BEGIN:VCALENDAR VERSION:2.0 PRODID:-//Meetup//Meetup Calendar 1.0//EN CALSCALE:GREGORIAN METHOD:PUBLISH NAME:Rome AI, Machine Learning and Computer Vision Meetup X-WR-CALNAME:Rome AI, Machine Learning and Computer Vision Meetup BEGIN:VTIMEZONE TZID:Europe/Rome TZURL:http://tzurl.org/zoneinfo-outlook/Europe/Rome X-LIC-LOCATION:Europe/Rome BEGIN:DAYLIGHT TZOFFSETFROM:+0100 TZOFFSETTO:+0200 TZNAME:CEST DTSTART:19700329T020000 RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU END:DAYLIGHT BEGIN:STANDARD TZOFFSETFROM:+0200 TZOFFSETTO:+0100 TZNAME:CET DTSTART:19701025T030000 RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU END:STANDARD END:VTIMEZONE BEGIN:VEVENT UID:event_315682798@meetup.com SEQUENCE:1 DTSTAMP:20260729T190735Z DTSTART;TZID=Europe/Rome:20260804T180000 DTEND;TZID=Europe/Rome:20260804T200000 SUMMARY:Aug 4 - Visual AI in Manufacturing DESCRIPTION:Rome AI\, Machine Learning and Computer Vision Meetup\nJoin ou r virtual meetup to hear talks from experts on cutting-edge topics at the intersection of manufacturing\, AI\, ML\, and computer vision.\n\n**Date\, Time and Location**\n\nAug 04\, 2026\n9:00 AM - 11:00 AM PST\nOnline. **[ Register for the Zoom!](https://voxel51.com/events/visual-ai-in-manufactur ing-meetup-august-4-2026)**\n\n**Enabling Multimodal Agents on the Edge**\ n\nThe next generation of AI agents is moving beyond cloud-based text-only models and will interact with the physical multimodal world in real-time. For example in the vision domain\, AI agents rely on Vision-Language Mode ls (VLMs) in their backbone. However\, deploying massive VLMs with billion s of parameters on the edge devices remains a significant engineering hurd le.\n\nDrawing on our recent ICML and CVPR research papers\, this session explores advancements in agentic model optimizations\, specifically how di stillation and pruning transform 'heavyweight' models into lean\, edge-rea dy engines. Lastly\, I present our UI agent running on the actual phone th at is being developed by our lab's team.\n\n*About the Speaker*\n\n[Denis Gudovskiy](https://www.linkedin.com/in/gudovskiy/) is a Distinguished AI E ngineer at Panasonic North America where he conducts R&D activities of var ious core AI methods\, including multimodal and hardware-efficient agents\ , supervised and RL training pipelines\, and robustness to out-of-distribu tion scenarios.\n\n**When the Camera Can’t Be Trusted: Health-Aware Visu al AI for Reliable Near-Miss Detection**\n\nNear-miss detection systems ar e often evaluated as though every camera frame is equally trustworthy\, ev en though blur\, poor exposure\, occlusion\, contamination\, and changing lighting can silently degrade the visual evidence used to make safety deci sions. This talk presents an online camera-health framework that estimates visual reliability before downstream perception performance significantly deteriorates.\n\nI will discuss how camera-health signals can support con dition-aware evaluation\, prioritize human review\, reduce unreliable aler ts\, and trigger appropriate fallback behavior. Drawing from research in s afety-critical visual perception\, the talk will demonstrate how these pri nciples can be adapted to industrial video systems operating across differ ent cameras\, shifts\, layouts\, and environmental conditions.\n\nThe pres entation will also connect camera-health monitoring with rare-event discov ery and failure-driven dataset improvement for more trustworthy near-miss detection.\n\n*About the Speaker*\n\n[Shiva Aher](https://www.linkedin.com /in/shivaaher/) is a computer vision researcher with a graduate background in computer science from the Georgia Institute of Technology\, specializi ng in artificial intelligence.\n\n**Agentic VLM applications in manufactur ing**\n\nVision Language Models (VLMs) introduce net-new functionality to vision workloads in manufacturing that traditional computer vision models simply do not offer (e.g.\, open-vocabulary detection\, in-context-learnin g). Even so\, fine-tuned models like YOLO offer a level of precision and r ecall that today's VLMs struggle to match out-of-the-box.\n\nThrough agent ic harnesses that coordinate calls to VLMs\, we can start to deliver simil ar reliability on manufacturing-relevant tasks (e.g.\, many-class\, many-i nstance detection)\, while also supporting the net new functionalities (e. g.\, multimodal search) that make VLMs distinct. In this talk\, we walk th rough the design of these harnesses\, how you serve them efficiently\, and how they deliver value in manufacturing.