, perpetually coming out with news and updates. 

 of the software at the beginning of September, which was 

, with whom we collaborated on a blogpost released soon after. Now, 

 — features that allow unprecedented customisation of the software for all the peculiar and unique needs our clients have. This is the first step towards next-gen simulation software that learns and adapts while it is being used. We now have a lot to show. So we did. We went to the biggest tech conference in Europe, Web Summit 2023, and showed 

Web Summit 2023 took place in the vibrant city of Lisbon. This global technology conference is renowned for bringing together innovators, technology leaders, and industry pioneers worldwide. We were looking forward to the Summit as it is a unique occasion to meet future clients, collaborators, and investors in person, engage with the latest trends, gain valuable insights, and foster meaningful connections within the tech community. Web Summit provides an unparalleled platform for networking, collaboration, and staying at the forefront of emerging technologies. Throughout the three days we spent there, we immersed ourselves in the energy of innovation and actively seized the opportunities that Web Summit 2023 has to offer.

Web Summit 2023 brought together 70,236 people and companies, redefining the tech industry. We presented 

 to hundreds of interested visitors and established some vital connections, it is a great pleasure to meet people who need your product, seek it, and are excited to meet you and finally find what they’re looking for.

Eventually, the shared experiences, visionary ideas we exchanged and the dynamic discussions we had will shape the future of our industry and bridge the worlds of “real-life” physics and digital twins through simulation. 

 attracted a lot of attention through global partnerships and connections to other representatives of emerging technologies. It bridges industries, integrating AI-driven numerical simulations with scientific learning principles to revolutionise how industries approach solving engineering and scientific challenges. The team from DimensionLab was right at the forefront of the AI & machine learning companies area, represented by our team consisting of Braňo Kršák, Head of R&D Partnerships, Peťo Macinsky from the sales team and Dávid Jopek from the marketing team.

During the event, our team unravelled the core mechanics and key components of the architectures of state-of-the-art scientific machine learning models. We demonstrated the vast potential and the groundbreaking opportunities PhysicsML brings to the table. We proudly displayed the potential of the technology we are leveraging and the power of the software we develop.

WebSummit 2023 was a powerful networking event where we started building new partnerships and opened numerous collaborations — yet that was just the first step. We look forward to welcoming our connections from Lisbon to the 

 platform, where we can roll up our sleeves and help the industry get faster, save costs, and make simulation far more sustainable and intuitive than it used to be.

websummit2023-titlepicture

DimensionLab at Web Summit 2023 - Lisbon: Spreading the word about Siml.ai

 more effective and user-friendly, we are more than happy to introduce the 

 feature, which gives users a hassle-free experience in making payments using their preferred mode and buying top-up credits in just a few clicks. The usage of the Environments in Siml.ai will be based on the user’s available credits (although, 

 – still, unlocking different functionality across the Siml.ai platform is conditioned based on the monthly or yearly subscription plan the user is on).

The Billing page has been spruced up to view available credits in real time while also enabling users to buy credits at any point in time. The whole process is then handled by Stripe Checkout, which will ensure a quick and secure payment experience. Users can change the amount of credits to buy on the checkout page (it supports setting an amount between $10 to $10,000 in one transaction).

Apart from this, there is also the feature of providing an estimate of required credits based on the current usage. This forecast aims to give users a fair idea of how many credits they would need to complete their ongoing training, inference, and simulation tasks so they can maintain the credits as required.

The compute credits feature helps users concentrate more on the business usage of the product and pay only for the amount of credit they intend to use. This makes using Environments, which is the core part of the Siml.ai training and inference infrastructure, much more predictable. Deployed Environments will now only incur a charge each minute from already available credit in the user’s account from the time it goes into a running state – we are not charging for the time it takes to get fully deployed. Also, when the user runs out of available credits, all currently running Environments will be auto-stopped. You can then buy more compute credits, start your Environments again and continue training.

