Mattermost
(Chat-Communication during the hackathon)
The link to the chat will be provided the day before start of the hackathon.

GitHub
The datasets for the challenges are provided in the following private respository. You will get access to the repository the day before the event.
https://github.com/Kiel-AI/Coding.Waterkant21

Datasets and Challenges
Work together with other participants on one of the following listed challenges or define a different project you would like to work on and present at the end of the hackathon.

Title	Patron	Description	What we think	Data
Prediction of Surfing Conditions	Kay Sörnsen & Jonas Kaufmann (Windfinder)	The local weather influences on a surf spot (thermals, wind direction, cloudiness etc.) are very unique and best known to the local surfers. They know best when the combination of different elements in the forecast is promising for a good day at their local spot. On such a day, the number of views on the forecast for this spot increases correspondingly, and the number of pageviews is therefore a good indicator for a good surfing day.	This is a classic supervised learning task and gives you the opportunity to test your skills on a task that has not yet been completed. It might even result in a new product that you could then further develop together with Windfinder after the hackathon.	Windfinder will provide weather station data from sevaral surfspots as well as the page views from the coresponding websites
Time Series Prediction for Bakery Turnovers	Thies Schönfeldt (Meteolytix)	Bakery sales largely depend on to the day of the week, on holidays or vacation, local festivals that tale place and many more. An important factor though is also the weather; depending on the location of the branch, this might even be one of the most important factors, which is why the inclusion of weather forecasts into the sales prediction has shown to be very beneficial.	Whether you want to gain first experience with time series predictions, train your first multidimensional LSTM, or maybe even a transformer model for time series prediction, this dataset gives you all the opportunities. And your challenge patron Thies from Meteolytics can give you direct feedback from the praxis in his company.	Meteolytics will provide daily turnover data for 13 different sales groups and three diffrent bakery branches (beach location, city center and periphery of the city) over a period of four years as well as correpsonding daily weather data.
Diagnosis of Vertebral Body Fractures	Claus Glueer (University of Kiel) & Valentina Pedoia (University of California, San Francisco)	Using radiological image data, vertebral bodies are to be automatically examined for the presence of fractures. The Genant Score (https://asbmr.onlinelibrary.wiley.com/doi/abs/10.1002/jbmr.5650080915) is used to classify the fractures.	Take the opportunity to get in contact with researchers from Kiel and San Francisco University and work together with them on a current research topic.	The University of Kiel will provide labelled X-ray scans of vertebral bodies to conduct the trainings.
Text Generation Using GPT	Doris Weßels (Kiel University of Applied Sciences)	Today’s AI-based language models allow to generate text that is basically not to distinguish from text written by humans anymore. In this challenge we will try to use the GPT-Neo to solve text writing tasks by automatically generating it with GPT-Neo and other freely available tools. GPT-Neo is one of the largest language models currently available and will be provided via a playground by Kiel.AI. To use the playground you don’t need to have any programming knowledge but you instruct the model via natural language to write about a given topic and in the form or structure needed. In order to get first knowledge on how to write these instructions (also called “prompts” in the machine learning context), we additionally provide a specific workshop at the beginning of the hackathon. If you prefer to use a programming interface with GPT-Neo, Kiel.AI can also provide you with an API access to GPT-Neo.	Instructing AI models via natural language is a very recent development and makes AI accessible for everyone. Take the opportunity to learn the corresponding brand new skill called "prompt designing" using the most powerful publicly available language model that has been published just a few weeks ago.	Training data is not needed but you will get access to the GPT-Neo model hosted by Kiel.AI.
Automated Essay Scoring	Sabrina Ludwig (Universität Mannheim) & Thorben Jansen (IPN Kiel)	In any learning context it is crucial to provide fast feedback to the learner. Being able to quickly analyze and categorize open text answers produced by the learner is therfore a a very important step in providing better and more efficient learning experiences. In this challenge we will therefore as a first step try to categorize open text answers that learners were providing on complex problem solving tasks according to the quality of their structure.	By fine-tuning current state-of-the-art language models and using them for text classification, this challenge gives you the opportunity to support the research in education and maybe go beyond what is currently known to be possible in this field of research.	The IPN will provide 900 scored texts and the University of Mannheim another set of about 1,800 scored texts.
Your Own Challenge	You	Besides the challenges provided by us, you are of cause also very welcome to work on your own challenge and maybe find fellow participants that want to work on it as well.	The access to GPT-Neo provided by Kiel.AI might be the start for your callenge. Whats's the task you want to automate with state-of-the-art language models?