Artificial Intelligence in Materials Science
Over the past 10 years artificial intelligence and machine learning have made significant contributions and disruptions to the way we live our daily lives. From automated driving & parking of vehicles ( Tesla ) to recommending movies to us on a particular day and time ( Netflix ) and beyond have been made possible through the advancements in the field of deep learning where intelligent algorithms are constantly learning and interpreting underlying information in large, diverse data.
Although many areas of businesses and our day-to-day lives are impacted by the advances in the field of artificial intelligence, there are many industries and processes that are yet to cross the chasm to adopt artificial intelligence and machine learning.
With artificial intelligence and machine learning disrupting many industries, there's reasons for why these processes have not been able to impact each industry by 2020. Firstly, the talent required to architect, research, and implement artificial intelligence solutions in each industry have funnelled into the internet technology (IT) industry, leaving gaps in other industries. Secondly, management style in certain industries are quite traditional, making it difficult to convince and allocate additional money and manpower into research and development to develop advanced solutions. Thirdly, and perhaps the most important reason that has left certain industries unaffected by the rapid advances of artificial intelligence is that utilizing "small data" for machine learning and artificial intelligence is an incredibly difficult challenge.
Many of the advances in artificial intelligence have come strictly from deep learning being able to extract deeper relationships and understanding from utilizing more data that is able to achieve higher-and-higher accuracy, such as GPT-3 As many industries and processes inherently have very small amounts of data, these industries are frequently left out from the glamour and growth of the implementation of AI in these other industries, as the data from these processes is frankly too sparse and small.
Deep Learning is getting really good on Big Data/millions of images. But Small Data is important too. Am seeing many exciting applications at Landing AI where you can get good results w/100 images. Hope more researchers work on Small Data--ML needs more innovations there.— Andrew Ng (@AndrewYNg) September 27, 2018
A specific industry where this problem of small data is apparent is materials formulation and development. Small datasets cite exist for these processes as experiments are done pragmatically, inefficient methods exist for tracking experiment data, and many experiments are novel and do not have existing data available. In these development processes, experiments are executed to achieve new specific formulation conditions. Oftentimes to achieve these specifications, a scientist may resort to trial-and-error experiments from the beginning until reaching their specifications. In such a case, time, resources, and products are ineffectively utilized in achieving business goals, wasting precious ti and allowing time for competitors to develop and market the same materials.
This leaves an important question, are there methods that can be exploited in order to learn and grow with this "small data"? In processes such as Computer Vision, there are methods that exist for extracting additional value with small data, such as transfer learning transfer learning. However, transfer learning is often specific in dealing with annotating images and lacks practicability to applicability in other industries, especially material development. Therefore, there exists a strong presence compulsion for companies that can utilize proprietary artificial intelligence that is able to extract underlying information from small data sufficed with key domain knowledge in order to help companies achieve novel material formulation and research in record time. This will enable them to reduce development time, environmental impact, free-up resources and costs while increasing their competitive advantage over their peers.
Polymerize is utilizing years of industry experience gained in the North American, Asian and European markets in combination with award winning AI researchers to develop novel artificial intelligence that learns from your experiment data, allowing you to achieve new product formulations 500% faster than the conventional methods. We utilize state-of-the-art artificial intelligence that learns underlying features from your data combined with our proprietary domain knowledge. This is then combined with a step-by-step learning approach that allows your company to achieve better and better accuracy with each experiment.
Our goal is to help companies achieve efficient and faster development timelines while saving the environment.