\n\n*About the Speaker*\n\n[Subra iz Ahmed](https://www.linkedin.com/in/subraiz/) is a member of the Technic al Staff at Perceptron AI. He builds the infrastructure to serve frontier vision models. He previously founded a series of startups. URL;VALUE=URI:https://www.meetup.com/rome-ai-machine-learning-and-computer -vision-meetup/events/315682798/ STATUS:CONFIRMED CREATED:20260714T180942Z LAST-MODIFIED:20260714T180942Z CLASS:PUBLIC END:VEVENT BEGIN:VEVENT UID:event_315389154@meetup.com SEQUENCE:1 DTSTAMP:20260729T190735Z DTSTART;TZID=Europe/Rome:20260806T180000 DTEND;TZID=Europe/Rome:20260806T200000 SUMMARY:Aug 6 - Audio and AI Meetup DESCRIPTION:Rome AI\, Machine Learning and Computer Vision Meetup\nJoin ou r virtual meetup to hear talks from experts on cutting-edge topics across AI\, ML\, and computer vision.\n\n**Date\, Time and Location**\n\nAug 06\, 2026\n9:00 AM - 11:00 AM PST\n**[Online. Register for the Zoom!](https:// voxel51.com/events/audio-and-ai-meetup-august-6-2026)**\n\n**Do Speech Mod els Actually Understand Speech? Evaluating Speech LLMs Under Realistic Spo ken Instruction Conditions**\n\nSpeech Large Language Models (SLLMs) are i ncreasingly capable\; but are we evaluating them the right way? Most bench marks rely on text prompts\, yet real users interact with these systems th rough speech\, a modality that introduces noise\, disfluencies\, and styli stic variation that text simply doesn't capture.\nIn this talk\, we presen t findings from a systematic study across 11 tasks\, 12 languages\, and fi ve prompt styles\, examining how prompt modality\, language\, and task typ e shape SLLM performance.\n\n*About the Speaker*\n\n[Maike Züfle ](https: //voxel51.com/events/www.linkedin.com/in/maike-z%C3%BCfle)is a PhD student at the Karlsruhe Institute of Technology (KIT)\, working in Prof. Jan Nie hues's group on interactive speech systems for more natural human–machin e communication. Her research focuses on instruction-following speech mode ls with speech as both input and output\, with a recent emphasis on full-d uplex systems. Beyond her research\, she co-organises the instruction-foll owing and speech translation metrics shared tasks at IWSLT. She is a 2026 Apple Scholar in AI/ML.\n\n**AI based Audio Forensics**\n\nIn this present ation\, attendees will discover several modules developed by Gradiant for the detection and analysis of synthetically generated or manipulated audio . The session will be delivered by one of the developers involved in the d esign and implementation of these technologies\, providing first-hand insi ght into their capabilities and underlying methodology.\n\nThe presentatio n will cover the traceability module\, which helps identify the origin of AI-generated content. It will also cover the segment detection tool\, desi gned to locate manipulated regions within an audio recording\, as well as the complete audio detection tool\, which assesses whether an entire recor ding has been synthetically generated.\n\n*About the Speaker*\n\n[Daniel P aniagua Ares ](https://voxel51.com/events/www.linkedin.com/in/daniel-pania gua-ares-96252918a)is a research engineer at Gradiant. Graduated in comput er engineering from the FIC and with a master's degree in AI from the VIU. \n\n**Curating\, Searching\, and Evaluating Audio Datasets in FiftyOne**\n \nIn this talk\, we'll start with the ESC-50 environmental-sound dataset t o show how FiftyOne represents audio: browsing clips in the tabular view\, rendering spectrograms directly in the sample grid with a custom renderer \, and turning sounds into searchable vectors with CLAP embeddings. Then w e'll demo a similarity-search panel that lets you query an entire audio co llection by example clip or a natural-language prompt to quickly find matc hing sounds.\n\nWe'll conclude with a live research problem: Audio Moment Retrieval from the DCASE 2026 Challenge\, where the goal is to localize th e exact moment in a long recording that matches a text query. We'll frame this as temporal detection\, evaluate predictions\, and visualize ground-t ruth vs. predicted moments on an interactive timeline to intuitively expos e model failure modes.\n\nAttendees will leave with a concrete blueprint a nd open code for applying visual data-centric AI practices to their own au dio and multimodal datasets.\n\n*About the Speaker*\n\n[John Duncan](https ://www.linkedin.com/in/john-a-duncan/) is a Machine Learning Engineer\, Cu stomer Success at Voxel51. His research interests include vision\, LiDAR\, and audio perception for robots and intelligent systems.\n\n**Real-Time A SR at 4x on Consumer Hardware: The Meetily Architecture**\n\nThis talk cov ers the engineering behind Meetily\, an open-source meeting assistant that runs Whisper and NVIDIA Parakeet transcription entirely on-device. We'll walk through how we got Parakeet to roughly 4x real-time on consumer hardw are\, and the specific points where it still falls over.