JSExperts - Advanced Tech Lab

About JSExperts Advanced Tech Lab

We are a small team of young people, with different backgrounds, but same passion for solving problems in new ways.
We take inspiration from the world around us, how it works, interacts and evolves to fuel our ideas and development no matter if they are computer vision algorithms for object detection and recognition, or data analytics.

Research directions

Kalray MPPA 256 Manycore application development

www.kalray.eu

We've been interested in multicore processors since 2009, and recently we have been evaluating the Kalray MPPA-256 Manycore platform for high-performance computing, computer vision as well as 4K HEVC video encoding.

Object Analysis,
Object Recognition,
Automated Recognition

Computer Vision is one of our main areas of research nowadays, mainly focusing on vehicle traffic analysis and computing different properties.

Neural Networks

Taking inspiration from nature to solve practical problems, we are focusing our research on running Neural networks on Parallel/Multicore CPUs

Big Data Analytics

Evaluating performance of Kalray 256 and 1024 core systems on huge datasets

Custom software development for TIBCO Statistica Enterprise for big data applications.

Humanoid Robots

We are partnering with Sanbot in delivering beautiful robotic augmented experiences

Performance Monitoring

A client-server system, monitoring in real time the staff entry/edit data performance

Projects

Computer Vision research

Georgi Botev, Iliyan Gochev,
Stefan Chahanov

We are researching different computer vision algorithms (CVA) applicable in the areas of traffic analysis.

Color spaces and how they affect CVA
Shapes detection, for example for road signs detection
Lines and features extraction
Heuristics methods: mathematical
Heuristics methods: physical, nature inspired

Currently we are focused on finding out stable methods for feature extraction specific to the traffic analysis domain.

Technologies: C++, OpenCV, Kalray MPPA-256

Automated image recognition system

Stefan Chahanov

A thesis for Stefan Chahanov's high school graduation. The system combines automatic image acquisition, clustering and classification, learning and producing results.
The basic idea is that we have a picture, and we ask the system if there is a specific object in that picture. The system may or may not have prior knowledge of what that object is. If it doesn't know it, it searches for information on the Internet and tries to learn what that object is and then detect it.
As an output the system gave predictions in the form of images, what it considered the requested object was.

Technologies: Java, TileEncore-Gx 36 multicore board, Linux.

Evaluating TilePro CPUs for Neural Networks

Iliyan Gochev

A diploma thesis for Iliyan Gochev's BS.C. degree in applied mathematics, this project was our first dive into the world of multicore CPUs. The main goal of the project was to try out and evaluate Tilera's TilePro architecture for executing artificial neural networks (ANNs).
Two different tasks were selected for the evaluation: the first was for optical character recognition (20,000 records, 18 variables used in the model building), the second was predicting credit default (1000 records, 15 variables used in the model building). It was both educational on the platform specifics, as well as the theory of ANNs. The diploma thesis was considered as innovative by the board of examiners at the Technical University - Sofia.

The source was written in C++, which allowed for easy port to the Kalray MPPA-256 architecture recently and evaluating the platforms side by side.
Future investigations include using Kalray's SigmaC for developing more natural neural data processing.

Technologies: C++, TilePro 64 core CPU in 1U server, Linux.