Maven.Pet’s Post

🔗 Exciting news! Check out the latest blog post on "Multi-Task Media-Bias Analysis Generalization for Pre-Trained Identification of Expressions." This groundbreaking post introduces MAGPIE, a large-scale multi-task pre-training approach explicitly tailored for media bias detection. MAGPIE outperforms previous approaches in media bias detection, achieving a 3.3% F1-score improvement on the Bias Annotation By Experts (BABE) dataset. Dive into the details and discover how MAGPIE confirms that MTL is a promising approach for addressing media bias detection. Read the full post here: https://bit.ly/43jpEFb #MediaBias #MAGPIE #PreTrainingApproach

One more important benefit of being current on the coming industry changes to fight deepfakes and re-establish trust in photos & videos – whether you’re a camera vendor; photo or video app developer; photo sharing, marketplace or stock provider; photo print producer; or anything in between. *Staying ahead of the curve benefit no. 2: Establishing yourself as an image-trusting solution for your clients and partners.* Demand for verifiable imagery is surging. With deepfake methods constantly evolving, understanding the latest techniques used to authenticate images and to detect deepfakes will empower you to be ahead of the curve in developing features that anticipate market needs. This will not only give you a competitive edge but could also enable you to establish your company as the trusted solution for your (potential) clients and partners. Join us April 11 for our live-streamed Spotlight event and hear from some of the world's leading experts in image content credentials, digital certificates, forensic watermarks and deepfake detection: Jeffrey McGregor, CEO of Truepic; Joseph DeGol, CTO at steg.ai, Michael Matias, CEO of Clarity; and Paul Melcher, Managing Director of Melcher System LLC . More info: https://lnkd.in/gCVdktA6 Bring your questions and comments!

Creating evidence is important. Creating usable evidence is even more so. But is it enough to just create high-quality usable evidence? If the evidence we generate is not used, we minimize our ability to achieve our shared mission of improving the lives of children and families. Learn more about how we can establish systems to support evidence use in the latest blog post from OPRE's Lauren Supplee. https://buff.ly/4aWSj5s

In our last study, we developed a novel machine learning-based strategy to identify relevant patterns associated with predicting blastocyst developmental stage at day 5. Take a look: https://lnkd.in/dNNn-myk #IVF #artificialintelligence #blastocyst

Hey all 🎉 Thrilled to announce that my team NeuralNinjasPSG (with Anrutha Kamalanathan) finished 44th out of 3300+ teams in the Fibe - Hack the Vibe! 2.0 - ML Challenge 2024! This year's challenge focused on classic multi-class text classification of news articles using IAB (Interactive Advertising Bureau) categories. Working with 6.7 Lakh + news articles train set across 26 categories, we developed a lightweight solution (finetuning of DistilBERT with LoRA + MetaClassifier). When I examined the data and with the initial modelling I found for certain categories model was struggling to classifying it. So we tried to bucket the no of categories into k (k << No of Categories) buckets by having a meta classifier then final category classification.

Creating evidence is important. Creating usable evidence is even more so. But is it enough to just create high-quality usable evidence? If the evidence we generate is not used, we minimize our ability to achieve our shared mission of improving the lives of children and families. Learn more about how we can establish systems to support evidence use in the latest blog post from OPRE's Lauren Supplee. https://buff.ly/4aWSj5s

Delighted to share that my team "Cookie Monsters" won the 2nd runner-up title in the "Biomed Datathon 2024" organized by mHealth Lab, BME-BUET yesterday. Participants were given a multi-label classification task, that is to detect 4 heart abnormalities solely from the heart audio recordings of patients. The main challenge was to make the train set (comprising only ~60 samples) usable for effective learning pipelines. Along with basic exploratory data analysis and data augmentations, our approach outlines both classical machine learning techniques (decision tree, random forest, xgboost, multi-layer perceptron) and attention-based modern deep learning techniques (audio spectrogram transformer) to achieve ~60-65% macro F1 score and ICBHI score. Solution Notebooks: https://lnkd.in/gUH5wAWu Presentation Slide: https://lnkd.in/gi4ZgXDh

New tutorial on Ultralytics datasets overview 😍 Ultralytics offers support for a variety of datasets, helping with computer vision tasks like detection, instance segmentation, pose estimation, classification, and multi-object tracking. In this video, Nicolai Nielsen walks us through the diverse datasets highlighted in Ultralytics documentation and showcases how to use them for fine-tuning Ultralytics YOLOv8 for specific tasks. What's Covered 🚀 ✅ Datasets for object detection ✅ Datasets for instance segmentation ✅ Datasets for pose estimation ✅ Datasets for oriented bounding boxes (OBB) ✅ Overview of DOTAv1 dataset Watch Now 👇 https://lnkd.in/ePK-B5T7 #computervision #youtubetutorial #objectdetection #segmentation #yolov8

Watch out for this episode, as our founder @Himanshu Tyagi explains the role of Witness Chain watchtowers in assuring the security of txns on Optimistic Rollups !

Since yesterday, 18 teams of highly motivated hackers have been tackling our case at START Summit x Hack 2024. Their task: "How to make food tracking attractive?" Nicolas Hesse, our PhD student who developed the case, describes in the video why existing food tracking apps are not attractive enough to inspire users in the long term for two reasons. Check out the video to learn more about our case.  We look forward to many innovative solutions!😎 💡 Thanks to Luisa Lager for the great film. 🎥 Are you currently attending the START Summit? Then make sure you stop by our stand at the St.Gallen Health area today or tomorrow and try out our Physio Coach ALEX. 🤖 START Global, Universität St.Gallen (HSG), Yannick Reichen, Nils Tiedemann