\n\nWe'll also get into the honest trade-offs between local and cloud inference: latency\, a ccuracy\, cost\, and what you actually give up by choosing one over the ot her. Wrapping ML inference in a Rust/Tauri desktop app came with its own c osts\, which we'll unpack as well.\n\nFinally\, we'll look at what "fully local" really means at an architecture level\, where that boundary sits\, and how easily it leaks once you add model downloads\, integrations\, or a pluggable LLM backend.\n\n*About the Speaker*\n\n[Sandeep Zachariah](http s://www.linkedin.com/in/sandeepzachariah/) is the Founder and CEO of Zackr iya Solutions and the leads the team behind Meetily\, an open-source\, pri vacy-first meeting assistant that runs Whisper and NVIDIA Parakeet transcr iption entirely on-device. He brings a rare full-stack perspective on audi o AI — from low-level embedded systems and hardware acceleration up thro ugh real-time ASR and local LLM summarization — with deep experience dep loying speech and ML models across servers\, GPUs and consumer hardware. URL;VALUE=URI:https://www.meetup.com/rome-ai-machine-learning-and-computer -vision-meetup/events/315389154/ STATUS:CONFIRMED CREATED:20260623T223549Z LAST-MODIFIED:20260623T223549Z CLASS:PUBLIC END:VEVENT BEGIN:VEVENT UID:event_315700954@meetup.com SEQUENCE:1 DTSTAMP:20260729T190735Z DTSTART;TZID=Europe/Rome:20260806T180000 DTEND;TZID=Europe/Rome:20260806T200000 SUMMARY:Aug 6 - Audio and AI Meetup DESCRIPTION:Rome AI\, Machine Learning and Computer Vision Meetup\nJoin us on Aug 6 for a special edition of the AI\, ML\, and Computer Vision Meetu p focused on audio use cases!\n\n**Date\, Time and Location**\n\nAug 06\, 2026\n9:00 AM - 11:00 AM PST\nOnline. **[Register for the Zoom](https://vo xel51.com/events/audio-and-ai-meetup-august-6-2026)**\n\n**Do Speech Model s Actually Understand Speech? Evaluating Speech LLMs Under Realistic Spoke n Instruction Conditions**\n\nSpeech Large Language Models (SLLMs) are inc reasingly capable\; but are we evaluating them the right way? Most benchma rks rely on text prompts\, yet real users interact with these systems thro ugh speech\, a modality that introduces noise\, disfluencies\, and stylist ic variation that text simply doesn't capture.\nIn this talk\, we present findings from a systematic study across 11 tasks\, 12 languages\, and five prompt styles\, examining how prompt modality\, language\, and task type shape SLLM performance.\n\n*About the Speaker*\n\n[Maike Züfle](https://v oxel51.com/events/www.linkedin.com/in/maike-z%C3%BCfle) is a PhD student a t the Karlsruhe Institute of Technology (KIT)\, working in Prof. Jan Niehu es's group on interactive speech systems for more natural human–machine communication.\n\n**AI based Audio Forensics**\n\nIn this presentation\, a ttendees will discover several modules developed by Gradiant for the detec tion and analysis of synthetically generated or manipulated audio. The ses sion will be delivered by one of the developers involved in the design and implementation of these technologies\, providing first-hand insight into their capabilities and underlying methodology.\n\nThe presentation will co ver the traceability module\, which helps identify the origin of AI-genera ted content. It will also cover the segment detection tool\, designed to l ocate manipulated regions within an audio recording\, as well as the compl ete audio detection tool\, which assesses whether an entire recording has been synthetically generated.\n\n*About the Speaker*\n\n[Daniel Paniagua A res](https://voxel51.com/events/www.linkedin.com/in/daniel-paniagua-ares-9 6252918a) is a research engineer at Gradiant. Graduated in computer engine ering from the FIC and with a master's degree in AI from the VIU.\n\n**Cur ating\, Searching\, and Evaluating Audio Datasets in FiftyOne**\n\nIn this talk\, we'll start with the ESC-50 environmental-sound dataset to show ho w FiftyOne represents audio: browsing clips in the tabular view\, renderin g spectrograms directly in the sample grid with a custom renderer\, and tu rning sounds into searchable vectors with CLAP embeddings. Then we'll demo a similarity-search panel that lets you query an entire audio collection by example clip or a natural-language prompt to quickly find matching soun ds.\n\nWe'll conclude with a live research problem: Audio Moment Retrieval from the DCASE 2026 Challenge\, where the goal is to localize the exact m oment in a long recording that matches a text query. We'll frame this as t emporal detection\, evaluate predictions\, and visualize ground-truth vs. predicted moments on an interactive timeline to intuitively expose model f ailure modes.