To make the available credits more visible when creating new Environments, we have added the credits  information in the Environments listing page as well:

Now, when you try to deploy an environment with a selected server Tier whose price exceeds your current amount of credit, Siml.ai will prevent you from doing this, and an error message will be displayed:

To sum it up, we hope that by changing Environments system in Siml.ai to a credits-based system, we’ll make using Siml.ai more approachable and predictable, but also add an ability to gain credits in different ways - for example, buying it through a classic invoice (if your business or academic institution requires this) or gather them for free through our 

. As always, we’re eager to listen to your 

 and hope this new feature will help you make the most out of Siml.ai for your use cases!

Compute credits closeup

Introducing Compute Credits

Today marks a significant milestone for the simulation software landscape as we at DimensionLab unveil the latest 

 version 0.3, featuring the most requested “Custom Equations” support. But we wanted to take this even further, and we’re super excited to also introduce 'Plugins', with their first usage categorized as the templates for custom equations, which can be shared with other users! These features are set to transform the way professionals and enthusiasts alike create and utilize AI-driven simulations and will allow them to bring their own mathematical models into 

Custom equations: bring your own mathematical models

The newly-introduced custom equations capability is a game-changer. It allows users to define their own mathematical models using partial differential equations (PDEs), lifting the limitations of relying only on predefined ones. With the integration of the 

, building much more complex simulations is now more sophisticated but still very accessible. The process is simplified with pre-defined Python code templates, enabling users to develop their own equations without starting from scratch.

Equation node in Siml.ai's Model Engineer no-code visual editor now shows a "Change equation" button, which will open up a browser-based code editor in a popup-modal. It's designed for efficiency and ease, while still offering flexibility for code customization.

The introduction of plugins opens doors to collaboration and knowledge sharing within the Siml.ai community. Users can save their custom equations as plugins, making it easier to not only reuse them in their own simulators, but also share their templates publicly. This feature fosters a collaborative environment for innovation be it for personal use in future projects or to contribute to the community's growing library. We’re excited to introduce the first use case for plugins and our team is already working on the next iterations to extend the plugins to each part of Siml.ai’s Model Engineer, to allow everybody to extend the platform in many ways.

To ensure users get the most out of these features, our software engineer 

 has prepared a concise tutorial video. This video guides users on integrating custom equations into their simulations with a practical example.

Our commitment to supporting our users on their journey of discovering how Siml.ai can help them in their simulation workflows is unwavering. The Siml.ai documentation at 

 will be continually updated to provide detailed guidance with the addition of how to leverage the custom equations feature.

Try it out and let us know about your experience using these new features. We’re looking forward to the next era of Siml.ai by opening up the platform to people who would like to help us with the process of extending it and start building the foundation of a marketplace for the “AI-driven engineering” community.

Custom equations in Siml.ai

Siml.ai v0.3 released: Custom equations and plugins

In one of the previous posts we have talked about brand new 

, this time we focus on extention of well-known Deep Operator Network (DeepONet). DeepONet is a deep learning framework for solving forward and inverse problems involving nonlinear operators. It can learn to map a function to another function by using pairs of input-output functions as training data. DeepONet can handle complex and high-dimensional problems without requiring any knowledge of the underlying equations or boundary conditions.

A recent paper by Liu et al. named "A novel deeponet model for learning moving-solution operators with applications to earthquake hypocenter localization" proposes a novel method for solving inverse problems using deep neural networks. The method is called Extended DeepONet (X-DeepONet).

The Extended DeepONet method enhances the DeepONet framework by incorporating two key changes to the architecture of the DeepONet. While the DeepONet uses only the dot product operator for the output, X-DeepONet incorporates also subtraction and sum operations. The results of all of these operations are fed to a single neural network named Root Net, which produces output. This extended architecture is better at generalization of more complex tasks, such as variable source position determination, as in earthquake hypocenter localization. 

The paper demonstrates the effectiveness of the Extended DeepONet method on several velocity models, namely Marmousi velocity model, smooth velocity model and OpenFWI velocity model. While with the first two models the architecture proved to predict highly accurate positions of the source, the third model is order of magnitude more complex, and the error reached with this model opens space for further improvements. 

