classifying responses. This pipeline is available via the R-based Shiny app of ReCo (shinyReCoR) through a graphical user interface to people without knowledge of how to use NLP or ML coding frameworks. Therefore, our app focuses on making the classification process transparent and offers many interactive ways to diagnose resulting models/classifiers, among others through visualizing the semantic space.
At the Coding.Waterkant, we plan to develop two new diagnostic and visualization features.
Literature Andersen, N., & Zehner, F. (2021). shinyReCoR: A Shiny Application for Automatically Coding Text Responses Using R. Psych, 3(3), 422–446. doi: 10.3390/psych3030030 Zehner, F., Sälzer, C., & Goldhammer, F. (2016). Automatic Coding of Short Text Responses via Clustering in Educational Assessment. Educational and Psychological Measurement, 76(2), 280–303. doi: 10.1177/0013164415590022
We use a publically released demo data set from ReCo with about 4,000 text responses to a demo item for demonstration purposes. For internal purposes, we work with confidential text responses from the PISA assessment.
We are happy to exchange ideas on text classification and how to provide easy access to current NLP and ML methods.
Generating Database Queries from Natural Language Inputs
Lennart Büttner, Sourceboat GmbH & Co. KG
We are currently developing a database interaction & creation tool for users without a programming background. In this context we want to evaluate the possibilites to use NLP to generate queries from natural languages, especially German and English.
Ideas reach from a chatbot that simply answers questions about your data, e.g. “How many paying customers do we have” over the generation of endpoints in ensemble with a functions as a service provider up to schema creation.
At this very first phase of experimentation and ideation we want to estimate scope and requirements of this project. Primary challenges to overcome are the dynamic generation of data based on a database schema as well as the tokenization and handling of Out-of-Vocabluary words.
We started using databases in different languages from http://paraphrase.org/#/download to generate data based on table and column names of a database schema.
I am looking forward to discuss techniques and exchange ideas with everyone who is interested.
Developing and Testing Super-Resolution AI Techniques for Refined Depiction of the Bronchial Tree
Niklas Koser, University of Kiel Claus Glüer, University of Kiel
The Intelligent Imaging Lab (i2Lab) of the Section Biomedical Imaging, Department of Radiology and Neuroradiology, UKSH, and CAU Kiel is developing innovative artificial intelligence methods for radiological imaging. Here we address super-resolution imaging. This AI approach may allow to improve the image quality (without increasing the radiation dose) or reduce the radiation dose (at the same image quality).
In the challenge, participating teams will use a pretrained Artificial Neural Network to segment the bronchial tree and try to program a super-resolution approach for improving image quality to improve the segmentation result.
60 lung CT images and segmentation mask from public repositories will be provided by us (40 for training/validation and 20 for testing)
Single competitors or teams with some experience in building ML models are invited to participate. Experience in PyTorch or TensorFlow would be beneficial.
Contact
Lena Krüger, Muthesius University of Fine Arts and Design
Language and linguistic contexts are to be visualized multidimensionally on the basis of geometries generated by language algorithms. This project’s aim is to explore how the algorithm can be used creatively and how its output can be visualized.
Your creativity and fantasy. :-)
I look forward to participants who are interested in combining artificial intelligence and creativity.
Here the Link to the Streamlit Docker Example:
If you have any questions, feel free to contact Matthias under mail@matthiasnannt.com