Rafael Hostettler

It's almost time for the last seminar of the month! ⏰ Tuesday 30 April, Emre Neftci from the Forschungszentrum Jülich will be in Groningen to tell us more about 'Meta-learning for Memristive Devices'. In this talk, he will present methods to pre-train neural networks to be largely insensitive to non-idealities during learning tasks. 🧠 These include non-linearities in conductance updates, variations during operation, fabrication mismatch, conductance drift, and the realities of gradient descent training. If you're in or around Zernike, feel free to join us! More information ➡ https://lnkd.in/eFudQaN4 #physics #neural #networks #memristors #science #neuromorphic #brain #greencomputing #energyefficiency #CogniGron #FutureProofComputing

34

1 Kommentar

⭐ We аre happy to share a paper by researchers from INSAIT and ETH Zürich accepted at CVPR’24 (highlight), called “Matching Anything By Segmenting Anything” - see why! 🔍 The robust association of the same objects across video frames in complex scenes is crucial for many applications, especially Multiple Object Tracking (MOT). Current methods predominantly rely on labeled domain-specific video datasets, which limits the cross-domain generalisation of learned similarity embeddings. 💡 The paper proposes “MASA” -  a novel method for robust instance association learning, capable of matching any objects within videos across diverse domains without tracking labels. Leveraging the rich object segmentation from the Segment Anything Model (SAM), MASA learns instance-level correspondence through exhaustive data transformations. 📊 Extensive tests on multiple challenging MOT and MOTS benchmarks indicate that the proposed method, using only unlabeled static images, achieves even better performance than state-of-the-art methods trained with fully annotated in-domain video sequences, in zero-shot association. 🔗 More info: https://lnkd.in/dqN_zdfn 

186

2 Kommentare

Unleashing Next-Level Agility with Robotic Monkeys! Meet Charlie, the innovative robotic monkey from Deutsches Forschungszentrum für Künstliche Intelligenz (DFKI) that's pushing the boundaries of bioinspired robotics. Key features of Charlie: Seamless Locomotion: Transitions effortlessly between walking on four legs and standing upright on two. iStruct Spine: Mimics natural movement with its groundbreaking, flexible artificial spine. Fluid-Powered Muscles: Utilizes fluid-filled cylinders to replicate the power and control of real muscles. Why Charlie Matters: Transformative Potential: This technology offers exciting possibilities for advancements in healthcare and prosthetics. Biomimicry for Good: Charlie goes beyond simply mimicking nature; it paves the way for robots that can assist humans in various tasks. The Future of Human-Robot Interaction: Charlie's creation brings us closer to a world where robots seamlessly integrate into our lives, blurring the lines between organic and artificial movement. Stay tuned for more exciting robot news! video-credits - Fossbytes #ai #innovation #engineering #science #technology #machinelearning #germany #research #development #nextgen #future #automation #leggedrobots #animalrobots #bioinspireddesign #movement #flexibility #agility #futureofwork #artificialintelligence #robotics

4

This is our #LiDAR approach by 3D scanning landscapes, plants and ancient ruins. 🆓❗️In the comment section below I will provide you links with the #3dscanned ancient ruins that are available for free to everybody.

5

4 Kommentare

Today Quantum annealer improves understanding of quantum many-body systems by Forschungszentrum Juelich Artistic rendition of a quantum simulation of 1T-TaS2 being performed on the quantum processing unit of a quantum annealer Physicists have long been pursuing the idea of simulating quantum particles with a computer that is itself made up of quantum particles. This is exactly what scientists at Forschungszentrum Jülich have done together with colleagues from Slovenia. They used a quantum annealer to model a real-life quantum material and showed that the quantum annealer can directly mirror the microscopic interactions of electrons in the material. The result is a significant advancement in the field, showcasing the practical applicability of quantum computing in solving complex material science problems. Furthermore, the researchers discovered factors that can improve the durability and energy efficiency of quantum memory devices. The findings have been published in Nature Communications.

6

1 Kommentar

Are you an optical designer that writes custom code with #python, #matlab, or C++ (#cpp)? If so, Quadoa Optical CAD is the optical design software for you. From the first line of code, Quadoa was written with a modern codebase to support comprehensive and efficient scripting/API access. So much so that it can be run independent of the UI "out of the box". This enables the raytracing and optimization algorithms to be incorporated into other software workflows and even optical instruments. Check out this video to see Python scripting in action. #quadoa #opticaldesign https://lnkd.in/gB7zzDpU

11

🚀 Discover the world of 3D modeling with RealityKit Object Capture! In our newest blog post we’ll tell you how to create custom 3D models for visionOS with ease using RealityKit's Object Capture API. 🔎 Understanding photogrammetry? 🔎 Hands-on tips for capturing and processing images? 🔎 A guide to transforming simple photographs into stunning 3D assets? It has it all! #3D #RealityKit #visionOS