\n\nAttendees will leave with a concrete blueprint and open c ode for applying visual data-centric AI practices to their own audio and m ultimodal datasets.\n\n*About the Speaker*\n\n[John Duncan](https://www.li nkedin.com/in/john-a-duncan/) is a Machine Learning Engineer\, Customer Su ccess at Voxel51. His research interests include vision\, LiDAR\, and audi o perception for robots and intelligent systems. URL;VALUE=URI:https://www.meetup.com/rome-ai-machine-learning-and-computer -vision-meetup/events/315700954/ STATUS:CONFIRMED CREATED:20260715T231958Z LAST-MODIFIED:20260715T231958Z CLASS:PUBLIC END:VEVENT BEGIN:VEVENT UID:event_315683203@meetup.com SEQUENCE:1 DTSTAMP:20260729T190735Z DTSTART;TZID=Europe/Rome:20260811T180000 DTEND;TZID=Europe/Rome:20260811T200000 SUMMARY:Aug 11 - Debugging Physical AI Models at Scale with Multimodal Dat a Workshop DESCRIPTION:Rome AI\, Machine Learning and Computer Vision Meetup\nJoin Vo xel51 for a live workshop on how multimodal data workflows in FiftyOne hel p teams inspect\, search\, and debug complex Physical AI datasets and expl ain black-box model behavior at scale. We’ll show how teams can work wit h synchronized video and sensor data\, query for similar scenarios across their datasets\, and uncover patterns behind model failures faster than pl ayback-only visualization tools allow.\n\n**Date\, Time and Location**\n\n Aug 11\, 2026\n9:00 AM - 10:00 AM PST\nOnline. **[Register for the Zoom!]( https://voxel51.com/events/debugging-physical-ai-models-at-scale-with-mult imodal-data-august-11-2026)**\n\nAs robotics and autonomous vehicle teams move from traditional perception models to end-to-end Physical AI systems\ , understanding model behavior is becoming harder than ever. These models ingest synchronized inputs from cameras\, sensors\, and other data streams \, but their decisions can be difficult to explain\, reproduce\, and impro ve.\n\nYou’ll learn how to use multimodal data to investigate questions like: when did the model swerve\, miss an object\, misinterpret a scene\, or behave unexpectedly — and how can you find every similar moment acros s your dataset?\n\nDesigned for robotics\, AV\, and machine learning teams \, this session will show how FiftyOne helps turn multimodal data into a s calable workflow for model evaluation\, debugging\, and improvement. URL;VALUE=URI:https://www.meetup.com/rome-ai-machine-learning-and-computer -vision-meetup/events/315683203/ STATUS:CONFIRMED CREATED:20260714T183337Z LAST-MODIFIED:20260714T183337Z CLASS:PUBLIC END:VEVENT BEGIN:VEVENT UID:event_315684124@meetup.com SEQUENCE:1 DTSTAMP:20260729T190735Z DTSTART;TZID=Europe/Rome:20260813T180000 DTEND;TZID=Europe/Rome:20260813T190000 SUMMARY:Aug 13 - How to Build Vision Data Agents with Tools\, Skills\, and MCP DESCRIPTION:Rome AI\, Machine Learning and Computer Vision Meetup\nIn this session\, you’ll learn how to build production-ready AI agents that can reason over your data\, automate complex tasks\, and integrate seamlessly into your existing stack using tools\, skills\, and the Model Context Pro tocol (MCP).\n\n**Date\, Time and Location**\n\nAug 13\, 2026\n9:00 AM - 1 0:00 AM PST\nOnline. **[Register for the Zoom!](https://voxel51.com/events /how-to-build-vision-data-agents-with-tools-skills-and-mcp-august-13-2026) **\n\nWe’ll walk through how modern agentic systems move beyond simple p rompts—leveraging structured tools like dataset operations\, embeddings\ , evaluation pipelines\, and model execution to take real action. You’ll see how these agents can tag data\, run inference\, evaluate performance\ , and surface insights automatically\, all within a unified workflow.\n\nB y combining natural language interfaces with programmable building blocks\ , teams can dramatically reduce manual effort\, accelerate experimentation \, and unlock faster decision-making across the ML lifecycle.\n\nWhether y ou're building data-centric AI systems\, managing large-scale vision datas ets\, or exploring agentic workflows for the first time\, this session wil l give you a practical blueprint for getting started.