Technology

 

Cogito’s massively parallel architecture delivers an extremely fast, compact and energy efficient pattern learning and recognition solution.

 

 

Neural Network Technology

 

Cogito Instruments products are based on the NeuroMem® hardware neural network technology.

Some of the most important characteristics of this technology are :

 

Truly massively parallel pattern matching capability

Traditional software-based technologies are sequential.
Even the most modern CPUs, DSPs or GPUs, cannot compare thousands of patterns simultaneously.
On the other side, the technology integrated in Cogito Instruments products enables true 100% parallellism.
The response time is therefore extremely low.

Compact and low power

Thanks to its unique pattern learning and matching hardware acceleration engine, Cogito Instruments products pack tremendous punch in a small form factor.
In addition, because our neural network hardware does not need to run at GHz frequencies to achieve superior performance, the resulting power consumption is extremely low.
We can typical deliver the equivalent of 250GOps per Watt.
In other words, we deliver the processing power of a high-end PC equipped with GPU acceleration, in a credit-card format. And we don’t need cooling!

Purely digital technology (no analog spiking neurons)

Unlike other implementations of artificial neural networks, our technology is purely digital.
This means that it’s fast, reproducible, observable and that we can easily save and restore the content of our neurons.
Once a machine has been trained, it’s very easy to take the training data and copy it to other machines without having to go through the entire learning again.

Extremely low and constant latency. Recognition time is typically 10us no matter how large the dataset is

The performance of Cogito Instruments products is very impressive thanks to their true parallel nature,
«1 out of n» recognition happens in microseconds no matter how large «n» is.
As your system learns and accumulates more and more data, your system performance remains constant.

Ability to deal with novelty or conflicting responses and to learn in real time to improve quality of response “on the job”

This is probably the most important feature Cogito Instruments brings for industrial systems.
Once the initial learning phase has reached the desired level of recognition quality, your system can still continue to learn even when it has been deployed “on the job”.
When a pattern is not recognized properly (false positive, false negative, unknown or doubt in the classification), new knowledge can be learned without having to re-compile the entire knowledge base. Unlike Deep Learning, the NeuroMem® technology offers “on-the-job training” capability. New knowledge acquired on one machine can be shared quasi-instantly between all machines thanks to the save and restore capability and the networkability of National Instruments platforms.

The benefits for your applications

Neuron content (knowledge) can be downloaded and uploaded. One machine can train other machines.

Real-time data analytics close to the sensors, embedded in your system.

Short response time makes it possible to integrate intelligence inside a real-time control loop.

Behavior is reproducible and traceable.

Precious data stays local, in your machines. No need to send your secrets to the Cloud.

Performance stays constant as more knowledge is accumulated. Productivity does not degrade over time.