Author | Lightnews

Reposted by Djordje Slijepčević

TU Wien @tuwien.at · 7d

🎉 Drei Auszeichnungen für TU Wien-Forschung!
Rémi Hocq, Hermann Hofbauer und Djordje Slijepčević wurden für ihre Forschung zu Industrieabgas-Nutzung, Energietechnologien und KI in der Medizin geehrt. Wissenschaft mit Wirkung – made at TU Wien.
🔗 www.tuwien.at/tu-wien/aktu...

Auszeichnungen für Forschende der TU Wien

Ob Industrieabgase, Energietechnologien oder KI in der Medizin – Forschende der TU Wien liefern Antworten auf zentrale Fragen unserer Zeit. Drei aktuelle Auszeichnungen für Rémi Hocq, Hermann Hofbauer...

www.tuwien.at

2 4

Djordje Slijepčević @dslijepcevic.bsky.social · 15d

There are certain limitations but our results add to the #verification vs. #validation debate in #gait #analysis. #Marker-based and #markerless systems may never yield perfectly matching #kinematics, but maybe they don't need to. We must understand their differences to use them where they work best.

Djordje Slijepčević @dslijepcevic.bsky.social · 15d

📊 Key results:
• Comparable performance to #OpenCap
• Challenges in #ankle #kinematics
• Reliable kinematics (RMSD ≈ 3°)
• Reasonable accuracy (RMSD ≈ 5.5°)
→ promising for #remote & #home-based #gait #assessment

1

Djordje Slijepčević @dslijepcevic.bsky.social · 15d

#Markerless 3D #motion #capture keeps moving forward and #monocular methods could make gait analysis even more accessible. In this study, we propose a low-cost monocular system, combining #CameraHMR & #OpenSim against both #OpenCap and a marker-based reference under various gait conditions.

1

Djordje Slijepčević @dslijepcevic.bsky.social · 15d

🚀 New paper:
“Validity and reliability of monocular 3D #markerless #gait #analysis in simulated pathological gait: A comparative study with #OpenCap” @jbiomech.bsky.social :
📄 doi.org/10.1016/j.jb...

👥 @bhorsak.bsky.social, M. Simonlehner, V. Quehenberger, B. Dumphart & A. Kranzl

#BiomechSky

Redirecting

doi.org

1 2

Djordje Slijepčević @dslijepcevic.bsky.social · 23d

Example of the functioning of our Speech Concept Bottleneck Models #SCBM

📄https://doi.org/10.1016/j.ipm.2025.104309

#XAI #explainability #interpretability #ML #AI #hatespeech #hate #counterspeech #toxic

Djordje Slijepčević @dslijepcevic.bsky.social · 23d

This work is a step toward AI systems people can:
✔️ Understand
✔️ Trust
✔️ Use responsibly

Big thanks to my coauthors:
Roberto Labadie, Xihui Chen, Adrian Jaques Böck, Andreas Babic, Liz Freimann, Christiane Atzmüller & Matthias Zeppelzauer

Djordje Slijepčević @dslijepcevic.bsky.social · 23d

📊 Results:
✅ #SCBM balance interpretability & performance
✅ They achieve (or surpass) state-of-the-art results
✅ Compete even with large #Transformers

1

Djordje Slijepčević @dslijepcevic.bsky.social · 23d

✨ What’s inside?
🔹 Distilling knowledge from #LLMs into transparent #SCBM
🔹 Using domain-specific human-understandable adjectives as concepts
🔹 A novel class-discriminative loss for #interpretability
🔹 Applied to #hate & #counterspeech recognition

1

Djordje Slijepčević @dslijepcevic.bsky.social · 23d

🧩The idea: Instead of predicting directly, #SCBM goes through a layer of human-understandable concepts. We leverage #LLMs to evaluate adjective-based concepts that capture the intention & tone on #socialmedia .
This way, the model explains why it decides.