\n\n*About the Speake r*\n\n[Adonai Vera](https://www.linkedin.com/in/adonai-vera/) \\- Machine Learning Engineer & DevRel at Voxel51\\. With over 7 years of experience b uilding computer vision and machine learning models using TensorFlow\\\, D ocker\\\, and OpenCV\\. I started as a software developer\\\, moved into A I\\\, led teams\\\, and served as CTO\\. Today\\\, I connect code and comm unity to build open\\\, production\\-ready AI\\\, making technology simple \\\, accessible\\\, and reliable\\. URL;VALUE=URI:https://www.meetup.com/rome-ai-machine-learning-and-computer -vision-meetup/events/315684124/ STATUS:CONFIRMED CREATED:20260714T192455Z LAST-MODIFIED:20260714T192455Z CLASS:PUBLIC END:VEVENT BEGIN:VEVENT UID:event_314739436@meetup.com SEQUENCE:1 DTSTAMP:20260729T190735Z DTSTART;TZID=Europe/Rome:20260820T180000 DTEND;TZID=Europe/Rome:20260820T200000 SUMMARY:Aug 20 - Cold Pool to Hot Queue: Annotation Curation with FiftyOne DESCRIPTION:Rome AI\, Machine Learning and Computer Vision Meetup\nIn this hands-on workshop\, you'll use [FiftyOne ](https://docs.voxel51.com/)to r un the full rare-class mining loop end-to-end on a large unlabeled image p ool: compress the pool with near-duplicate detection\, embed images with a modern vision backbone\, mine candidate positives via seeded similarity f rom a tiny labeled set\, confirm them through targeted human review\, and prioritize the survivors for annotation using representativeness and uniqu eness scores.\n\n**Time\, Date and Location**\n\nAug 20\, 2026\n9:00 AM - 11:00 AM PST\, 2026\nOnline. **[Register for the Zoom!](https://voxel51.co m/events/from-cold-pool-to-hot-queue-annotation-curation-with-fiftyone-aug ust-20-2026)**\n\n**What You'll Walk Away With**\n\n* A working FiftyOne p ipeline for finding rare classes in any visual dataset you own\n* A repeat able four-stage funnel — compress\, mine\, confirm\, prioritize — with a clear objective at each stage\n* A fine-tuned detector that demonstrabl y outperforms one trained on the same number of randomly sampled images\n* The mental model that data curation — not architecture or hyperparamete rs — is the highest-leverage thing you can do to improve a rare-class de tector URL;VALUE=URI:https://www.meetup.com/rome-ai-machine-learning-and-computer -vision-meetup/events/314739436/ STATUS:CONFIRMED CREATED:20260511T191209Z LAST-MODIFIED:20260511T191209Z CLASS:PUBLIC END:VEVENT BEGIN:VEVENT UID:event_315684972@meetup.com SEQUENCE:1 DTSTAMP:20260729T190735Z DTSTART;TZID=Europe/Rome:20260825T180000 DTEND;TZID=Europe/Rome:20260825T200000 SUMMARY:Aug 25 - Advances in AI at NYU DESCRIPTION:Rome AI\, Machine Learning and Computer Vision Meetup\nJoin ou r virtual meetup to hear talks from researchers at NYU on cutting-edge top ics across AI\, ML\, and computer vision.\n\n**Date\, Time and Location**\ n\nAug 25\, 2026\n9:00 AM - 11:00 AM PST\nOnline. **[Register for the Zoom !](https://voxel51.com/events/advances-in-ai-at-nyu-august-25-2026)**\n\n* *Using Computer Vision to Advance the Sciences**\n\nI'll present some of o ur ongoing work on using computer vision to create impact in the sciences. These target a two areas\, solar physics and evolutionary biology\, that deal with objects of radically different sizes but are unified by a need f or high quality\, trustworthy data.\n\nI'll show off our efforts\, done in collaboration with domain experts\, that aim to produce the best possible maps of the Sun's powerful magnetic field and have created some of the wo rld's largest repositories of data about bird morphology.\n\n*About the Sp eaker*\n\n[David Fouhey](https://www.linkedin.com/in/david-fouhey-64212037 9/) is an Associate Professor at New York University and a research scient ist at Polymathic AI. Before joining NYU\, he received a PhD in robotics f rom Carnegie Mellon\, was a postdoc at UC Berkeley\, and was a professor a t University of Michigan.\n\n**Solaris: Building a Multiplayer Video World Model in Minecraft**\n\nThis talk will introduce Solaris: a multiplayer v ideo world model in Minecraft. I will first present SolarisEngine\, the so ftware platform we built to simulate realistic multiplayer gameplay betwee n bots at scale\, enabling us to collect a large training dataset of align ed multiplayer actions and frames.\n\nI will then discuss our staged train ing pipeline\, starting with single-player pre-training before converting the model into a long-horizon multiplayer generator through bidirectional training\, followed by causal training\, and concluding with Self Forcing. I will also cover our memory-efficient implementation of Self Forcing\, c alled Checkpointed Self Forcing.\n\nFinally\, I will showcase generated vi deos illustrating how Solaris maintains coherent long-horizon multiplayer interactions.\n\n*About the Speaker*\n\n[Oscar Michel ](https://www.linked in.com/in/oscar-michel-82a162145/)is a PhD student at NYU advised by Prof. Saining Xie. His research studies world models: generative models of agen ts interacting in an environment.\n\n**Closing the human to robot gap for dexterous hands**\n\nCollecting task-specific robot data for multi-fingere d hands is challenging due to the many difficulties that arise in teleoper ation. That is why recently there has been a major focus on learning robot policies directly from human demonstrations. However\, human demonstratio ns are difficult to work with\; there is a major morphological and visual gap between human and robot hands\, as well as between the environments th ey operate in.