MInD – Machine Intelligence Designer
Deep learning platform
for engineers
On the MInD platform, you can create, operate, and maintain deep learning solutions based on image, video, or sensor time series data.
5 business reasons to use MInD
Simplified AI development
All software in one place from data management to operation, making AI development on a unified platform simpler and more cost-effective.
For business leaders, the question is no longer whether AI solutions should be implemented, but rather HOW and WHEN?
The MInD platform includes all development tools from data management to maintenance, making the implementation and development of AI on the unified platform faster and more cost-effective.
The platform is designed for engineers, so we offer the opportunity to develop AI with your own engineering resources without software development qualifications. In our experience, an employed engineer supported by the MInD Platform can be as productive as 3 highly qualified data scientists.
We stand on the shoulders of giants
The MInD platform offers Google and Microsoft AI technologies on convenient user interfaces. We carefully select the most time-tested software components of the tech giants that are immediately available to our customers on the MInD platform.
We maintain a close relationship with academic researchers.
We follow the development of the latest AI methods that we incorporate into the MInD platform if they also work well on our customers’ data.
Continuous availability and training
On-line, on-site support with a professional Central-European team.
- With a professional Hungarian team, we help to assess and solve business and engineering problems, either in personal meetings or at an online conference.
- Using AI is not always the shortest way to solve a problem. We will tell you when it is worth using AI and when it is more practical to work with classical methods.
- We provide detailed and searchable written and video documentation for development and maintenance.
- We enter into business specific Service Level Agreements (SLAs) so your company’s digital transformation can be planned, tracked, and reproduced.
Cloud and on-premise benefits together
MInD solutions can be developed and run on both cloud and on-premise machines, and can be combined to suit customer needs.
Benefits of cloud-based MInD services:
- Resources and databases that can be shared across multiple sites
- Fast scalable storage and computing capacities
If necessary, the MInD services can also be run on on-premise computers, which has the following benefits:
- Fast and secure in-house data transfer
- User interfaces without any delay
- Network independence
Joint digital product development
We serve the introduction of new digital products with customized development.- We design and develop the AI component of your new product on the MInD platform.
- We work and think together with your company’s engineering team. Together, we test, improve and integrate the AI solution into the product.
- We deployed 17 MInD solutions worldwide, developed in collaboration with local engineers.
- We assume some of the growth risk by offering payment structures related to product launch, such as e.g.
- Delayed license fees,
- Sales-related license fees
Tech background of MInD
Data management
For us, data is above all. This is the basis of our philosophy. We have created a system in which we store all the information together with its relations, be it raw measurement data, annotations, training parameters, or the results and logs of the deployed solutions.- Millions of images or time series are dynamically cached, so our clients can work seamlessly, as if everything were running on their own computer.
- Our databases can operate on both cloud and on-premise servers, so you can decide about the storage and the security of your data.
- You can export or import your data or any subset of it at any time, either through a graphical interface or API.
Annotation
We know from experience that cleaning and annotating the raw data can often take 90 percent of the total development time. We do our best to perform these tasks quickly and with high quality.- Our user interface can be customized from interactive panels.
- There are equally convenient tools for left- and right-handers, and dynamically selectable hotkeys speed up the manual annotation.
- Automatic and semi-automatic annotating models can be trained, which can speed up the process up to ten times.
- Any number of annotators can work together on our servers. Any past state of the annotations can be restored using the versioned history of the data.
Modelling
A transparent model can be created only if its internal state can be traced step by step. To this end, we created an interactive modeling user interface in which any function of the model can be examined during operation.- Our neural networks can be supplemented with pre- and post-processing data operations.
- Built-in AutoML features automatically suggest a neural architecture for the given data quantity and quality.
- A pipeline can be built from augmentations to achieve the amount of data needed for robust training.
- The results of the training are automatically reported, which can be reviewed at any time.
- No programming knowledge is required, but custom code can be inserted if necessary.
Installation
The central element of the Service Oriented Architecture (SOA) is the Manager Service, where all the development and production services register themselves. Using the Manager Service, the models stored on the Database Service can be deployed to Brain Services running in production.- A model (Brain) can be installed on either an on-premise, cloud-based or edge computer.
- Deployment into production systems can be scheduled and can be subject to conditions.
- The process of deployment is versioned, traceable and restorable.
Maintenance
A central maintenance server can be used to monitor the performance, load and health status of the deployed models, whether they run on an industrial computer, a Raspberry Pi device or Android-based systems.- The results of the models and their aggregate statistics can be evaluated on the central server.
- If the performance of a model deteriorates, new training data can be collected to complement the original training set, and a more robust model can be deployed back to the device.
- In the event of a hardware problem or misuse, the device will automatically request maintenance from the central server.
Hardware requirements
The following minimum hardware requirements must be met to run MInD services.| Brain service (Developer) | Brain service (Runtime) | Database service | Manager service | |
|---|---|---|---|---|
|
CPU |
Intel Core i7 Comet Lake or newer |
Intel Core i3 Kaby Lake or newer |
Intel Core i7 Comet Lake or newer* |
Intel Core i7 Comet Lake or newer* |
|
RAM |
16 GB |
8 GB |
32 GB |
16 GB |
|
GPU compute capability |
6.1+ ** |
- |
- |
- |
|
GPU RAM |
4+ GB |
- |
- |
- |
|
HDD |
1 TB |
- |
4 TB (RAID)/NAS/SAN |
1 TB |
|
SSD |
512 GB |
128 GB |
512 GB |
128 GB |
|
Cable bandwith |
1 Gb/sec*** |
100 Mb/sec**** |
1 Gb/sec |
1 Gb/sec |
* or equivalent AMD processor
** https://developer.nvidia.com/cuda-gpus
*** it depends on the number of clients and access to the database server
**** it depends on the transmission requirements and the measurement format