idw – Informationsdienst Wissenschaft
Nachrichten, Termine, Experten
Nachrichten, Termine, Experten
https://www.hswt.de/newsroom/veranstaltungskalender/detail/fachsymposium-artific...
Key-Note-Speaker: Prof. Dr. Sepp Hochreiter
Johannes Kepler University Linz
European AI pioneer, inventor of the LSTM technology, and author of the book „What Can AI Do?“
Dexiong Chen
Max Planck Institute of Biochemistry
From language modeling to graph modeling: Transformers are Scalable Graph Generators
Building on recent advances in language modeling and graph learning, this talk introduces a novel framework for graph generation that leverages language models. Our approach converts graphs into random sequences through a reversible random walk sampling process, thereby reframing the generation task as language modeling. This approach opens new possibilities for applications in molecular and protein design.
Christoph Behr and Tobias Kolb
University of Applied Sciences Weihenstephan-Triesdorf
Development and Validation of a Vibration sensor for state recognition of CHP-Machinery
Increasing costs for maintenance and downtime, along with the need for more flexible and optimized operating schedules, are becoming major challenges in CHP operation. Therefore, methods and systems for monitoring and evaluation of machinery are needed. At the core of the developed system was an acceleration sensor, mounted on the engine block. The acceleration data was fast Fourier transformed into a frequency spectrum. On this spectral data a random forest model was trained. The training data consisted of various fault conditions. The goal was the determination of type and detail on which the different fault conditions could be recognized. Results show possibilities of reliable state recognition and engine part assessment. While the results show promising potential, they also reveal the current boundaries of the system.
Finja Schaumann and Sebastian Burkhart
University of Applied Sciences Weihenstephan-Triesdorf
Identifying Heat Stress in Bumblebee Colonies Based on Audio Recordings
Habitat loss, pesticide use, and climate change have been identified as the main risk factors for pollinators, which are ecologically and economically important yet globally declining. Improved monitoring methods are urgently needed to gather real-time, multi-scale information across spatial and temporal dimensions to better understand how pollinators respond to these stressors. Within the EU project PHENET (https://www.phenet.eu/en/about-phenet/use_cases/uc8), we are investigating the potential of Bombus terrestris colonies equipped with sensor arrays - including microphones and thermometers - as bioindicators for detecting and disentangling multiple stressors. AI-driven models are trained on datasets from controlled field experiments. In collaboration with Hochschule Weihenstephan-Triesdorf (HSWT), this work leverages the AI infrastructure provided by the AI4Life project (https://www.hswt.de/forschung/projekte/1755-ai4life) for the computationally intensive data processing required for training. Preliminary findings suggest that audio frequency spectra enable differentiation between control and heat-stressed bumblebee colonies. This pilot indicates the potential of non-invasive stress detection via acoustic patterns, laying the groundwork for scalable pollinator monitoring.
Ehsan Yaghoubi
University of Applied Sciences Weihenstephan-Triesdorf
Cross-modal uncertainty-based fusion for animal behavior analysis using audio
Animal vocalizations carry rich information about their emotional state, context (e.g., playing, eating, isolation), health condition, and more. In this study, we propose a dual-stream model that leverages both frequency-domain and time-series features of animal sounds to classify their behavioral context. One stream analyzes spectrogram representations, while the other processes raw audio waveforms. We incorporate uncertainty analysis to estimate the confidence of each stream and use this information to guide feature fusion. We evaluate our approach on two publicly available datasets, SoundWel (pig calls) and DogBarks, and compare it with existing methods. The results demonstrate the effectiveness of our proposed method in improving classification accuracy and robustness.
Malte von Bloh
Technical University Munich
EcoMapper: Generative Modeling for Climate-Aware Satellite Imagery
Satellite imagery plays a central role in applications such as crop yield prediction, environmental monitoring, and climate change assessment, yet its integration with climate data remains limited. We present EcoMapper, a novel dataset of 2.9 million Sentinel-2 images linked with climate variables across 15 land cover types, enabling scalable climate-aware remote sensing. Building on this dataset, we fine-tune Stable Diffusion 3 models for two generative tasks: (i) text-to-image generation from structured geographic-climate prompts, and (ii) multi-conditional generation with ControlNet, which preserves spatial structure and supports time-series synthesis. Our models generate realistic, climate-conditioned satellite images at up to 10-meter resolution, filling gaps in cloud-covered regions and enabling scenario-based simulations of land surface change. This work advances generative modeling for Earth observation, offering new opportunities for forecasting, climate adaptation, and geospatial analysis.
Information on participating / attending:
Online Event
free of charge
Register here to the symposium: https://hswt.zoom.us/meeting/register/sFRgUP4ORGOv0k4BYQ23ow#/registration
Date:
10/24/2025 10:00 - 10/24/2025 15:00
Event venue:
University of Applied Sciences
Weihenstephan-Triesdorf
Online Event
Bayern
Germany
Target group:
Business and commerce, Scientists and scholars
Email address:
Relevance:
international
Subject areas:
Environment / ecology, Information technology, Zoology / agricultural and forest sciences
Types of events:
Conference / symposium / (annual) conference
Entry:
10/20/2025
Sender/author:
Gerhard Radlmayr
Department:
Zentrum für Forschung und Wissenstransfer
Event is free:
yes
Language of the text:
English
URL of this event: http://idw-online.de/en/event80265
Programme You can combine search terms with and, or and/or not, e.g. Philo not logy.
You can use brackets to separate combinations from each other, e.g. (Philo not logy) or (Psycho and logy).
Coherent groups of words will be located as complete phrases if you put them into quotation marks, e.g. “Federal Republic of Germany”.
You can also use the advanced search without entering search terms. It will then follow the criteria you have selected (e.g. country or subject area).
If you have not selected any criteria in a given category, the entire category will be searched (e.g. all subject areas or all countries).
