This year’s DataFest will focus on the analysis of sensor data streams for tool tracking applications. The usage of manual tools in industry, especially hand tools, continues to be necessary due to the flexibility of human labor. While automated processes assure quality and traceability, yet manual labor introduces “gaps” into the quality assurance process, i.e., we cannot ensure high quality standards automatically. This is not only undesirable but even intolerable in many cases. We introduce a system that monitors the process using inertial, magnetic field and audio sensors that we attach as add-ons to hand-held tools. The sensor data is analyzed via embedded classification algorithms and our system directly provides feedback to workers during the execution of work processes.
This year, for the projects the Tool Tracking Dataset serves as a basis. On the respective pages you will find documentation of the datasets and sample source code that enables you to load and visualize data quickly.
Possible Research Directions
There is a number of research direction to work with on the given dataset. We will not provide any strict tasks to work on as it is your task to decide what to do. Research areas that might be of interest cover the wide area of machine learning. Some of them might include the robustness to anomalies, active learning, self-supervised learning, multi-task learning, deep compression, automatic machine learning, innovative visualizations, and explainable machine learning. However, this is a non-exclusive list. Feel free to come up with an unconventional and different idea!
There is also a variety of performance metrics to judge the research impact. Besides classification performance, there is also a need to have small models that allow an embedded execution and/or feature extraction pipelines that are optimized for energy consumption. But also in the field of active learning and multi-task learning it is crucial to provide the classifiers with a minimum amount of labeled data samples as data annotation is costly. In summary, there are many metrics in this application that can be optimized.
Form a team of 3-5 people, inspire yourself by looking at the data.
How to find a team?
- Register as a team (write e-mail to email@example.com and put your names in and your teammates in CC)
- Register alone (write e-mail to firstname.lastname@example.org and I will put your name and mail address here (you will receive the password upon registration); if you already have ideas/research interests feel free to write them in the mail and I also put them on this page). You can get in contact with the people whose names we list here.
- We aim at having a few balanced teams (but max 10). Participation will be granted on a first-come-first-served basis.
What do we expect from the teams and what is the timeline?
- Position Paper & Pitch at a kickoff zoom meeting (24.08.2020 10:00). (1) prepare a one-page position paper that outlines the general idea that is to be investigated by the team (Deadline: 17.08.2020 EOD (end-of-day)), and (2) pitch the idea with max. 3 slides in max. 5 minutes to the audience.
- Work on your topic. You and your team have almost 4 weeks to work on your idea and topic. We will provide weekly zoom sessions for status updates and Q&A (although you can always ask questions via e-mail, especially about the provided data).
- Final Paper: document and sell your contribution. Write a paper with max. 8 pages (ICLR template in latex from https://github.com/ICLR/Master-Template/blob/master/archive/iclr2020.zip) and outline your motivation and idea, explain your contribution and provide experimental results and evaluation. The paper must be submitted via e-mail to email@example.com by (16.09.2020 EOD)
- Demo Presentation: present your results (presentation with “interactive” demo; 10 minutes presentation + 5 minutes questions) in a final zoom meeting on (18.09.2020 10:00).
|17.08.2020||Submission deadline for position paper and pitch slides|
|24.08.2020 10:00||Pitches of the ideas|
|16.09.2020||Paper submission deadline for solutions|
|18.09.2020 10:00||Presentations and demos|
How do we form grades?
We will give points for ideation, innovation & potential, demo presentation, paper and coding results.
Depending on your study programme you can use this project:
- Data Science: “Data Fest” (part of Current Research)
- Statistics: 3 ECTS (“Ausgewählte Gebiete…”)
- Computer Science: 3 ECTS Elective Masters Module (“ein halbes Masterwahlmodul”)
See the internal page here
Frequently Asked Questions (FAQ)
No questions so far.