The Extended DeepONet method is a promising technique for solving inverse problems using deep learning. It can potentially be applied to various engineering domains, such as fluid mechanics, solid mechanics, electromagnetics, and heat transfer. The paper opens up new possibilities for combining physics-based modeling and data-driven learning for solving challenging engineering problems.

deepOnet-picture

Example article sub-heading

Earthquake hypocenter localization with X-DeepONet

Today, we are thrilled to announce the beta release of 

, the next-generation software platform that is designed to transform the landscape of AI-driven numerical simulation. Offering a host of new and enhanced features, 

 is primed to redefine the boundaries of simulation and analysis. 

Let us walk you through the exciting features of the upcoming 

 beta that make it the trailblazer in the simulation industry.

New design of Model Engineer’s visual flow editor

The intuitive visual editor for simulators has been redesigned for slightly better UX. Additionally, there’s now the ability to easily copy nodes, which accelerates the simulation setup process, enabling users to replicate nodes with just a few clicks, saving time and effort. However, this redesign is still a work in progress, so expect huge improvements and direct visualizations of 3D geometries in next versions.

Support for parameterized inputs and variables in equations

 introduces the much-anticipated feature of parameterizing inputs into the simulator’s neural network (defined in neural network node settings and usable in constraint node settings) and variables in equations (by toggling the “Parameterized?” checkbox in equation node settings). This is an important capability that facilitates a much more flexible simulator training and fast inference across potentially thousands of configurations of initial and boundary conditions, enabling users to easily tweak variables and observe the effects in real-time, thereby speeding up the analysis and optimization processes.

Parameterized inference in Simulation Studio

Following the introduction of the ability to train simulators with parameterized inputs, it will be straightforward to use them in Simulation Studio. This addition will enable users to conduct multiple high-fidelity multiphysics simulations for different conditions in a few minutes, pushing the boundaries of what is possible in the realm of numerical simulation. Furthermore, we have expanded Simulation Studio by integrating the ability to segment simulation data into multiple planes in 3D. We've also added functionality to adjust the source of magnitude according to the outputs from the simulator.

Web-based documents with embedded interactive simulations

To showcase an ability for connecting additional services to 

, in a way of a web app, through which you can do quick simulator inferencing and visualization, we’re piloting a new approach in 

 that we think will redefine how we’re building real-time and interactive analysis tools. Underlying infrastructure within Environments now comes with a bundled web app that leverages the powerful Panel library from Holoviz ecosystem, with basic inferencing configurations and visualizations of 2D plane sections of 3D simulation outputs. In the near future, you’ll be able to create all kinds of applications with tools like Panel or Streamlit that connect to 

’s ecosystem of trained AI-based physics models.

 now empowers users with the option to fine-tune publicly available or their own simulator models. This feature ensures that you can tailor simulators to your specific needs, achieving results that are more aligned with your objectives. Since it's a pretty significant feature, we’ll delve deeper into this topic in a separate post in upcoming days.

In our pursuit to foster learning and innovation, we are excited to start offering academic licenses. Institutions and academics can reach out to us for requests at 

. We are excited to be a part of the academic journey of numerous researchers and students around the globe.

To assist our users in making the most of these exciting new features, we have updated our documentation which can be accessed at 

. Here you will find comprehensive information on the latest features and how to use them effectively to maximize the potential of your simulations.

As we usher in a new era of numerical simulation with the beta launch of 

, we invite you to be a part of this journey -

 in fact, we’re giving out 1 month for free additionally to our 14-day free trial when you go through our 

Join us in exploring the unprecedented capabilities and features that 

 has to offer. Together, let's forge a future where simulation is not just a tool, but a catalyst for rapid innovation and progress. 

 community. We look forward to fostering a collaborative and innovative environment with you. 

feature_img_simlai

Announcing Siml.ai beta: On the road to the future of AI-Driven numerical simulation

In parallel to working hard on our upcoming beta release of 

, which will launch an upgraded architecture and improved features, we took part at 

 2023 that was held in Košice during 19th and 20th September. The pulse of innovation was palpable, underscoring Slovakia's position as a hub of technological advancements. The team from DimensionLab was right at the heart of the buzz, taking center stage where Michal Takáč, our CEO, sparked conversation on harnessing the power of AI for engineering and science, specifically, how we merge artificial intelligence and nuanced neural networks for simulating physics in virtual environments. The forum served as a platform to forge new partnerships, foster discussions, and shed light on the prospects of applying the latest approaches in scientific machine learning across diverse industries.