\n\nIn this talk\, I'd like to discuss my efforts on closing this gap.\n\n*About the Speaker*\n\n[Irmak Guzey](https://www.linkedin.co m/in/irmak-guzey-6a9010175) I'm Irmak (she/her)\, a rising 3rd year PhD st udent at New York University\, currently advised by Lerrel Pinto. My resea rch focuses on robot learning for dexterous manipulation. I have been awar ded a Fulbright scholarship and NYU's Best Master's Thesis Award in the pa st. URL;VALUE=URI:https://www.meetup.com/rome-ai-machine-learning-and-computer -vision-meetup/events/315684972/ STATUS:CONFIRMED CREATED:20260714T204416Z LAST-MODIFIED:20260714T204416Z CLASS:PUBLIC END:VEVENT BEGIN:VEVENT UID:event_314999424@meetup.com SEQUENCE:1 DTSTAMP:20260729T190735Z DTSTART;TZID=Europe/Rome:20260827T180000 DTEND;TZID=Europe/Rome:20260827T200000 SUMMARY:Aug 27 - AI\, ML\, and Computer Vision Meetup DESCRIPTION:Rome AI\, Machine Learning and Computer Vision Meetup\nJoin ou r virtual meetup to hear talks from experts on cutting-edge topics across AI\, ML\, and computer vision.\n\nDate\, Time\, and Location\n\nAug 27\, 2 026\n9:00 AM - 11:00 AM PST\nOnline. **[Register for the Zoom!](https://vo xel51.com/events/ai-ml-and-computer-vision-meetup-august-27-2026)**\n\n**R obust Concept Protection against Diffusion-Based Image Editing and Persona lization**\n\nDiffusion-based image editing and personalization models hav e made it increasingly easy to manipulate and replicate visual concepts fr om only a few reference images. However\, existing protection methods ofte n overfit to a single attack model and fail to generalize across diverse e diting pipelines.\n\nIn this presentation\, I will discuss recent advances in concept protection for generative AI systems\, focusing on targeted pe rturbation strategies and style-sensitive diffusion representations. I wil l also present experimental findings across multiple editing and fine-tuni ng scenarios\, highlighting the challenges of robustness\, transferability \, and imperceptibility in practical protection settings. Finally\, I will discuss open problems and future directions toward trustworthy generative content ownership.\n\n*About the Speaker*\n\n[Qiuyu Tang](https://www.lin kedin.com/in/qiuyutang/) is a Ph.D. student in Computer Science and Engine ering at Lehigh University. Her research focuses on trustworthy AI\, media forensics\, and robust protection methods against diffusion-based image e diting and personalization systems. Her recent work explores concept prote ction\, style safeguarding\, semantic image manipulation\, and generative AI robustness. She has contributed to multiple publications in computer vi sion and AI safety\, including research on diffusion model protection and manipulation detection\, and has also served as a conference workshop orga nizer.\n\n**From Pixels to the Planet: Building Scalable and Grounded AI f or Science**\n\nAI has demonstrated a lot of new possibilities\, from draf ting emails to image editing and generation. The efficacy of AI models is largely built upon a standard machine learning pipeline\, where data is fe d into models to get representations and predictions\, and the performance is evaluated with controlled benchmarks and metrics. However\, the mismat ch arises when we try to transit this pipeline to the interaction with the real world and use AI for scientific discovery. Beyond close-set decision s\, scientists want to discover new categories and propose new hypotheses. In this talk\, I will share how I address the challenges of AI for scienc e from the perspectives of data-centric methods and interpretability appro aches.\n\n*About the Speaker*\n\n[Jianyang Gu](https://www.linkedin.com/in /jianyang-gu-9235151b9/) is a postdoctoral scholar at The Ohio State Unive rsity. His research focuses on using data-centric methods to build scalabl e and interpretable foundation models for science.\n\n**Beyond the Barn: N on-Invasive Acidosis Detection in Dairy Cattle Through Multimodal Gas Emis sion Intelligence**\n\nRumen acidosis silently costs the global dairy indu stry billions annually and compromises animal welfare\, yet current detect ion methods remain invasive\, delayed\, and impractical at scale. Our lab has pioneered a fundamentally new approach: capturing and analyzing exhale d CO₂ and CH₄ gas emission patterns through synchronized RGB-thermal i maging\, turning every breath into a diagnostic signal. We developed DualG asNet\, a dual-stream deep learning architecture with cross-attention fusi on that detects acidosis non-invasively and in real time\, achieving state -of-the-art accuracy on a first-of-its-kind livestock gas emission dataset we constructed from scratch.