Machine learning is used in a variety of ways, including artificial intelligence and image recognition technology. In recent years, however, in the United States, it has succeeded in finding a metallic glass composite pattern 200 times faster than before by utilizing machine learning.
Metallic glasses (metallic glasses) are materials that cause glass transition even among amorphous metals that have no regularity in the arrangement of elements. It has high corrosion resistance and abrasion resistance, which is one of the next generation materials expected to realize higher rigidity and lighter material than steel. However, metal glass has been difficult to commercialize because of its limited combination for 50 years after its discovery. There are millions of combinations of these metals, so it is not easy to experiment with all the great candidates.
However, Northwestern University and the Stanford Linear Accelerator Center, NIST, have introduced machine learning for metal glass production and have succeeded in quickly creating and examining hundreds of samples, which have found three new metal glasses: Science Advances ) Reported in a research paper.
There have been instances where machine learning has been used to predict production. However, this announcement differs from the measurement results in that it quickly predicts and reflects the results in the next machine learning process or experiment. First, we collected 6,000 samples of metal glass production data collected over 50 years using a machine learning algorithm. After learning this, we create sample alloys in two ways and conduct X-ray inspection of these alloys. The results are collected and used for machine learning and other samples. Through this process, it is now possible to find one metal glass per two or three pieces of metal glass in 300 to 400 pieces of data. It is said that the introduction of machine learning increased the speed needed to discover metal glass 200 times faster than before.
The machine learning algorithm used in this experiment has the advantage that it can be applied to other researches as it does not need to understand the existing theory. In the past, human beings can be liberated in the process of non-creative experiment that human beings have to make, so that human beings can concentrate on other tasks requiring intuition and creativity.
In addition to these fields, machine learning is also expected to be used in the medical field. Last year, MIT researchers also attempted to apply machine learning to the medical field. Professor Regina Barzilay, who was diagnosed with breast cancer in 2014, found out that she lacked information in the medical field in search of information about her disease treatment. The doctor writes the information obtained from the patient and handles the correlation based on the basic statistical analysis. But this process can be said to be primitive on the computer science side.
Only 1.7 million people are diagnosed with cancer each year in the United States. Only 3% of clinical trials are registered. Currently, medical research is dependent only on the data obtained through it. To treat cancer, the remaining 97% of patients need to be analyzed for treatment-related information.
Professor Basilay has been working on cancer treatment by combining machine learning with medical data. The research team used natural language processing tools to extract clinical information and DB from 108,000 cancer treatment pathology reports. The database accuracy is as high as 98%, which means that it has efficiently managed a huge amount of work that can not be done with the power of the person. The information in this database is an attempt to make a model that can be deduced through machine learning. In addition, we are trying to apply it to preventive medicine through machine learning. It is to find signs that lead to breast cancer by combining deep learning with information that is hard to decipher with eyes. It is expected that it will help early breast cancer patients who can not be identified by the doctor ‘s eyes and to predict patients who can easily relapse cancer. Providing physicians with data and data analysis techniques through machine learning is expected to improve medical care and benefit more patients.
In fact, in 2016, Google researchers reported that they had done more than just expertise in early detection of diabetes-related eye disease through machine learning. Diabetic retinopathy caused by diabetic complications is also the number one cause of blindness in adults. There is a risk of blindness if left untreated, but it can be treated by early detection. It is important to find the disease as soon as possible to avoid blindness.
Of course, ophthalmologists can detect the onset of diabetic retinopathy by looking at the retinal state, but the problem is that the number of specialists is not enough. The team worked with 54 US eye surgeons to try diabetic retinopathy through deep running with 128,000 retinal images. As a result, 9,963 people were surveyed through deep learning, and the Google algorithm showed an index of 0.95, which is a combination of specificity and detection sensitivity. This is higher than the ophthalmologist average score of 0.91. Of course, this finding is only a way to detect diabetic retinopathy. The need for additional clinical work in ophthalmologists, such as combining other approaches, is further needed. Nonetheless, combining machine learning techniques with new materials, such as synthesis of new materials and medical care, can help alleviate a significant portion of non-creative work, while at the same time helping to expand medical benefits and early detection of patients in the medical field Noteworthy.