# Real world example that HTM algorithm or Numenta's framework can be applied to

Hi,

I am interested in bring Numenta’s framework (location-based framework) to some real world problems

Such as autonomous car can analyze the traffic patterns.

Thank you

1 Like

Here’a a good one showing HTM’s anomaly detection applied across data domains.
Of course the more you can specify your application goal – and even show sample data – the more we can advise you.

2 Likes

Hi,

Thank you so much for the information!

Currrently, we are interested in applying machine learning into auto-driving car. We want to build a model or a system which can let car learn the traffic pattern of a road (such as the car in front of it is decreasing the speed within a certain distance then this car should also decrease its speed).

Do you have any advice of where I should look for and whether HTM theory and Numenta’s framework can do that?

Thank you so much again

I think this particular show of machine intelligence may be most reliably done with a simple logic block, basically:

``````distance = get_distance(sensor_values)
if distance < safe_distance:
car.slow_down()
``````

I say that because this logic (obvious as it is to us humans) relies on an understanding of the natural world – knowing that fast-moving objects too close together means danger for us. We innately know that our survival depends on being afraid of certain kinds of scenarios, like car crashes.

No ML model (that I know of) has that innate intuition. It could understand that distance is low or decreasing at a certain rate, and use that pattern/trend to predict the next distance values, but it doesn’t reflexively understand why it’s bad like we do.

This is just a general point, and I’m no expert of self-driving ML.

HTM theory is based on the idea that the Neo-cortex builds a “sensory-motor” model of the world. So each sensory patch (like your eye, hand or ear) is constantly predicting what its next inputs will be, and how they’'ll be effected by different actions.

The current out-of-the-box HTM applications for anomaly detection and forecasting (HTM Studio & NuPIC) just do the sensory part – passively taking in sensor values and learning the sequential patterns continually as they appear.

The HTM framework does have the building blocks for including the motor part – accounting for movements of the sensor itself in one research code here for example: https://github.com/numenta/htmresearch/blob/master/htmresearch/frameworks/location/location_network_creation.py)

Including these elements in application is a lot more complex, calling for a deep understand of the algorithm structure and function. I’d recommend looking for a way to apply the out-of-the-box HTM systems. Maybe there are internal car components which could be monitored for anomalies, and do early detection of abnormal behaviors to avoid some engine failure or something?

I’m no expert on cars either. I’m just thinking of anything that generates sequences which are likely to have patterns over time. It can learn those patterns and detect when they are changing or turning to chaos.

3 Likes