Chair of Artificial Intelligence and Machine Learning’s Post

Excited to announce our new blog post on Few-Shot Classification of Interactive Activities of Daily Living (InteractADL) on arXiv. Understanding ADLs, especially those involving multi-person interactions in home environments, is crucial for various applications. Learn about the proposed InteractADL dataset and benchmark, along with the novel method for fine-grained few-shot video classification called Name Tuning. Check out the full post at https://bit.ly/3VqUbPc.

Recently I presented my work on basis functions nonlinear DeePC at the European Control Conference 2024 and I received positive feedback for the presentation, which gives a tutorial and intuitive overview of the technical results in the paper. If you are interested in basis functions formulations of nonlinear DeePC, relation to Koopman MPC and other subtleties have a look at the slides!

Delighted to share that our paper “Video Anomaly Detection via Spatio-Temporal Pseudo-Anomaly Generation : A Unified Approach” is accepted at #CVPR2024 Workshop on Visual Anomaly and Novelty Detection (VAND 2.0). We propose a novel and generic spatio-temporal pseudo anomaly generator for Video Anomaly Detection (VAD) encompassing inpainting of a masked out region in frames using a latent diffusion model and applying mixup augmentation to distort the optical flow. We also introduce a unified VAD framework that measures and aggregates three different indicators of anomalous behaviour namely reconstruction quality, temporal irregularity and semantic inconsistency under a one class classification setting. Paper : https://lnkd.in/eG9CJmx4 Kudos to the team Tarun Krishna Feiyan Hu Alexandru Drimbarean Kevin McGuinness Alan Smeaton Noel O'Connor

Statistical analysis and machine learning algorithms often assume that samples are independent and that most contain genuine information relevant to the issue at hand. In this blog post, I challenge this assumption. Special thanks to my two dear friends Aleksander Molak and Nadav Kedem. For their help and friendship.

Wave 👋 to alert fatigue with modern Anomaly Detection! Andreas Grabner highlights key learnings from his recent "Introduction to Anomaly Detection based on DQL" Observability Lab with David Bründl. Watch the full episode on your YouTube channel here: https://dynatr.ac/4aQYBDy

Here's something you can do with https://tryhelix.ai: you can fine-tune an LLM (Mistral-7B) on new arxiv papers and then chat to your newly minted expert Try it today on a paper that just came out! There's no way the base model will know about it (Thanks to Phil Winder for the demo!)

So here is news on the intersection of my past and present: using the same technology that powers ChatGPT (transformers) to solve a very academic problem in theoretical high energy physics. As a proof of concept this is awesome! Note however that this is only a first step towards using this to predict actual scattering data. #ai #aiimpact #theoreticalphysics #firstlove

🔍 Join me at the upcoming WiDS conference on April 15th, where I'll be hosting a roundtable discussion on "Navigating the Uncharted: Robust Prompt Quality Assurance in the Age of Language Models." 🚀 In today's world dominated by large language models, ensuring prompt quality and robustness is more crucial than ever. During our roundtable, we'll dive deep into the challenges posed by prompt modifications, navigating through model upgrades, handling model deprecation, and selecting the best models for various tasks. I'm passionate about the roundtable format because it allows for genuine exchange among diverse perspectives. Together, we'll enrich each other's understanding and learn valuable coping methods.

I am very pleased to announce our new paper A Meta-analysis of Vulnerability and Trust in Human-Robot Interaction https://lnkd.in/gtGQcmjG. I think it shows how considerate the field is.