1

Djordje Slijepčević @dslijepcevic.bsky.social · 23d

In our new paper "Distilling knowledge from large language models: A #concept #bottleneck #model for #hate and #counterspeech recognition" we introduce Speech Concept Bottleneck Models #SCBM - a step toward #interpretable #LLM.
📄https://doi.org/10.1016/j.ipm.2025.104309

1

Djordje Slijepčević @dslijepcevic.bsky.social · 23d

AI systems must be accurate but also #understandable. Trust in #AI depends on #transparency. Yet Large Language Models (#LLMs) remain #blackboxes.
How do we make them #interpretable without losing performance? 👇

#Explainability #NLP #XAI #OnlineHate #ML #Interpretability

1 1

Djordje Slijepčević @dslijepcevic.bsky.social · Jul 4

Congrats, Anne!

1

Djordje Slijepčević @dslijepcevic.bsky.social · Jul 4

☕ I'll be around until Wednesday – happy to connect over coffee, just drop me a DM!

#ESB2025 #ESB #Biomechanics #AIinBiomechanics #ExplainableAI #GaitAnalysis #HumanMovement #ML #XAI

Djordje Slijepčević @dslijepcevic.bsky.social · Jul 4

Make sure to catch @bhorsak.bsky.social talk on:
🎤 “𝗕𝗥𝗘𝗔𝗞𝗜𝗡𝗚 𝗕𝗔𝗥𝗥𝗜𝗘𝗥𝗦 𝗜𝗡 𝗛𝗨𝗠𝗔𝗡 𝗠𝗢𝗧𝗜𝗢𝗡 𝗖𝗔𝗣𝗧𝗨𝗥𝗘 𝗪𝗜𝗧𝗛 𝗔𝗜-𝗣𝗢𝗪𝗘𝗥𝗘𝗗 𝗟𝗢𝗪-𝗖𝗢𝗦𝗧 𝗠𝗔𝗥𝗞𝗘𝗥𝗟𝗘𝗦𝗦 𝗦𝗬𝗦𝗧𝗘𝗠𝗦”
🕥 Room G5 | Tuesday 08/July/2025, 10:25 – 10:45am

#ESB2025 #ESB

1 1

Djordje Slijepčević @dslijepcevic.bsky.social · Jul 4

👣 Looking forward to attending the @esbcongress.bsky.social at @ethz.ch next week!

I'd like to invite you to my perspective talk:

🎤 “𝗨𝗡𝗕𝗢𝗫𝗜𝗡𝗚 𝗧𝗛𝗘 𝗕𝗟𝗔𝗖𝗞-𝗕𝗢𝗫: 𝗘𝗫𝗣𝗟𝗔𝗜𝗡𝗜𝗡𝗚 𝗠𝗔𝗖𝗛𝗜𝗡𝗘 𝗟𝗘𝗔𝗥𝗡𝗜𝗡𝗚 𝗠𝗢𝗗𝗘𝗟𝗦 𝗜𝗡 𝗛𝗨𝗠𝗔𝗡 𝗠𝗢𝗩𝗘𝗠𝗘𝗡𝗧 𝗔𝗡𝗔𝗟𝗬𝗦𝗜𝗦”
🕗 Audi Max | Tuesday 08/July/2025, 8:30 – 8:50am

#ESB2025 #ESB

1 1

Djordje Slijepčević @dslijepcevic.bsky.social · May 2

This is a new approach to #personalized #rehabilitation based on #ML, unlocking novel opportunities in #injury #prevention and therapy. Our primary goal is to support clinicians and #patients in improving therapeutic outcomes.

#AI #MachineLearning #Rehabilitation #Sonification #GaitAnalysis

Djordje Slijepčević @dslijepcevic.bsky.social · May 2

A huge thanks to my co-authors for the great collaboration: De-Jesus-Oliveira V., Dumphart B., Ferstl S., Joschua Reis, Raberger A.-M., Heller M., @bhorsak.bsky.social, Michael Iber, Simonlehner M., Prock K., Siragy T.