6

The new MISRA-C++:2023 Safety Guidelines by Peter Sommerlad at using std::cpp 2024. MISRA guidelines and corresponding static analysis tooling exist to foster programming practices that are considered important for safety-critical systems, especially in the automotive industry. In application areas where human live or capital is at risk, safe and secure code is a must, but even in other domains the internal quality of C++ programs is an important factor for programmers' effectiveness. Adhering to MISRA-C++ can help to achieve better internal quality, reducing the risk of undefined behavior and portability issues through implementation-defined behavior. Since the original MISRA-C++ 2008 guidelines that addressed C++03, five new C++ standards were released. While inherently conservative, an update to the MISRA-C++ guidelines that the presenter contributed to was released in 2023 adressing C++17. This talk will give an overview of why and what MISRA-C++:2023 attempts to achieve as well as demonstrate some example MISRA-C++:2023 rules and how contemporary C++ design already fulfils them. Some seemingly surprising guidelines are shown and explained, why such "unachievalbe" guidelines exist and the rationale behind those guidelines that one seemingly must violate. https://lnkd.in/dp9xbGPT

25

“Practical tips for DeepLabCut markerless tracking” workshop led by its developers. You can’t miss this opportunity! See you on August 16th in Bern ✨ Register now 🙌 ##swissneuroscience #deeplabcut

2

EEG Testing Day at Data Science Group Frankfurt: A Glimpse into Brain-Computer Interface Research🧠🔬 Today is a thrilling day for the Data Science Group Frankfurt as we delve into the fascinating world of EEG (electroencephalogram) testing! Team members, Catalina Jaramillo, Michael Nordine, Markus Ketomäki, Johanna Keim and Hans Magnus Haveraaen have volunteered as test subjects for our Master’s student and data scientist; Lea Grebe’s thesis on Non-invasive, EEG-assisted Brain Computer Interfaces, developed in collaboration with Johannes Gehrig from the Department of Neurology, aimed at improving patient communication. 🧠🩺💬 Equipment 🔬 The testing procedure involves equipping participants with the Unicorn Hybrid Black EEG cap, which is integral to capturing the brain’s electrical activity. The initial setup may require adjustments, including the application of conductive gel to ensure optimal signal transmission. Procedure 🔍 Our software undergoes a series of tests, each lasting 10 minutes, to validate its effectiveness. In these tests, visual stimuli – circles labeled with letters ‘A’ or ‘B’, and patterns forming human shapes – are presented. The purpose is not to challenge the test subjects but to ensure the software accurately detects when a subject is focusing on a specific circle. This detection is facilitated by the P300 response, a brainwave event that exhibits increased amplitude for task-relevant stimuli, indicating the subject’s attention. The P300 response 💯 The P300 response occurs approximately 300 milliseconds after a person notices something unexpected or significant within a series of regular events. Essentially, the P300 is a signal that the brain generates when it recognizes something important or different, like spotting a friend in a crowd. This makes it useful for studying attention and perception. The goal 🎯 The primary goal of the project is to provide patients with an uncomplicated and efficient method of conveying their requirements. The development of non-invasive brain-computer interfaces marks a considerable progression in facilitating effortless communication for patients. We invite you to follow our journey as we delve deeper into the potential of #BCIs in patient care! Visit our website to discover more about our diverse research endeavors 👉http://dsgfrankfurt.de/ #EEGTesting #BrainComputerInterface #PatientCommunication #CognitiveScience #Innovation g.tec medical engineering GmbH Universitätsklinikum Frankfurt

38

2 Kommentare

🌐 Groundbreaking Compression Standard for Haptics Published! 🌐 The Technical University of Munich (TUM) has led an eight-year effort culminating in the publication of the first standard for the compression and transmission of the sense of touch, named "Haptic Codecs for the Tactile Internet" (HCTI). This pioneering standard lays the groundwork for advancements in tele-surgery, tele-driving, and immersive gaming experiences. Haptic codecs, much like JPEG for images and MP3 for audio, enable the efficient transmission of tactile information over the Internet. The new IEEE standard 1918.1.1 encodes and decodes data to allow users to experience remote tactile interactions, such as feeling the grip of a robot arm or the texture of materials like paper or metal. Prof. Eckehard Steinbach, leading the research at TUM, highlighted the challenge of transmitting haptic feedback, which requires bidirectional data flow and ultra-low latency to maintain realistic interactions. The new standard reduces the clock rate from 4,000 to 100 times per second, aligning closely with human perception thresholds. The applications of HCTI are vast: ⚕️ Telesurgery: Enables surgeons to perform operations remotely with precision, even across long distances. 🚚 Teledriving: Facilitates remote control of vehicles, enhancing safety and flexibility in autonomous driving. 💉 Emergency Medical Services: Allows remote ultrasound diagnostics by doctors while en route to hospitals. 🎦 Entertainment: Enhances realism in gaming and films by transmitting physical sensations. 💻 E-commerce: Provides tactile feedback for online shoppers, enhancing the shopping experience. This new standard is expected to spur innovation and new applications, much like JPEG and MP3 did for their respective fields. For more details, visit TUM's official website -https://lnkd.in/eDfvSz_x To access the standard, visit IEEE here - https://lnkd.in/eenTQAaH

