AI and Machine-Learning

#AI, #Artificial Intelligence, #Machine-Learning, #Deep-Learning, etc. are the trendy and streaming terms of today's social media and blogs. Companies are thinking and speculating on how AI should or would boost their processes, offering or performance and finding the correct answer seems to be a bit tricky. I dare to say AI is like a child that needs to be guided in the right direction. We give examples, we teach how to know right from wrong, and we gently guide him/her forward. We need to build or provide the best base for growing up — the ground where it is easy to take the steps into the future. Of course, it is good to recognize a child's strengths and weaknesses. Support him/her in the various stages of growth through the means we feel are the best at that moment. We should do the same when we incorporate AI into our services or systems. Teach to know right from wrong, build ground or base for learning, create different models that support autonomous learning.

Today there are plenty of different software libraries, pre-trained models and cloud computing services for machine-learning or deep-learning. You need to recognize what it is that you want the intelligence to guide, where to find enough high-quality data to teach the AI, and how to use the AI in your services.

Benemen R&D is studying and using intelligence already in the Benemen services. It used to route calls or to generate information that can be used to lead people and services. The key to machine-learning is the data, a lot of data, high-quality data. In the call services, it is possible to recognize the caller/client emotion during the call, more accurately than using Speech-to-text services and analyzing the context via used words and sentences. 

Of course, I can not give you examples from our R&D projects, but I can provide some sample. I try to avoid buzz-word-bingo as much as I can, but some keywords I have to use. The example video is from my home security camera, where I use YOLOv3 (Real-Time Object Detection, with Deep Learning), Darknet-CUDA-COCO dataset combined with OpenCV4 — scripted with Python. Yes, you can use SSD or Fast/Faster R-CNN, but I selected YOLO, where the speed and accuracy (one-stage detector) are in balance.