The HAIBAL deep learning toolkit project is an ambitious journey that aims to provide the best artificial intelligence development tools.
Launch of HAIBAL’s development
Every journey has a beginning and let’s bet that we will succeed in developing a complete deep learning library that is easy to use, ergonomic and suitable for academics and engineers.
Making our first convolution in LabVIEW
We are now able to do our first 2D convolution in LabVIEW
LabVIEW architecture project
All the main layers are now coded in native LabVIEW.
Successful signal prediction test
This example, implemented natively in the HAIBAL library, aims to understand how to train signal prediction model. Our idea is to help our novice users to start simply with machine learning and then hit the moon !
Successful Importation from keras test
Our first version of HDF5 importator is now functionnal and make possible to load, edit and run any model from Keras Tensorflow. We import a VGG 16 model from Keras with success.
Successful MNIST test
This example, implemented natively in the HAIBAL library, aims to understand how to train, predict, save and load a basic model. Our idea is to help our novice users to start simply with machine learning and then hit the moon !
Successful Tiny yolo import from keras and run test
Importing a Tiny YoloV3 Model from Keras with success.
In order to better optimize our inference, we have adopted a new approach to memory. The kernel memory system is now managed in C language.
Cuda integration v1
Model run now on Nvidia GPU.
Launch of our youtube channel
To improve our visibility and better communicate with our community, we are launching our youtube channel on which we will share training videos and applications using HAIBAL.
Release of HAIBAL 1.0
After one year of development we are proud to release the first version of HAIBAL, the LabVIEW deep learning toolkit.
Give the student access to the HAIBAL toolkit through our university sponsorship program.
A new premium program of selective partnership for independents allowing them to promote and distribute the HAIBAL toolkit.
Cuda integration improvment
The full parallelism process is now possible with our new cuda memory manager.
New save file system
We adopt H5 and JSON file format to save HAIBAL models.
Keras tensorflow compatibility
We integrate the load of H5 keras model format.
We integrate the load of pt Pytorch model format.
Reinforcement learning examples integrations
Release of some reinforcement learning examples.
XILINX FPGA ULTRASCALE +
One of our goal is to make HAIBAL running on Xilinx FPGA ultrascale+ offering our custom new possibilities.
We will integrate Intel oneAPI toolkit making HAIBAL compatible with the INTEL Multiarchitecture Open Accelerated Computing (CPU / GPU / FPGA)
HAIBAL will be available on UBUNTU
Release of our computer vision module
We will integrate our own vision module into the toolkit. It will allow our users to easily capture their data from a camera. This module will also allow to add overlay and segmentation mask.
HAIBAL deep learning toolkit will also be distributed by National Instuments.
Implementation of the training and our deep learning certifications.