A dive into the AI-driven numerical simulation

In a world that is increasingly reliant on data and algorithms, the focus has gradually shifted towards making machine learning more scientific and geared towards solving complex real-world problems. DimensionLab, a pioneer in this field, is developing 

 software platform that integrates AI-driven numerical simulations with scientific learning principles to revolutionize how industries approach solving engineering and scientific challenges.

During the conference, our team unveiled the core mechanics and key components of state-of-the-art scientific machine learning models’ architectures. We demonstrated the vast potential and the groundbreaking opportunities that PhysicsML brings to the table.

As we interacted with industry professionals and enthusiasts, it became increasingly clear that the bridging of physics and machine learning could herald a new era of innovation. The integration facilitates a synergistic relationship that promises to unlock untapped potential in various industries, from raw materials extraction, manufacturing and healthcare to energy and beyond (to space).

Our discussions revolved around how PhysicsML could serve as the linchpin in accelerating the transition from traditional simulation models to more dynamic, accurate, and predictive frameworks. The conference was a testament to the interest and the widespread acknowledgment of the potential impacts that this fusion of disciplines can have in reshaping industry norms and standards.

A notable highlight of the SlovakiaTech forum was the panel discussion about the ecosystem of financial support available to Slovak startups. There are many more ways of how startups can secure funding than there were a few years ago. We can see higher availability of VC funding, many accelerators, mentorships on how to secure international funding, and the narrative was that the state should put more emphasis on creating funding opportunities in forms of grants for early-stage startups, and cut back on bureaucracy in the application process. Branislav Kršák, our Head of R&D Partnerships, summarized the views from the startup’s point of view, and together it made for a fruitful discussion and Q&A session.

Looking ahead: DimensionLab as a strong force in AI-driven engineering

As we wrap up our experience at SlovakiaTech 2023, it’s clear that the technological innovation in Slovakia is accelerating at a rapid pace. Also, we’re delighted about the positive response and the fruitful discussions that ensued during the conference are a clear indication of the tremendous potential that lies ahead.

At DimensionLab, we are more committed than ever to spearheading the movement towards integrating AI-driven numerical simulations into tech companies relying on such simulations. We believe that by fostering a community of like-minded professionals and enthusiasts, we can pave the way for a brighter and more efficient future.

We extend our gratitude to all the participants and organizers of SlovakiaTech 2023. It was an honor to share our vision about a new era of technological advancements that are grounded in science and propelled by innovation in AI.

For all of you who couldn’t be there and would like to know more about physics simulation and how we can leverage AI in the industrial metaverse – ping us at 

Michal Presenting Solo

DimensionLab at SlovakiaTech 2023: Spreading the word about PhysicsML

Ever since the launch of large language models such as ChatGPT, scientists, engineers and public are trying to find various ways to automate complicated computing tasks with natural language processing. Researchers from Brown University have proposed an interesting framework which enables to use natural language processing (NLP) to help with scientific computing tasks. They called the framework MyCrunchGPT, which is a chatbot that can assist you with scientific machine learning (SciML) problems.

SciML can be challenging and time-consuming, as it involves many steps such as pre-processing, problem formulation, code generation, post-processing and analysis. Moreover, SciML may not be accessible to everyone, especially those who are not familiar with coding or physics.

