How does artificial intelligence learn?
Machine learning is the process of using computers to detect patterns in a large number of data sets, and then make predictions based on what the computer learns from these patterns. This makes machine learning a specific and narrow artificial intelligence. Fully artificial intelligence involves machines that can perform the thinking capabilities of humans and intelligent animals, such as perception, learning, and problem-solving.
All machine learning is based on algorithms. Generally speaking, an algorithm is a set of specific instructions that a computer uses to solve a problem. In machine learning, algorithms are the rules of how to use statistical data to analyze data. Machine learning systems use these rules to identify the relationship between data input and expected output. First, the scientist gives a set of training data to the machine learning system. These systems apply their own algorithms to this data to train themselves on how to analyze similar inputs received in the future.
One area where machine learning shows the potential is the detection of cancer in computed tomography (CT) imaging. First, the researchers collect as many CT images as possible as training data. Some images show cancerous tissue, and some show healthy tissue. The researchers also collected information on what to look for in the images to identify cancer. For example, this may include the boundaries of cancer tumors. Next, they formulate rules about the relationship between the data in the image and the doctor's knowledge of identifying cancer. Then they pass these rules and training data to the machine learning system. The system uses these rules and training data to teach itself how to recognize cancerous tissue. Finally, the system will get a CT image of a new patient. The system uses the knowledge it has learned to determine which images show signs of cancer faster than any human. Doctors can use the system's predictions to help determine whether a patient has cancer and how to treat it.
The establishment of training data divides machine learning systems into two categories: supervised and unsupervised. If the training data is labeled, the system is supervised. The labeled data tells the system what the data is. For example, CT images can be labeled as cancerous lesions or tumors next to healthy tissue. This means that the machine learning system learns by example.
If the training data is not labeled, the machine learning system is unsupervised. In the case of cancer scans, an unsupervised machine learning system will be given a lot of CT scan and tumor type information, and then let it find what to identify cancer. This saves people from having to label the data used in the training process. The disadvantage of unsupervised learning is that the results may not be as accurate due to the lack of clear labels.
Some machine learning systems can improve their abilities based on the predictive feedback they receive. These are called reinforcement machine learning systems. For example, the system can be informed by the doctor of the results of other tests on whether the patient has cancer. The system can then adjust its algorithm to produce more accurate predictions in the future. US Department of Energy Office of Science: Contributions to machine learning The US Department of Energy Office of Science supports machine learning research through its Advanced Scientific Computing Research (ASCR) program. ASCR has a series of data management, data analysis, computer technology, and related research, which all contribute to machine learning and artificial intelligence. As part of this portfolio, the Department of Energy has some of the most powerful supercomputers in the world.
As a whole, the Department of Energy’s Office of Science is committed to using machine learning to support scientific research. Science relies on big data, and the user facilities of the Office of Science, such as particle accelerators and x-ray light sources, generate a lot of big data. Using machine learning, researchers are identifying data patterns or designs from these devices that are difficult or impossible for humans to detect, at hundreds to thousands of times faster than traditional data analysis techniques.