BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//scienceofintelligence.de - ECPv6.15.12.2//NONSGML v1.0//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
X-ORIGINAL-URL:https://www.scienceofintelligence.de
X-WR-CALDESC:Events for scienceofintelligence.de
REFRESH-INTERVAL;VALUE=DURATION:PT1H
X-Robots-Tag:noindex
X-PUBLISHED-TTL:PT1H
BEGIN:VTIMEZONE
TZID:Europe/Berlin
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:20220327T010000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:20221030T010000
END:STANDARD
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:20230326T010000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:20231029T010000
END:STANDARD
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:20240331T010000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:20241027T010000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTART;TZID=Europe/Berlin:20230713T100000
DTEND;TZID=Europe/Berlin:20230713T110000
DTSTAMP:20260409T045342
CREATED:20230605T103302Z
LAST-MODIFIED:20240813T103032Z
UID:15693-1689242400-1689246000@www.scienceofintelligence.de
SUMMARY:Santiago Paternain\, "Safe Learning for Dynamical Systems and Control"
DESCRIPTION:Abstract: \nReinforcement learning has shown great success in controlling complex dynamical systems. However\, when training a policy\, most algorithms only consider a single objective function. While this may suffice in virtual domains\, physical systems must satisfy a set of operational constraints\, with safety being of crucial importance. It is natural to express these problems as constrained optimization problems since weighted combinations of different rewards are not guaranteed to find a solution that satisfies all the requirements. Furthermore\, these examples are not contrived\, and safety-constrained reinforcement learning is a vital area of research that needs to be tackled. \nAfter establishing the need to tackle safety-constrained reinforcement learning\, I will shift my focus to solving these generally non-convex problems. I will discuss different approaches that exploit duality theory to pave the way towards algorithms for general constrained reinforcement learning. In particular\, I will discuss that (i) despite their non-convexity these problems have zero duality gap\, (ii) a state-augmented approach that does not guarantee convergence to an optimal policy but\, it guarantees optimality and (iii) a safe policy-gradient theorem that allows us to consider constraints beyond time-averages. \nThis talk will take place in person at SCIoI. \n  \nPhoto by Jeswin Thomas on Unsplash \n 
URL:https://www.scienceofintelligence.de/event/thursday-morning-talk-santiago-paternain/
CATEGORIES:Thursday Morning Talk
ATTACH;FMTTYPE=image/jpeg:https://www.scienceofintelligence.de/wp-content/uploads/2023/06/jeswin-thomas-dfRrpfYD8Iw-unspla.jpg
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Europe/Berlin:20230720T100000
DTEND;TZID=Europe/Berlin:20230720T110000
DTSTAMP:20260409T045342
CREATED:20230605T103948Z
LAST-MODIFIED:20240813T103027Z
UID:15696-1689847200-1689850800@www.scienceofintelligence.de
SUMMARY:Lisa-Kristin Richter\, "Model Training for Facial Recognition of Raccoons"
DESCRIPTION:Machine learning tools have already been used to identify individual animals such as but not limited to pandas\, black bears\, cows and dogs. These tools can help to improve the quality of non-invasive wildlife monitoring and enhance the information on individual animal behaviour as well as on behaviour within social networks of the animals (Lynn 2019; Schofield et al. 2019). \nRaccoons (proctorloco) are considered an invasive species in Germany that has been introduced to many parts of the world outside of their native range in North America. \nIn order to train a model for facial recognition of raccoons\, we collected 7812 pictures of 133 individuals. After manual selection for quality focusing on sharpness\, image detail and light\, 111individuals with 4000 pictures remain in the dataset. The individuals were pictured in more than 10 different facilities and locations with different lights and angles. From this baseline dataset\, one data set using bounding boxes is created for training and one dataset using masks is also created for training. This is done to keep the influence of the background minimal. \nFinally\, this data is used to train different pre-trained deep learning models from Image Net\, namely ResNet50\, VGG19 and Mobile Net. While model training parameters like batch size\, number of epochs\, learning rate scheduler\, picture augmentation techniques and more are being varied. \nChallenges arise from the time and computer resources needed for training.Currently\, training is done via Google Colab\, which disconnects after a certain time. Furthermore\, input on dataset preprocessing\, model selection\, possible combination of models and variation in parameters would be very helpful. \nThis talk will take place in person at SCIoI. \nPhoto by Lukas Stoermer on Unsplash \n 
URL:https://www.scienceofintelligence.de/event/thursday-morning-talk-lisa-kristin-richter/
LOCATION:MAR 2.057
CATEGORIES:Thursday Morning Talk
ATTACH;FMTTYPE=image/jpeg:https://www.scienceofintelligence.de/wp-content/uploads/2023/06/lukas-stoermer-C0hX1cRQPOY-unspl.jpg
END:VEVENT
END:VCALENDAR