Paris AI, Machine Learning and Computer Vision Meetup

Register for the Zoom

Part 3: Teaching Machines to See and Click - Model Finetuning

From Foundation Models to GUI Specialists

Foundation models, such as Qwen2.5-VL, demonstrate impressive visual understanding, but they require specialized training to master GUI interactions. In this final session, you'll transform a general-purpose vision-language model into a GUI specialist that can navigate interfaces with human-like precision.

We'll explore modern fine-tuning strategies specifically designed for GUI tasks, from selecting the right architecture to handling the unique challenges of coordinate prediction and multi-step reasoning. You'll implement training pipelines that can handle the diverse formats and platforms in your dataset, evaluate models on metrics that actually matter for GUI automation, and deploy your trained model in a real-world testing environment.

About the Instructor

Harpreet Sahota is a hacker-in-residence and machine learning engineer with a passion for deep learning and generative AI. He’s got a deep interest in RAG, Agents, and Multimodal AI.

Virtual. Register for the Zoom!

Detecting the Unexpected: Practical Approaches to Anomaly Detection in Visual Data

Anomaly detection is one of computer vision's most exciting and essential challenges today. From spotting subtle defects in manufacturing to identifying edge cases in model behavior, it is one of computer vision's most exciting and crucial challenges. In this session, we’ll do a hands-on walkthrough using the MVTec AD dataset, showcasing real-world workflows for data curation, exploration, and model evaluation. We’ll also explore the power of embedding visualizations and similarity searches to uncover hidden patterns and surface anomalies that often go unnoticed.

This session is packed with actionable strategies to help you make sense of your data and build more robust, reliable models. Join us as we connect the dots between data, models, and real-world deployment—alongside other experts driving innovation in anomaly detection.

About the Speaker

Paula Ramos has a PhD in Computer Vision and Machine Learning, with more than 20 years of experience in the technological field. She has been developing novel integrated engineering technologies, mainly in Computer Vision, robotics, and Machine Learning applied to agriculture, since the early 2000s in Colombia.

Scaling Synthetic Data for Industrial AI: From CAD to Model in Hours

This talk explores how we generate high-performance computer vision datasets from CAD—without real-world images or manual labeling. We’ll walk through our synthetic data pipeline, including CPU-optimized defect simulation, material variation, and lighting workflows that scale to thousands of renders per part. While Blender plays a role, our focus is on how industrial data (like STEP files) and procedural generation unlock fast, flexible training sets for manufacturing QA, even on modest hardware. If you're working at the edge of 3D, automation, and vision AI—this is for you!

About the Speaker

Matt Puchalski is the founder and CEO of Bucket Robotics, A Y Combinator backed startup building self-serve computer vision systems for manufacturing. Previously, he led robotics reliability at Argo AI and helped build and deploy autonomous vehicles at Stack AV and Uber ATG.

Swarm Intelligence: Solving Complex Industrial Optimization in Seconds

Manufacturing and logistics companies face increasingly complex operational challenges that traditional AI and human planning struggle to solve effectively. Collide Technology harnesses Swarm Intelligence algorithms to transform intractable problems—like scheduling hundreds or thousands of maintenance employees while simultaneously optimizing production capacity, inventory levels, and cross-sector resource allocation—into solutions delivered in seconds rather than weeks.

Unlike rigid Operations Research approaches that require specialized expertise and expensive implementations, our platform democratizes industrial optimization by making sophisticated decision-making accessible to any factory or logistics operation. We deliver holistic, data-driven solutions that optimize across multiple business entities and sectors simultaneously, adapting to real-world constraints and evolving operational needs.

About the Speaker

Frederick Gertz, PhD has worked in AI for the manufacturing space for over a decade delivering data science insights for the medical and pharmaceutical manufacturing space. Prior to that he worked in nanotechnology with a focus on bio-physics and nanomagnetics with his dissertation research on Magnonic Holographic Devices being named as a runner-up for 2014 Physics Breakthrough of the Year by Physics World.

Virtual. Register for the Zoom!

Bringing Specialist Agents to the Physical World to Improve Manufacturing Output

U.S. manufacturing productivity (output per labor hour) has been stagnant since 2008, driven by a stall in technology integration as well as available workers. RIOS Agents are collaborative AI perception and control systems that act as plant managers' eyes on the ground. Our Agents become specialists in a process, observing process steps, reporting on them, and ultimately controlling them by integrating into new or existing equipment. This enables factory production to be optimized in a way that was previously not possible.

About the Speaker

Clinton Smith is the co-founder and CEO of RIOS, whose AI agents watch, optimize and control production in various industrial facilities, including deep penetration into wood products and lumber. Clinton previously was a Senior Member of the Research Staff at Xerox PARC, leading multiple Dept. of Energy & Dept. of Defense projects, and holds a PhD in Electrical Engineering from Princeton University and a BS in Computer Engineering from Georgia Tech.

Accelerating Robotics with Simulation

In this session, Steve Xie, CEO of Lightwheel, shares how simulation-first workflows and high-quality SimReady assets are transforming the development of visual AI in manufacturing. From warehouse anomaly detection to worker safety and object identification, Steve will explore how physics-accurate simulation and synthetic datasets can drive scalable AI training with minimal real-world data. Drawing from Lightwheel’s deployment of robot models like GR00T N1 in factory environments, the talk highlights how unifying vision, language, and action in simulation accelerates real-world deployment while improving safety, generalization, and efficiency.

About the Speaker

Dr. Steve Xie is founder and CEO of Lightwheel, a company leading simulation infrastructure for embodied AI. Steve is a pioneer in generative-AI-powered simulation for robotics. He holds a B.S. from Peking University and a Ph.D. from Columbia University. Steve has led simulation efforts at NVIDIA and Cruise, where he built end-to-end synthetic data pipelines that set industry benchmarks for realism, scalability, and sim2real transfer.