This is where MyCrunchGPT comes in. MyCrunchGPT is a chatbot that uses ChatGPT to communicate with the user and guide them through the SciML workflow. ChatGPT is based on GPT-3, which is one of the most advanced NLP models that can generate natural and coherent texts based on user prompts. CrunchGPT leverages ChatGPT to understand the user's queries, provide relevant information, generate code snippets, plot graphs, and produce summary reports.

To demonstrate the capabilities of CrunchGPT, the authors of the paper present two examples: (1) design and optimization of 2D NACA airfoils, and (2) solving prototype fluid mechanics problems using physics-informed neural networks (PINNs). In both cases, CrunchGPT can help the user with various tasks such as defining the problem, choosing the parameters, running the simulations, visualizing the results, and validating the accuracy. The user can interact with CrunchGPT using natural language or simple commands, and get instant feedback and guidance.

The authors also provide a web-based user interface for CrunchGPT, which allows the user to choose from different options and see the results in real time. The user can also request a comprehensive summary report that contains all the information and outputs generated by CrunchGPT. The authors claim that CrunchGPT can be a useful tool for researchers, engineers, students, and educators who want to use SciML for their problems.

CrunchGPT is an innovative and exciting framework that shows the potential of using NLP for scientific computing. By using ChatGPT as a conductor, CrunchGPT can orchestrate the entire SciML workflow and make it easier and more accessible for everyone. If you want to learn more about CrunchGPT, you can read the full paper 

CrunchGPT Feature Image

MyCrunchGPT - A ChatGPT Framework for scientific machine learning

Have you ever wondered how to solve complex equations that describe how things change over time and space? These equations are called partial differential equations (PDEs) and they appear in many engineering problems, such as modeling fluid flow, heat transfer, wave propagation, or chemical reactions. Solving PDEs can be very challenging, especially when the domain is irregular or the equation is nonlinear.

A new machine learning technique called Laplace Neural Operator (LNO) can help solve PDEs more efficiently and accurately than traditional methods. LNO is a type of neural network that learns to map the input parameters of a PDE to its solution, without requiring any meshing or discretization of the domain. LNO uses the Laplace transform to decompose the input space into frequency components, which makes it suitable for problems with non-periodic signals or transient responses.

In a paper "LNO: Laplace Neural Operator for Solving Differential Equations", researchers from Brown University and MIT demonstrated the superior performance of LNO over another neural network technique called Fourier Neural Operator (FNO) in solving various PDEs. FNO uses the Fourier transform to decompose the input space, but it has limitations when dealing with non-periodic signals or undamped scenarios. LNO, on the other hand, can handle these cases better and exhibit exponential convergence.

The researchers tested LNO on three ordinary differential equations (ODEs) and three PDEs that have applications in physics and engineering. For example, one of the ODEs is the Duffing oscillator, which models a nonlinear spring-mass system that can exhibit chaotic behavior. One of the PDEs is the Euler-Bernoulli beam equation, which describes the deflection of a beam under load. The results showed that LNO can approximate the solutions of these equations more accurately than FNO, especially in capturing transient responses or complex boundary conditions.

LNO is not only a powerful tool for solving PDEs, but also a more interpretable and generalizable one. LNO incorporates the pole-residue relationship between the input and output spaces, which means that it can learn the natural frequencies and modes of the system. This can help engineers understand the underlying physics and phenomena better. Moreover, LNO can generalize to unseen inputs or domains better than FNO, which means that it can solve new problems with less data.

LNO is an exciting and promising direction for machine learning and engineering research. It shows how neural networks can help solve hard engineering problems that have been traditionally tackled by numerical methods. By applying LNO to real-world problems, engineers can achieve faster, more accurate, and more reliable solutions, while also gaining insights into the underlying physics and phenomena.

If you are interested in learning more about LNO and how it works, you can read the full paper 

. At DimensionLab, we are working on implementation of LNO into 

laplace-neural-operator

Laplace Neural Operator

Imagine a world where the design of wind turbines, the aerodynamics of vehicles, the optimization of electronics, the efficient conversion and storage of hydrogen fuel, and the exploration of alternative energy sources took merely a fraction of the time it currently does. Traditionally, these tasks have relied heavily on the creation of mathematical models - a painstakingly meticulous process carried out by human experts - followed by rigorous validation through numerical simulations. This age-old approach, while robust, is increasingly facing its limitations in a rapidly advancing world.

