Cognitive Smart OoD(Out-of-Distribution detection)

The main articles are:

• Video analytics (terms, applications, technologies)
• Computer vision (machine vision)

2024: Cognitive Smart OoD System Introduction

Cognitive Pilot solved a difficult problem in training neural networks, which allows you to ensure the maximum level of security of autonomous control systems and save tens of man-years in their development. The company announced this on February 7, 2024.

The selection of rare interesting data from a video stream is an extremely important stage, especially in the late stages of neural network training. It is he who makes it possible to achieve maximum accuracy of object recognition and is extremely important in solving life critical class problems (critical for human life), which include autopilot systems by tram, tractor and other vehicles designed to ensure the safety of participants in traffic in any conditions.

Cognitive Pilot specialists have proposed a productive method for automatically searching for unusual data in a video stream. They developed Cognitive Smart OoD technology (from the international name of this direction - Out-of-Distribution detection), which allows such a search to be carried out automatically.

The essence of the approach developed by Cognitive Pilot is as follows. If you train the neural network on the most common images encountered during the movement of the vehicle, with which the developers usually have no problems, then the system in this case can only provide a basic, close to industrial level of safety. But in order for the autopilot system to guarantee the safety of the vehicle, it is necessary that the neural network be trained in all situations that can only arise during its operation. To take this into account, the neural network must be trained in all kinds of rare situations, with which great difficulties arise.

The discovery of scientists and engineers Cognitive Pilot means that a tool has been found to automatically train autopilot in all sorts of accidents that can occur when a vehicle moves. This means that in almost all cases our autopilot system is able to clearly understand the situation on the road, in the field, on the rails and adequately respond to it, ensuring maximum traffic safety. This is a huge technological breakthrough, an extremely big event in the AND systems market, "said Olga Uskova, CEO of Cognitive Pilot.

An example of rare, atypical data could be a partially obscured traffic light, a flash of lightning while driving, an animal in a field, an unusual car and even a meteorite.

𝓲

Фото: Cognitive Pilot

Searching for this kind of object in a video stream in a traditional, "manual" way is an extremely time-consuming and tedious procedure that takes AI developers many weeks, or even months.

It's like mining radium, it can take tens of man-years, "says Gennady Savitsky, head of unmanned vehicles at Cognitive Pilot. It is simply unrealistic for a markup specialist to make a high-quality selection of images of this kind by hand. Unique personnel can be one in a million, they are simply worth their weight in gold. We do this, in fact, by pressing the key.

Cognitive Smart OoD technology allows you to correctly classify and organize rare and unusual objects. At the first stage, the datacet is evaluated for the content of rare images. This is carried out using a neural network trained on a sufficiently large diverse data set.

At the second stage, statistical methods are used to search for atypical data and data with rare features in Cognitive Smart OoD.

The ability to automate the selection of rare data is a huge step forward, continuing our developments in the direction of machine learning. In this zone of the AI market for unmanned vehicles, we have accumulated one of the most solid examinations on training neural networks on our own datacets in several segments at once: agriculture, rail transport and driver assistance systems, - concluded Olga Uskova.

Cognitive Smart OoD technology is already implemented in Cognitive Pilot autopilot systems, both for agro-transport and smart trams.