Anomalib 2.0: Edge Inference and Model Deployment

When deploying models for inference, just exporting the models and calling them via the inferencers do not work. There are challenges related to pre-processing and post-processing. Any deviation in these steps during inference impacts performance. This talk is about how we re-designed components of Anomalib to integrate pre and post-processing steps in the model graph.

About the Speaker

Samet Akcay is an AI Research Engineer at Intel who leads ML research and development efforts across multiple Open Edge Platform libraries, including Intel Geti, Datumaro, Anomalib, Training Extensions, and Vision Inference libraries. His research specializes in semi/self-supervised learning, zero/few-shot learning, and multi-modal object and anomaly detection. He is the creator of Anomalib, a major open-source anomaly detection library.

Exploring Robotic Manipulation Datasets using FiftyOne: DROID and Amazon Armbench

About the Speaker

Allen Lee is currently a Machine Learning Engineer at Voxel51. Before that, Allen was the Co-Founder and Consulting Engineer at Leap Scientific LLC, where they provided scientific software consultancy services related to computation, machine learning, and computer vision.

Virtual. Register for the Zoom!

Towards Robotics Foundation Models that Can Reason

In recent years, we have witnessed remarkable progress in generative AI, particularly in language and visual understanding and generation. This leap has been fueled by unprecedentedly large image–text datasets and the scaling of large language and vision models trained on them. Increasingly, these advances are being leveraged to equip and empower robots with open-world visual understanding and reasoning capabilities.

Yet, despite these advances, scaling such models for robotics remains challenging due to the scarcity of large-scale, high-quality robot interaction data, limiting their ability to generalize and truly reason about actions in the real world. Nonetheless, promising results are emerging from using multimodal large language models (MLLMs) as the backbone of robotic systems, especially in enabling the acquisition of low-level skills required for robust deployment in everyday household settings.

In this talk, I will present three recent works that aim to bridge the gap between rich semantic world knowledge in MLLMs and actionable robot control. I will begin with AHA, a vision-language model that reasons about failures in robotic manipulation and improves the robustness of existing systems. Building on this, I will introduce SAM2Act, a 3D generalist robotic model with a memory-centric architecture capable of performing high-precision manipulation tasks while retaining and reasoning over past observations. Finally, I will present MolmoAct, AI2’s flagship robotic foundation model for action reasoning, designed as a generalist system that can be post-trained for a wide range of downstream manipulation tasks.

About the Speaker

Jiafei Duan is a Ph.D. candidate in Computer Science & Engineering at the University of Washington, advised by Professors Dieter Fox and Ranjay Krishna. His research focuses on foundation models for robotics, with an emphasis on developing scalable data collection and generation methods, grounding vision-language models in robotic reasoning, and advancing robust generalization in robot learning. His work has been featured in MIT Technology Review, GreekWire, VentureBeat, and Business Wire.

Beyond Academic Benchmarks: Critical Analysis and Best Practices for Visual Industrial Anomaly Detection

In this talk, I will share our recent research efforts in visual industrial anomaly detection. It will present a comprehensive empirical analysis with a focus on real-world applications, demonstrating that recent SOTA methods perform worse than methods from 2021 when evaluated on a variety of datasets. We will also investigate how different practical aspects, such as input size, distribution shift, data contamination, and having a validation set, affect the results.

About the Speaker

Aimira Baitieva is a Research Engineer at Valeo, where she works primarily on computer vision problems. Her recent work has been focused on deep learning anomaly detection for automating visual inspection, incorporating both research and practical applications in the manufacturing sector.

The Digital Reasoning Thread in Manufacturing: Orchestrating Vision, Simulation, and Robotics

Manufacturing is entering a new phase where AI is no longer confined to isolated tasks like defect detection or predictive maintenance. Advances in reasoning AI, simulation, and robotics are converging to create end-to-end systems that can perceive, decide, and act – in both digital and physical environments.

This talk introduces the Digital Reasoning Thread – a consistent layer of AI reasoning that runs through every stage of manufacturing, connecting visual intelligence, digital twins, simulation environments, and robotic execution. By linking perception with advanced reasoning and action, this approach enables faster, higher-quality decisions across the entire value chain.

We will explore real-world examples of applying reasoning AI in industrial settings, combining simulation-driven analysis, orchestration frameworks, and the foundations needed for robotic execution in the physical world. Along the way, we will examine the key technical building blocks – from data pipelines and interoperability standards to agentic AI architectures – that make this level of integration possible.

Attendees will gain a clear understanding of how to bridge AI-driven perception with simulation and robotics, and what it takes to move from isolated pilots to orchestrated, autonomous manufacturing systems.

About the Speaker

Vlad Larichev is an Industrial AI Lead at Accenture Industry X, specializing in applying AI, generative AI, and agentic AI to engineering, manufacturing, and large-scale industrial operations. With a background as an engineer, solution architect, and software developer, he has led AI initiatives across sectors including automotive, energy, and consumer goods, integrating advanced analytics, computer vision, and simulation into complex industrial environments.

Vlad is the creator of the Digital Reasoning Thread – a framework for connecting AI reasoning across visual intelligence, simulation, and physical execution. He is an active public speaker, podcast host, and community builder, sharing practical insights on scaling AI from pilot projects to enterprise-wide adoption.

The Road to Useful Robots

This talk explores the current state of AI-enabled robots and the issues with deploying more advanced models on constrained hardware, including limited compute and power budgets. It then moves on to what's next for developing useful, intelligent robots.

About the Speaker

Michael Hart, also known as Mike Likes Robots. is a robotics software engineer and content creator. His mission is to share knowledge to accelerate robotics. @mikelikesrobots