Use quantum computing to protect AI from attack

Machine learning is a field of artificial intelligence (AI) where computer models become experts at a variety of tasks by consuming large amounts of data. This instead of humans explicitly programming this level of expertise.

For example, modern chess AIs do not need to be taught chess strategies by human masters but can ‘learn’ them independently by playing millions of games against copies of themselves.

This is valuable in situations where writing down explicit instructions is impractical, if not impossible – how to define a mathematical function that can tell you a picture contains a cat or a dog?

Children never learn any such function but instead watch many examples of cats and dogs, and eventually understand their differences.

Machine learning is about replicating this process in a computer.

But despite their incredible successes and increasingly widespread deployment, machine learning-based frameworks are still vulnerable to adversarial attacks – that is, malicious tampering with data causes them to fail. in surprising ways.

For example, image classification models (which analyze images to identify and recognize a variety of criteria) can often be fooled by adding cleverly crafted changes (called perturbations). into the input image is so small as to be unrecognizable. human eyes. And this can be exploited.

The continued vulnerability to attacks like this also raises serious questions about the safety of deploying machine learning neural networks in potentially life-threatening situations. This includes applications such as self-driving cars, where the system could be confused when driving through an intersection by a harmless graffiti on a stop sign.

At a critical time when AI development and deployment are rapidly evolving, our research team is looking at ways quantum computing can be used to protect AI from these vulnerabilities,

MACHINE LEARNING AND QUANTUM COMPUTING

Recent advances in quantum computing have created much excitement about the prospect of enhancing machine learning capabilities with quantum computers. Various ‘quantum machine learning’ algorithms have been proposed, including quantum generalizations of standard classical methods.

Generalization refers to the ability of a learning model to properly adapt to new, never-seen-before data.

It is believed that quantum machine learning models can learn certain types of data significantly faster than any models designed for current or ‘classical’ computers.

Conventional computers operate with bits of data that can be either ‘zero’ or ‘one’ – a classic two-level system.

Quantum computers work with ‘qubits’, states of two-level quantum systems, that exhibit strange additional properties that can be exploited to solve certain problems more effectively than conventional machines. their classicity

What’s less clear, however, is how widespread these speedups will be and how useful quantum machine learning will be in practice.

This is because although quantum computers are expected to learn a broader class of models efficiently than classical computers, there is no guarantee that these new models will be useful to most people. all the machine learning tasks that people really care about. These could include medical classification problems or general AI systems.

These challenges prompted our team to consider what other benefits quantum computing could bring to machine learning tasks – beyond the usual goals of improving efficiency or accuracy.

Shield AI from attacks

In our latest research, we propose that quantum machine learning models can be better protected against adversarial attacks generated by classical computers.

Adverse attacks work by identifying and exploiting features used by machine learning models.

However, the features used by general quantum machine learning models are inaccessible to classical computers and are therefore invisible to adversaries equipped only with classical computing resources.

These ideas can also be used to detect the presence of adversarial attacks using classical and quantum networks simultaneously.

Under normal conditions, both networks will make the same prediction, but when there is an attack – their output will be different.

While this is encouraging, quantum machine learning continues to face significant challenges. Most prominent among them is the huge gap in capabilities that separates classical and quantum computing hardware.

Today’s quantum computers are still significantly limited by their size and high error rates, making them unable to perform lengthy calculations.

Tremendous technical challenges remain, but if these can be overcome, the unique capabilities of large-scale quantum computers will certainly bring surprising benefits across many fields.

Banner: Getty Images