\n\nTo push toward explainable\, farm-ready A I\, we integrate vision-language models — CLIP and LLaVA-1.5 — enablin g zero-shot diagnostic reasoning that bridges the gap between deep learnin g predictions and actionable veterinary insight. This talk will walk throu gh the full pipeline from custom dataset creation to multimodal fusion to VLM-powered interpretation\, offering the audience a compelling case study in how computer vision can solve high-impact\, real-world problems outsid e traditional benchmarks.\n\n*About the Speaker*\n\n[Taminul Islam](https: //www.linkedin.com/in/taminul-islam/) is a Doctoral Research Fellow and Ph D candidate at Southern Illinois University Carbondale with 40+ publicatio ns\, 740+ citations\, and an h-index of 16 — with publications in CVPR 2 026\, WACV 2026 (Oral)\, ICCV 2025\, and Nature Scientific Reports\, inclu ding a Highly Cited Paper for 2024–25.\n\n**Building Real-World Computer Vision Systems with Voxel51**\n\nThis talk will explore practical workflo ws for building\, evaluating\, and improving modern computer vision system s. We’ll dive into real-world approaches to dataset curation\, model ana lysis\, multimodal AI workflows\, and production-ready vision pipelines us ing open-source technologies.\n\nThe session is designed for engineers\, r esearchers\, and AI practitioners looking to better understand how teams a re developing and scaling computer vision applications today. Expect pract ical demos\, technical insights\, and discussions around the evolving AI t ooling ecosystem.\n\n*About the Speaker*\n\n[Daniel Gural](https://www.lin kedin.com/in/daniel-gural/) is an expert in Physical AI and has been worki ng in the field for over 8 years. Working across healthcare he has experie nce in both operating use case as well as using Visual AI as an aid in psy chology applications as well. URL;VALUE=URI:https://www.meetup.com/rome-ai-machine-learning-and-computer -vision-meetup/events/314999424/ STATUS:CONFIRMED CREATED:20260528T185051Z LAST-MODIFIED:20260528T185051Z CLASS:PUBLIC END:VEVENT BEGIN:VEVENT UID:event_314739577@meetup.com SEQUENCE:1 DTSTAMP:20260729T190735Z DTSTART;TZID=Europe/Rome:20260902T180000 DTEND;TZID=Europe/Rome:20260902T200000 SUMMARY:Sept 2 - Document Visual AI Workshop DESCRIPTION:Rome AI\, Machine Learning and Computer Vision Meetup\nIn this hands-on workshop\, you'll use [FiftyOne](https://docs.voxel51.com/) and the High Quality Invoice Images for OCR dataset to run the full data-centr ic loop end-to-end: embed invoices with a modern visual document model\, c luster them by structure\, run LightOnOCR as your base model\, and use per -sample evaluation scores layered onto embedding space to find \\*where\\* and \\*why\\* it fails.\n\n**Time\, Date and Location**\n\nSep 02\, 2026\ n9:00 AM - 11:00 AM PST\nOnline. **[Register for the Zoom!](https://voxel5 1.com/events/document-visual-ai-workshop-september-2-2026)**\n\n**What You 'll Walk Away With**\n\n* A working FiftyOne pipeline for any document col lection you own\n* A repeatable curation query that combines evaluation + embedding signals\n* A fine-tuned LightOnOCR checkpoint that demonstrably outperforms the base model on your invoices\n* The mental model that data curation — not architecture or hyperparameters — is the highest-levera ge thing you can do to improve a document AI system URL;VALUE=URI:https://www.meetup.com/rome-ai-machine-learning-and-computer -vision-meetup/events/314739577/ STATUS:CONFIRMED CREATED:20260511T191801Z LAST-MODIFIED:20260511T191801Z CLASS:PUBLIC END:VEVENT BEGIN:VEVENT UID:event_315388191@meetup.com SEQUENCE:1 DTSTAMP:20260729T190735Z DTSTART;TZID=Europe/Rome:20260924T180000 DTEND;TZID=Europe/Rome:20260924T200000 SUMMARY:Sept 24 - AI\, ML and Computer Vision Meetup DESCRIPTION:Rome AI\, Machine Learning and Computer Vision Meetup\nJoin ou r virtual meetup on September 24 to hear talks from experts on cutting-edg e topics across AI\, ML\, and computer vision.\n\n**Date\, Time and Locati on**\n\nSep 24\, 2026\n9:00 AM - 11:00 AM PST\nOnline. **[Register for the Zoom!](https://voxel51.com/events/ai-ml-and-computer-vision-meetup-septem ber-24-2026)**\n\n**How Do Mercedes-Benz AI Principles Drive our Innovatio n?