Enter the age of Artificial Intelligence, paving an unprecedented path in the realm of physics. From the seed of an incomplete problem description, AI has the ability to generate a mathematical model of extraordinary accuracy - a model so complex it could confound even the most brilliant of human minds. Our role, as humans, now pivots to becoming the validators of these AI-created models, rigorously testing and scrutinizing them from all perspectives to ensure their validity.

AI's disruptive presence extends further, offering a tremendous advantage when it comes to the utilization of these models in numerical simulations. The computational speed of these AI-generated models outpaces traditional methods by several thousand times, resulting in faster visualizations of physical phenomena for problem-solving, optimization, fault prediction, or technological advancement.

Embracing this accelerated process, we stand on the precipice of a research and development revolution. With the capacity to explore a vastly larger realm of possibilities in the same amount of time as before, we are pushing the boundaries of innovation and discovery. The future is here, and it is powered by AI.

It's essential to see how this revolutionary technology is applied in real-world scenarios. The exciting fusion of AI and physics isn't just theoretical; it's being leveraged right now, disrupting industries and accelerating progress. To truly appreciate the transformative power of AI in this field, let's delve into three compelling use-cases that exemplify this cutting-edge amalgamation at work.

Digital twin of Earth with fast weather prediction

An AI-driven approach was used to model global Earth weather leveraging 

work), which accurately forecasts high-resolution, fast-timescale variables such as the surface wind speed, precipitation, and atmospheric water vapor. It has important implications for planning wind energy resources, predicting extreme weather events such as tropical cyclones, extra-tropical cyclones, and atmospheric rivers.

, generating a week-long, accurate, 3-dimensional global weather prediction in under 10 seconds, reaching 

’s (IFS) state-of-the-art Numerical Weather Prediction (NWP) model, with 

 when dealing with model trained on meteorological dataset of global 30km vertical weather data. Once trained, however, FourCastNet 

 to generate a forecast than the IFS model. FourCastNet could be trained on currently available GPU hardware in about two months (~1,400 compute hours on a single A100 GPU) with available meteorological dataset from 40 years of global meteorological data collection (

Thanks to the speed of such a weather prediction model, it can be used in a wide variety of applications, such as energy and agricultural planning, transportation, and disaster mitigation. Here, we present few animations from visualizing the interactive simulation of FourCastNet model built with 

 framework and transformed into digital twin of Earth through 

In a world facing the dire realities of climate change, the quest for cleaner, greener energy is of paramount importance. 

, a leading name in renewable energy, has taken a giant leap in this direction, 

harnessing the power of physics-informed neural networks (PINNs)

 to revolutionize wind energy generation.

Before we delve into their breakthrough, it's important to understand the challenges of wind farm optimization. High-fidelity, transient simulation of a single wind turbine using traditional simulation software with LES turbulence modeling could take an astonishing 30-40 days. Now, consider a wind farm comprising tens, if not hundreds, of such turbines. Examining and reducing possible wake effects in real time, in tandem with fine-tuning wind farms to handle a multitude of wind and weather conditions, necessitates hundreds or even thousands of simulation runs and iterations. Traditionally, this was rendered infeasible due to the formidable time limitations and financial costs involved.

Enter the world of PINNs and AI-assisted positioning. Siemens Gamesa leveraged AI-based physics simulators trained using PINNs, leading to a dramatic acceleration in simulation times. Today, they can simulate tens of wind turbines in mere minutes, unlocking the power of inverse physics simulation for lightning-fast optimization of energy output.

But what does this breakthrough mean for our future? Imagine a world where efficient, green energy generation isn't just a distant ideal, but an everyday reality. The swift simulation and optimization process enabled by PINNs will allow us to rapidly set up and optimize wind farms, harnessing the untapped potential of wind energy at an unprecedented scale and speed.