1 1

Djordje Slijepčević @dslijepcevic.bsky.social · May 2

📚 Simonlehner et al. (2024). 𝗦𝗼𝗻𝗶𝗳𝗶𝗰𝗮𝘁𝗶𝗼𝗻 𝗰𝗮𝗻 𝗮𝗹𝘁𝗲𝗿 𝗝𝗼𝗶𝗻𝘁 𝗔𝗹𝗶𝗴𝗻𝗺𝗲𝗻𝘁 𝗳𝗼𝗿 𝗣𝗲𝗿𝘀𝗼𝗻𝗮𝗹𝗶𝘇𝗲𝗱 𝗥𝗲𝗵𝗮𝗯𝗶𝗹𝗶𝘁𝗮𝘁𝗶𝗼𝗻: 𝗘𝘃𝗶𝗱𝗲𝗻𝗰𝗲 𝗳𝗿𝗼𝗺 𝗮 𝗖𝗼𝗻𝘁𝗿𝗼𝗹𝗹𝗲𝗱 𝗣𝗶𝗹𝗼𝘁 𝗦𝘁𝘂𝗱𝘆. 𝘎𝘢𝘪𝘵 & 𝘗𝘰𝘴𝘵𝘶𝘳𝘦, 113, 212–213.
doi.org/10.1016/j.ga...

Redirecting

doi.org

1

Djordje Slijepčević @dslijepcevic.bsky.social · May 2

Our contributions include:

📚 de Jesus Oliveira et al. (2023). 𝗔𝘂𝗱𝗶𝘁𝗼𝗿𝘆 𝗳𝗲𝗲𝗱𝗯𝗮𝗰𝗸 𝗶𝗻 𝘁𝗲𝗹𝗲-𝗿𝗲𝗵𝗮𝗯𝗶𝗹𝗶𝘁𝗮𝘁𝗶𝗼𝗻 𝗯𝗮𝘀𝗲𝗱 𝗼𝗻 𝗮𝘂𝘁𝗼𝗺𝗮𝘁𝗲𝗱 𝗴𝗮𝗶𝘁 𝗰𝗹𝗮𝘀𝘀𝗶𝗳𝗶𝗰𝗮𝘁𝗶𝗼𝗻. 𝘗𝘦𝘳𝘴𝘰𝘯𝘢𝘭 𝘢𝘯𝘥 𝘜𝘣𝘪𝘲𝘶𝘪𝘵𝘰𝘶𝘴 𝘊𝘰𝘮𝘱𝘶𝘵𝘪𝘯𝘨 27, 1873–1886 (2023).
doi.org/10.1007/s007...

Auditory feedback in tele-rehabilitation based on automated gait classification - Personal and Ubiquitous Computing

In this paper, we describe a proof-of-concept for the implementation of a wearable auditory biofeedback system based on a sensor-instrumented insole. Such a system aims to assist everyday users with s...

doi.org

1

Djordje Slijepčević @dslijepcevic.bsky.social · May 2

We #𝘀𝗼𝗻𝗶𝗳𝘆 𝗵𝘂𝗺𝗮𝗻 #𝗺𝗼𝘃𝗲𝗺𝗲𝗻𝘁 and 𝗮𝗹𝗲𝗿𝘁 𝗽𝗮𝘁𝗶𝗲𝗻𝘁𝘀 𝘄𝗵𝗲𝗻 𝗺𝗶𝘀𝗮𝗹𝗶𝗴𝗻𝗺𝗲𝗻𝘁𝘀 𝗼𝗰𝗰𝘂𝗿, whether during #walking or #rehabilitation exercises. Using sensor-based #insoles and #ML, we aim to detect deviations from ideal posture or movement and provide 𝗿𝗲𝗮𝗹-𝘁𝗶𝗺𝗲 𝗮𝘀𝘀𝗶𝘀𝘁𝗮𝗻𝗰𝗲 𝘁𝗵𝗿𝗼𝘂𝗴𝗵 𝘀𝗶𝗺𝗽𝗹𝗲 𝗮𝗻𝗱 𝗶𝗻𝘁𝘂𝗶𝘁𝗶𝘃𝗲 𝗮𝘂𝗱𝗶𝗼 𝗰𝘂𝗲𝘀.

