Pasteur Labs & ISI

When asked "Do your clients primarily use #RealWorldData, #SyntheticData, or a combination of both in their work? How do your tools handle each type, and how do you integrate them into your #MachineLearning #algorithms?", our Founder Alexander Lavin shed some light on the engineering reality (https://lnkd.in/e49B4Jav): "That may be the #1 fault of existing technologies that we’re addressing: the inability to utilize real-world data. One cannot simply plug-in data—this is a fundamental flaw with CAE software, and it’s muddied with incumbents trying to throw ‘AI this’ and ‘chatbot that’ into the legacy CAE software toolchains... Industries are always naive about how hard it is to operationalize data in ways that build value — firehoses are expensive, and hard to tame in environments that matter... You’ll see in the upcoming Pasteur Labs platform launch, that we’ve built automated dataflow pipelines for interoperable #CAE toolsets and ML-based problem solving — that is, the necessary parameterization and labeling for algorithms (learning, optimization, and data assimilation), data structures befitting differentiable physics and interfaces for efficient, end-to-end modeling workflows." 👉 follow us Pasteur Labs & ISI (@SimAI4Science on https://lnkd.in/gznPy7Bh) for more demos, product rollouts, and real-world results.

We are thrilled for Hannah Bollar, who's been nominated as Developer of the Year for the 5th Annual Polys! Congratulations, Hannah!!

We launched our new blog last week at Pasteur Labs & ISI, and I'm super proud to share a slice of my colleagues' brilliant minds! 🤖 Niklas Heim posted the first of a two-part dive into how JAX, the differentiable programming framework, implements automatic differentiation —the backbone of AI. 🔬 Dion Häfner explained how we approach answering fundamental science questions in ways that hold up in the real world. He deep-dives into one of the paradigms that shapes how we do our work — namely, execution-driven science. Excited to hear what you think! https://lnkd.in/e2UwpExN

*New Insights* 🚀 Check out the latest from Pasteur Labs & ISI, where R&D Lead Engineer Dion Häfner describes how key ingredients, + understanding of problems from multiple perspectives and across multiple dimensions; + pragmatic engineering for high-quality execution; + touching the real world, doing it at scale; + demonstration and iteration, always; together can enable "execution-driven science", and thus progress our path to reshaping the scientific method for the machine age (and in ways that matter!): https://lnkd.in/eXtiJDZ8 #SimulationIntelligence #AI4Science #PasteursQuadrant #FrontierTech

Pasteur Labs Insights is officially live! https://lnkd.in/emaevUtC Insights is our blog, where we’ll share our perspectives and learnings on topics like: scientific machine learning (SciML), digital engineering, human-AI teaming, the future of industrial R&D, AI/ML ‘in the wild’, and all things Simulation Intelligence! We’ll also share updates about our products, talks, and research. Check out our first few posts: 🧮 [technical] The first of a two-part dive into how JAX, the differentiable programming framework, implements automatic differentiation, the backbone of AI 🏟️ [industry] A recap of our inaugural workshop on the industrialization of SciML, co-organized with and hosted by ICERM at Brown University (repost from spring 2024) 👩🏫 [talks] A tour of the many talks and presentations by Pasteurians during summer and fall of last year. ...and more coming soon: product demos, expert talks, research results... ! We’d love to hear what you think, and what Insights you’re eager to read about. Keep an eye on this space (and X @SimAI4Science ) for more — enjoy.

We're excited for another year of field-leading dev tutorials from our Simulation Intelligence Advocate, Felix Koehler 🚀 https://lnkd.in/e6NHS3mQ #SimulationIntelligence #AI4Science #SciML