This is more than just a technological triumph; it's a beacon of hope for a cleaner future. As we march forward, these advancements in AI and physics simulation will play a pivotal role in driving our transition towards sustainable, renewable energy sources.

Blood flow simulation in intracranial aneurysm

Computer simulation is highly beneficial for understanding the physics of blood flows within our bodies, particularly when dealing with dangerous health issues like aneurysms. Thanks to modern technology, we can create 3D geometries from the images of these anomalies using MRI and CT scans. However, such 3D representations vary for each patient, making it quite challenging to simulate blood flow and pressure for every unique case using traditional simulation software. It's a time-consuming process, but in these life-or-death situations, every minute counts. Quick understanding and decision-making can mean the difference between life and death.

With the help of physics-informed neural networks, doctors could analyze each anomaly in a few minutes instead of days or weeks. It’s like having a fast-forward button for crucial medical decisions. They can even explore multiple 'what if' scenarios in a single day, visualizing the potential outcomes of various treatments.

Armed with a comprehensive report of anticipated physiological factors, surgeons can better prepare for complex procedures and help avert disastrous medical events. In essence, these AI-powered tools will provide our medical professionals with a detailed roadmap for each unique patient journey, potentially saving more lives along the way.

PhysicsML

Harnessing AI: The New Frontier in Physics Simulation

We are happy to announce a launch of Siml.ai alpha release to the public!

 is a software platform for AI-based numerical simulation. It provides a visual editor for building high-performance physics simulators that leverage deep learning techniques (also called surrogate models). Additionally, the platform features tools for training these simulator models in powerful GPU-based cloud servers and easy-to-use AI inferencing and 3D visualization pipelines. There's a paradigm shift happening in the world of physics simulation. Deep learning methods are outperforming even the most optimized classical numerical solvers. The key components of these methods are 

 variations of Fourier neural operators (FNOs) and deep operator networks (DeepONets)

eliminate the need for meshing the geometries.

In complex numerical simulations, it’s important to precisely quantify the underlying physical mechanisms in order to analyze them. High-dimensional scientific simulations are computationally very expensive to run, and solvers and parameters must often be tuned individually to each system studied. 

Machine learning approaches tend to considerably decrease the load on computational resources by reducing the dimensionality of the studied problem and by factoring in the fact that 

AI-based simulators need to be trained only once

numerical simulation is computed during the inference of the model

. This, compared to computing physics every time a simulation is re-run, brings a large efficiency benefits together with shortened time for the actual computations as trained models.

Siml.ai platform provides a set of tools to easily create, train and deploy powerful physics-informed neural network models with complex physics and custom partial differential equations. Building an AI-based simulator is extremely easy in our Siml.ai Model Engineer.

Siml.ai Model Engineer is a tool for visually creating and optimizing general-purpose "learnable simulators" (e.g. for simulating fluid flows, multiphase flows, mechanical deformation, solid-fluid interaction and so on) using deep learning techniques. 

Our aim is to make the process of building such simulators simple, visual and understandable

Within the Siml.ai platform, Model Engineer has a powerful position of being a funnel through which we plan to build a large database of trained models by individual engineers, researchers or companies. This will allow us to create a new market for ready-to-use, "packaged" AI simulators – essentially a marketplace in which these entities can earn money from simulators trained by them by allowing other users to use them in their projects through 

Simulation Studio is a hybrid between web-based and native application for solving engineering and scientific problems leveraging pre-trained and optimized models of AI-based physics simulators. Since the simulations are computed by inferring these models, the time it takes to compute one timestep even of highly irregular simulation domain is in low tens of milliseconds, resulting in real-time simulation and visualization of physical phenomena. High-fidelity visualization rendering is achieved by leveraging the powerful Unreal Engine under the hood.

3D interface for high-fidelity, interactive in-situ visualizations

If you want see what is planned for the future, make sure to check out our 

We're also giving out 100% discount for 1 month for everybody who completes our 

 - you can get more information in our community Discord, which you can join 

Blog