1

Djordje Slijepčević @dslijepcevic.bsky.social · May 2

Proud to announce that our team is one of the 𝘄𝗶𝗻𝗻𝗲𝗿𝘀 of the 𝟐𝟎𝟐𝟓 𝗗𝗮𝘁𝗮 𝗦𝗼𝗻𝗶𝗳𝗶𝗰𝗮𝘁𝗶𝗼𝗻 𝗔𝘄𝗮𝗿𝗱𝘀! 🏆 This recognition from the community affirms the impact of our interdisciplinary research in #𝗦𝗼𝗻𝗶𝗳𝗶𝗰𝗮𝘁𝗶𝗼𝗻, #𝗠𝗮𝗰𝗵𝗶𝗻𝗲𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴, #𝗚𝗮𝗶𝘁𝗔𝗻𝗮𝗹𝘆𝘀𝗶𝘀, and #𝗥𝗲𝗵𝗮𝗯𝗶𝗹𝗶𝘁𝗮𝘁𝗶𝗼𝗻.

phaidra.fhstp.ac.at/api/object/o...

phaidra.fhstp.ac.at

1 1

Djordje Slijepčević @dslijepcevic.bsky.social · Apr 3

🚀 𝗧𝗿𝘆 𝗶𝘁 𝗼𝘂𝘁 𝗳𝗼𝗿 𝘆𝗼𝘂𝗿𝘀𝗲𝗹𝗳!
🔗 𝗗𝗼𝘄𝗻𝗹𝗼𝗮𝗱 𝗵𝗲𝗿𝗲: github.com/fhstp/Intell...
💻 𝗡𝗼 𝗶𝗻𝘀𝘁𝗮𝗹𝗹𝗮𝘁𝗶𝗼𝗻 𝗻𝗲𝗲𝗱𝗲𝗱: just 𝗱𝗼𝘄𝗻𝗹𝗼𝗮𝗱, 𝘂𝗻𝗽𝗮𝗰𝗸, and 𝗿𝘂𝗻 in Vicon Nexus 2.14+

💡 𝗟𝗼𝗼𝗸𝗶𝗻𝗴 𝗮𝗵𝗲𝗮𝗱: Stay tuned for an even larger evaluation coming soon!

#Gait #GaitAnalysis #Biomechanics #Kinematics #MachineLearning #DeepLearning

This link will take you to a page that’s not on LinkedIn

lnkd.in

1

Djordje Slijepčević @dslijepcevic.bsky.social · Apr 3

Great work by Bernhard Dumphart and congrats to our entire team @bhorsak.bsky.social, Kranzl Andreas, Fabian Unglaube, Baca Arnold, Fabien Leboeuf, Stéphane Armand, Marys Franco Carvalho, Florian Dobler, Nathalie Alexander.

1 1

Djordje Slijepčević @dslijepcevic.bsky.social · Apr 3

In our multicenter study 𝗜𝗻𝘁𝗲𝗹𝗹𝗘𝘃𝗲𝗻𝘁 showed 𝘀𝘁𝗿𝗼𝗻𝗴 𝗴𝗲𝗻𝗲𝗿𝗮𝗹𝗶𝘇𝗮𝗯𝗶𝗹𝗶𝘁𝘆 in detecting initial contact and foot-off events and performed well even on 𝗽𝗿𝗲𝘃𝗶𝗼𝘂𝘀𝗹𝘆 𝘂𝗻𝘀𝗲𝗲𝗻 #𝗽𝗮𝘁𝗵𝗼𝗹𝗼𝗴𝗶𝗲𝘀 .

📄https://doi.org/10.1016/j.gaitpost.2025.01.049

#AI #ML #Gait #Biomechanics #Kinematics #MachineLearning #DeepLearning

1