The "Earth Intelligence Engine" creates realistic, simulated satellite imagery from the future, by combining a physics-based model with generative AI. These images can help users see what their region might look like after a hurricane or other flooding event, in much more vivid detail than traditional, color-coded flood maps. Developed by researchers from MIT, with collaborators from academia, government, and industry, the tool has the potential to help policymakers and residents make better decisions about how to prepare in advance of a storm, and when to evacuate. Björn Lütjens, a postdoc in the MIT Department of Earth, Atmospheric and Planetary Sciences (MIT EAPS) who led the research while he was a doctoral student in MIT AeroAstro, explains, “The idea is: One day, we could use this before a hurricane, where it provides an additional visualization layer for the public. One of the biggest challenges is encouraging people to evacuate when they are at risk. Maybe this could be another visualization to help increase that readiness.” Media Lab Director Dava Newman, a co-author on the paper, says, “People relate to their own zip code, their local environment where their family and friends live. Providing local climate simulations becomes intuitive, personal, and relatable.” The researchers’ paper, “Generating Physically-Consistent Satellite Imagery for Climate Visualizations,” is out today in IEEE Transactions on Geoscience and Remote Sensing (IEEE Xplore). Authors: Björn Lütjens (MIT EAPS); Océane Boulais (Scripps Institution of Oceanography); Farrukh Chishtie (The University of British Columbia and Peaceful Society, Science and Innovation Foundation); Natalia Díaz Rodríguez (Universidad de Granada); Margaux Masson-Forsythe 🏳️🌈 (Surgical Data Science Collective (SDSC)); Ana Mata-Payerro (MIT Human Systems Laboratory); Christian R. (Max Planck Institute for Biogeochemistry); Aruna Sankaranarayanan (CSAIL MIT); Aaron Piña (USDA Forest Service); Yarin Gal (University of Oxford); Chedy Raissi (Riot Games); Alexander Lavin (Pasteur Labs & ISI); Dava Newman (MIT Media Lab/MIT AeroAstro). https://lnkd.in/gQWiuKuq

Important food-for-thought on #AI #ML advances vs #DigitalEngineering practice in this lecture! Here are a few key quotes from our R&D Lead Dion Häfner: 👨🍳 "Ideally a useful AI model, like we're building as plug-n-play commodities for industrial R&D, would have a few key ingredients for [human-machine teaming]: an understanding of *causality* s.t. can transfer understanding to us humans, *compatibility* for fusing your various models with mixed data end existing simulators is really attractive if not fundamentally necessary, and *reliability* cannot be an afterthought... or AI that speaks our language." 🔍 "thousands of nodes [in Deepmind's GraphCast] is hardly a small textbook example, where in the millions of nodes it gets interesting, so this SOTA in literature is several orders of magnitude away from commercial scale." 🕸 "Nvidia is all about Fourier Neural Operators and their kinda-platform called Modulus, which has inspiring science works, like Nvidia weather forecasting models based on FNOs, that is published and promoted but oops! They cannot work on unstructured meshes!" 🚀 "Pasteur (pasteurlabs.ai) is building a pure platform of AI surrogate pipelines for real-world engineering: it's critically important to validate surrogate models on actual data, that work in actual environments that are not cherry-picked, that interface with existing simulators... don't stop at training the models–that's the easy part! Integrating these models such that end-users can interact with them, that requires much more effort and thought, and is really unchartered territory..." If this fascinates and motivates you, check out #AI4Science #SimulationIntelligence opportunities with the Pasteur Labs team: https://lnkd.in/gz38vSg7 💪

It was an honor to join David Barnes, PhD and Marne Marotta today in a discussion on the militarization of AI and its implications for the future battlefield. I can't think of a more relevant topic given the acceleration of the so-called "AI arms race" and the rise of dual-use AI — all with Veteran's Day coming up and our recently-concluded presidential race in the backdrop. It was an incredibly rich conversation! We touched on topics ranging from how AI might reduce human error on the battlefield ... to cybersecurity (both using and targeting AI) ... to over-reliance on AI systems ... to AI training and awareness ... to how AI might impact the nature of war altogether ... and everything in between. Several themes emerged — my personal favorite being the importance of human-AI interaction design. Thank you to Arnold & Porter’s Veterans and Affiliates Leadership Organization (VALOR) and to everyone who attended!

The word is out! Turns out AI is NOT a magical solution to all the things. 😉 But what about AI for simulating physics in the context of industrial design and engineering? Seems like using AI for something science-y should be fine, right? Well, like most things, it's a bit more complicated than that. Check out this panel, hosted by nTop, to find out what it takes to use AI in in these contexts so that it is reliable, trustworthy, and effective. When: 31 Oct, noon EDT 👻 Register: https://lnkd.in/eQRz4_6H Bonus: Pasteur Labs & ISI CEO, Alexander Lavin, will be speaking on the panel!