Whether it’s Deep Blue playing chess or any of the smart devices in your home, it seems that we’re seeing signs of Artificial Intelligence (AI) everywhere.
The truth is that the hype around AI is spreading faster than the actual science. Sure, machines are making huge steps forward. But, while they’re able to mimic behavior on specific tasks, they're not capable of thinking or acting like humans.
In the physical security industry and elsewhere, there are a lot of claims about what current versions of AI are capable of. Like other forms of technology, if we don’t have a fact-based understanding of its potential, AI will fail to meet our unrealistic expectations.
These days, AI, machine learning, and deep learning are buzzwords that get thrown around, but although related, they mean very different things.
What is AI?
In data science, AI refers to a fully functional artificial brain. Housed in a machine, it is a self-aware intelligence that can learn, reason, and understand. It can also increase its knowledge without human operator input.
The goal of AI research is to create an intelligence that can understand its surrounding world, acquire inputs, and, from these experiences, learn skills that were not previously programmed into it. While we're years away from this type of true AI, data scientists have made tremendous strides in several areas of research, including machine learning.
What is machine learning?
Machine learning involves teaching a machine to use inputs and historical information to improve its performance without being explicitly programmed to do so. Instead of coding instructions that don’t change over time, programmers use machine learning algorithms and datasets to train a computer to assess, alter, and ultimately improve its own computational processes.
What is deep learning?
One of the several types of machine learning is deep learning, which uses task-specific algorithms to help train a computer to properly classify inputs.
To do this, programmers work with data that has been organized or labeled in a predefined manner. They teach a computer to apply correlations to new inputs by associating thousands, often millions, of possible inputs with corresponding labels that a computer can understand.
Once the computer has ingested and classified a new input, programmers measure the accuracy of the machine’s classification. In the event that the machine gets it wrong, programmers assess the degree of the error and then make adjustments along the network to keep the machine from repeating this misclassification in the future. In this way, they train the computer to improve its ability to recognize new inputs.
How does Genetec use this science?
At Genetec, we use deep learning in our automatic license plate recognition (ALPR) solution and another area of machine learning—called unsupervised—in Genetec Citigraf™, our decision support system (DSS).
With our ALPR system, we're training our algorithms using a structured dataset of raw LPR images and a limited set of possible classes or outputs. Structured datasets consist of data that has been organized or labeled in a predefined manner. In this case, the datasets include labeled images of a wide variety of license plates.
The goal is to have the system take an image of the rear of a car that it has never seen before and be able to output the license plate characters, its state of origin, and the make of the vehicle. It does this by comparing the new image with labeled images in its database. It then calculates the probability that the new image belongs to a specific pre-determined set of classifications.
Our current ALPR offerings contain Deep Neural Net classifiers that are very efficient at reading characters, rejecting bad reads, and recognizing a license plate’s state of origin.
Where AutoVu takes advantage of supervised machine learning, Citigraf employs unsupervised machine learning.
Unsupervised machine learning tackles very narrow problems by analyzing data that has not been organized or labeled in advance in order to find patterns. In this case, the computer is looking for discernable patterns in the data and searching for an unknown output or “ground truth.”
How machine learning can keep your city safe
Citigraf Insights is a crime prediction and resource deployment software-as-a-service (SaaS) that helps cities and law enforcement deploy their physical resources more efficiently based on predicted trends in crime.
To do this, it utilizes an unsupervised machine learning algorithm to estimate how the occurrence of different types of crime can influence the risk of other crimes occurring in the future. In these cases, there are no “ground truths” in the original problem and the answers are learned from the data.
With its correlation engine, Citigraf Command identifies possible relationships between data points, including events. It works in real-time and associates data together to help law enforcement and other public safety personnel build a complete picture of an incident.
So, while machines are in no way ready to take over the planet, they’re certainly becoming an increasingly important tool to help keep us safe. For our part, we are committed to supporting research in data science that has the potential to provide benefits for people the world over.
Want a peek into a day in the life of a Genetec data scientist? Read Sean’s blog.