29

A New Generation of Networked UxV Systems Auterion, manufacturer of an open operating system for the control of autonomous robotic systems, and blackned gmbh, which provides a core element for the digitalization of land forces with its TACTICAL CORE middleware, are entering into a strategic partnership to develop an integrated technology for the implementation and control of #UxV in the digital battlefield. The aim of this partnership is to apply Auterion’s manufacturer-independent control software for autonomous robotic systems to the information and communication network of the digital battlefield. This enables a faster distribution of reconnaissance data and the acceleration of the sensor-to-shooter chain through the integration of UxV into the TACTICAL CORE system network. This makes the intelligence networks denser and more robust. The monitoring and control of UxV is therefore no longer limited to manufacturer-specific systems but can be taken over from different starting and location points of the battlefield via a common platform in accordance with the software-defined defense (SDD) principle. #tacticalcore #uxv #uav Read more: https://lnkd.in/enzmkPyN

63

2 Kommentare

【New Invention for Tech Transfer】 Efficient solution of constrained quadratic programming problem in quantum annealer 　 One Pager 　[ENG]　https://lnkd.in/gijrqNPv 　 Contact to request for tech transfer. https://lnkd.in/gz2dUar2 #technology #TLO #techtransfer #quantumcomputer #quantumannealing #quadraticprogramming

8

Turning smallest #sensor nodes into top #AI performers: Rick Pandey, Sebastian Uziel, Tino Hutschenreuther and Silvia Krug will present their new TEEMSC #algorithm that can be fully trained on a #sensor node. It outperforms existing methods for on-device learning. The talk will be at the 2024 IEEE International Workshop on #Metrology for #Industry40 & #IoT in Florence, Italy: https://metroind40iot.org ➡ Why: #predictivemaintenance (PdM) has been a challenge due to insufficient labelled data and dynamic #industrial environments. These problems require a data driven solution for machine degradation analysis which is capable of on-device learning and adapting to the changing industrial environments. Existing data driven unsupervised methods mostly do anomaly detection and the #neuralnetwork-based solutions either rely upon a cumbersome data acquisition and labelling process or suffer from concept drift caused due to changing environmental scenarios. ➡ How: The new TEEMSC algorithm uses singular values and canonical crosscorrelation and enables #energyefficient #machine #diagnosis. TEEMSC can be trained completely on a #sensor node. The learned parameters are adaptive to dynamic industrial scenarios. TEEMSC learns from unlabelled data in mainly unsupervised manner. ➡ #Test: Considering bearing degradation as a PdM scenario the authors have tested the algorithm on PRONOSTIA and XJTU bearing dataset and compared performance of TEEMSC with three existing degradation trend analysis methods namely Principal Component Analysis, Power Spectral Density and Kurtosis on the wavelet decomposition of the original vibration signal. ➡ Performance: Their results can clearly outperform existing unsupervised degradation trend analysis and anomaly detection methods for on-device learning. The algorithm has been developed in the HoLoDEC project. Learn more on the project under www.imms.de/holodec or contact Sebastian Uziel directly. The HoLoDEC project is funded by the German Federal Ministry of Education and Research under the reference 16ME0703.

13

💦💦SemanticSpray: self-driving in wet conditions💦💦 👉#BMW (+ULM) unveils SemanticSpray: novel dataset contains scenes in wet surface conditions captured by camera, LiDAR and radar. The following label types are provided: Camera: 2D Boxes | LiDAR: 3D Boxes, Semantic Labels | Radar: Semantic Labels. #Github available💙 𝐇𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬: ✅Adapt the energy-based outlier detection framework ✅Detecting adverse weather in LiDAR point clouds ✅New SOTA, robustness in unseen weather effects ✅Outlier detection/segmentation of the inlier classes ✅SemanticSpray dataset for research #artificialintelligence #machinelearning #ml #AI #deeplearning #computervision #AIwithPapers #metaverse 👉Discussion https://lnkd.in/dMgakzWm 👉Paper https://lnkd.in/dnN5MCZC 👉Project https://lnkd.in/dkUaxyEF 👉Data https://lnkd.in/ddhkyXv8

291

3 Kommentare