**\n\nAt Mercedes-Benz\, our AI Principles guide every step of innovatio n\, emphasizing responsible use\, safety and reliability\, explainability\ , and the protection of privacy. These principles go beyond statements and actively shape how we design\, test\, and deploy AI systems in real-world automotive and enterprise settings. In this talk\, I will present how the se principles inspired our recent research on when reusing LoRA (Low-Rank Adaptation) is effective. By combining theoretical analysis with synthetic data as a proxy for enterprise scenarios\, we uncovered the strengths and limitations of modular AI components under constrained data access. Our f indings provide practical guidance on when reused LoRAs could deliver high -quality results.\n\n*About the Speaker*\n\n[Mei-Yen Chen](https://www.lin kedin.com/in/mei-yen-chen-22937787) is a Senior Data Scientist at Mercedes -Benz Tech Innovation GmbH in Germany with 10 years of industry experience in AI and data solutions. She leads early-stage AI projects across busine ss functions and collaborates with research institutions on machine learni ng and responsible AI.\n\n**Region Tokens as the Visual Primitive: From Re cognition to World Modeling**\n\nPatch-based tokenization has become the d efault interface between vision encoders and downstream models\, yet patch es carry no semantic structure and scale poorly with resolution and tempor al extent. This talk presents a research program centered on replacing pat ch tokens with region-level representations — semantically dense tokens grounded in visual entities rather than arbitrary grid crops.\n\nI will de scribe RELOCATE\, REN\, and T-REN\, a progression of methods that produce region tokens via pooling\, train them with region-level objectives\, and extend them to video with temporal coherence. I will then present ongoing work integrating region tokens into VLMs to directly expand visual context capacity\, and preliminary results on future region trajectory prediction as a foundation for world modeling.\n\nThe broader thesis is that region- level tokens are a more natural unit of visual computation than patches\, and their advantage compounds as task complexity\, resolution\, and tempor al horizon increase.\n\n*About the Speaker*\n\n[Savya Khosla](https://www. linkedin.com/in/savyakhosla/) is a second-year Ph.D. student at the Univer sity of Illinois Urbana-Champaign\, advised by Prof. Derek Hoiem and Prof. Alex Schwing.\n\n**Leveraging Text-To-Image Diffusion Models for Consiste nt Set-to-Set Generation**\n\nImage collections are humans' primary way of capturing the world\, yet advances in generative editing remain largely i napplicable to this modality. We address this gap by introducing Match-and -Fuse - a zero-shot\, training-free method for consistent set-to-set gener ation from image collections that share a common visual element but differ in viewpoint\, capture time\, and surrounding content.\nOur key idea is a unified graph-based framework that combines dense correspondences with an emergent prior in text-to-image diffusion models to generate coherent can vases. We achieve state-of-the-art consistency and visual quality\, and un lock new creative capabilities for content generation.\n\n*About the Speak er*\n\n[Kate Feingold](https://www.linkedin.com/in/katefeingold/) is a PhD student in Computer Vision at the Weizmann Institute of Science. Her rese arch sits at the intersection of generative models\, 3D/4D perception\, an d multimodal learning\, focusing on problems where vision meets other moda lities or paradigms in creative tasks.\n\n**Yield Estimation of a Coffee i n a dense environment**\n\nThis presentation provides a detailed workflow related to coffee yield estimation in a dense environment. With photos of pre-harvest coffee plants from a couple of coffee estates\, details relate d to pre-processing\, annotation to detect regions of interest (ROI)\, obj ect detection training and inferencing results with various Yolo models an d finally segmentation with SAM2 and Yolo\\*-seg with training and inferen ce results to determine the count of raw\, pre-mature\, mature and over-ma ture coffee berries and finally the yield of the entire estate. All this i s based on real world data captured on iPhone and android phones.\n\n*Abou t the Speaker*\n\n[Raghu M. Rao](http://www.linkedin.com/in/raghumrao) is a consultant working on applications of computer vision AI models. He was previously with AMD and Xilinx. He has a Ph.D. in Wireless Communications from UCLA and is a Senior Member\, IEEE. His current interests are in appl ications of AI for agriculture\, health care and wireless communications. URL;VALUE=URI:https://www.meetup.com/rome-ai-machine-learning-and-computer -vision-meetup/events/315388191/ STATUS:CONFIRMED CREATED:20260623T210742Z LAST-MODIFIED:20260623T210742Z CLASS:PUBLIC END:VEVENT X-ORIGINAL-URL:https://www.meetup.com/rome-ai-machine-learning-and-compute r-vision-meetup/events/ical/ X-WR-CALNAME:Rome AI\, Machine Learning and Computer Vision Meetup END:VCALENDAR