Aalto University

Machine learning in chemistry - algorithms help finding minimum energy paths and saddle points more effectively

Reproduced from J. Chem. Phys. 147, 152720 (2017), with the permission of AIP Publishing.

Reproduced from J. Chem. Phys. 147, 152720 (2017), with the permission of AIP Publishing.

Machine learning brings new possibilities to many areas of research - also chemical research, which is shown in a recent dissertation written at the Finnish Center for Artificial Intelligence.

Olli-Pekka Koistinen, doctoral candidate at Aalto University, developed machine learning algorithms based on Gaussian process regression to enhance searches of minimum energy paths and saddle points, and tested how well the algorithms work.

In theoretical chemistry, finding minimum energy paths and saddle points is one of the problems that consume most time and computational resources. The bottleneck of the computation is the accurate evaluation of energy and forces for each atomic configuration, which typically needs to be performed at hundreds of points in the configuration space.

Algorithms utilizing machine learning can reduce the number of observation points and expensive energy evaluations to a fraction of what is required by conventional methods and thus speed up the computation.

Minimum energy paths lie on a potential energy surface that describes the energy of a particular system - a molecule, for example - in terms of particular parameters. Usually these parameters show the locations of the atoms. The local minimum points of the energy surface correspond to the stable states of the system. The minimum energy paths connect these points and describe possible reaction mechanisms.

“As an orienteer, I see this energy surface as a map. The stable atom configurations are shown as depressions on the map, and the minimum energy path is a route between two such depressions. It stays as low as possible all along the way. The highest point of the path is at a saddle point where you can get from one depression into another one staying as low as possible,” Koistinen explains.

Traditionally, researchers have searched for minimum energy paths and saddle points using iterative methods that proceed on an energy surface with small steps. With the help of machine learning and statistical models, previous observations can be utilized to model the energy surface, and the goal can be reached with significantly fewer iterations.

Therefore, machine learning offers a more effective and lighter and thus also cheaper and more ecological option. It can also open new possibilities for studying problems that have not been feasible with traditional methods. “This is another example of a research topic in which machine learning methods can be helpful,” Koistinen says.

Neural network for elderly care could save millions

Illustration by Matti Ahlgren / Aalto University

Illustration by Matti Ahlgren / Aalto University

A deep neural network model helps predict healthcare visits by elderly people, with the potential to save millions.

If healthcare providers could accurately predict how their services would be used, they could save large sums of money by not having to allocate funds unnecessarily. Deep learning artificial intelligence models can be good at predicting the future given previous behaviour, and researchers based in Finland have developed one that can predict when and why elderly people will use healthcare services.

Researchers at the Finnish Centre for Artificial Intelligence (FCAI), Aalto University, the University of Helsinki, and the Finnish Institute for Health and Welfare (THL) developed a so-called risk adjustment model to predict how often elderly people seek treatment in a healthcare centre or hospital. The results suggest that the new model is more accurate than traditional regression models commonly used for this task, and can reliably predict how the situation changes over the years.

Risk-adjustment models make use of data from previous years, and are used to allocate healthcare funds in a fair and effective way. These models are already used in countries like Germany, the Netherlands, and the US. However, this is the first proof-of-concept that deep neural networks have the potential to significantly improve the accuracy of such models.

‘Without a risk adjustment model, healthcare providers whose patients are ill more often than average people would be treated unfairly,’ Pekka Marttinen, Assistant Professor at Aalto University and FCAI says. Elderly people are a good example of such a patient group. The goal of the model is to take these differences between patient groups into account when making funding decisions.

According to Yogesh Kumar, the main author of the research article and a doctoral candidate at Aalto University and FCAI, the results show that deep learning may help design more accurate and reliable risk adjustment models. ‘Having an accurate model has the potential to save several millions of dollars,’ Kumar points out.

The researchers trained the model by using data from the Register of Primary Health Care Visits of THL. The data consists of out-patient visit information for every Finnish citizen aged 65 or above. The data has been pseudonymized, which means that individual persons can not be identified. This was the first time researchers used this database for training a deep machine learning model.

The results show that training a deep model does not necessarily require an enormous dataset in order to produce reliable results. Instead, the new model worked better than simpler, count-based models even when it made use of only one tenth of all available data. In other words, it provides accurate predictions even with a relatively small dataset, which is a remarkable finding, as acquiring large amounts of medical data is always difficult.

‘Our goal is not to put the model developed in this research into practice as such but to integrate features of deep learning models to existing models, combining the best sides of both. In the future, the goal is to make use of these models to support decision-making and allocate funds in a more reasonable way,’ explains Marttinen.

The implications of this research are not limited to predicting how often elderly people visit a healthcare centre or hospital. Instead, according to Kumar, the researchers’ work can easily be extended in many ways, for example, by focusing only on patient groups diagnosed with diseases that require highly expensive treatments or healthcare centers in specific locations across the country.

The research results were published in the scientific publication series of Proceedings of Machine Learning Research.

Further information
Yogesh Kumar
Doctoral Candidate
Aalto University, Finnish Centre for Artificial Intelligence
yogesh.kumar@aalto.fi

Pekka Marttinen
Assistant Professor
Aalto-yliopisto, Finnish Centre for Artificial Intelligence
Phone +358 50 5124362
pekka.marttinen@aalto.fi

How ecological is AI? How to compose with a neural network? Come to AI exhibition to find out!

ctd-opening-ad.png

A new exhibition by Aalto Digi Platform and FCAI combines science and arts and sheds light on the history, state of art, and future of AI.

Aalto Digi Platform and FCAI are hosting a science and art exhibition that takes its visitors to a journey in the world of Artificial Intelligence. The Connecting the Dots exhibition, taking place at Aalto University’s campus in Espoo, Finland, aims to increase our understanding about the history of AI, the current state of art, and what the future holds.

“AI is already part of our everyday life, but what we can do with it and how it works is still quite a mystery to most people. With this exhibition, we hope to give answers to questions they may have,” says Saara Halmetoja, the exhibition coordinator from Aalto Digi Platform.

The exhibition is suitable for people of all ages. Visitors can step into a sound environment in which they can control the surrounding voices with their movements; compete against AI in a music game; and compose together with a neural network, among other activities.

The exhibition dives into the core of AI research. Visitors will see how a machine learns to classify and predict things, as well as how it forms sounds, pictures, text, and molecules. The exhibition presents robotics and shows how AI adapts to real-world situations.

Connecting the Dots discusses the history and ecological aspects of AI. While AI can help the clothing industry in getting rid of textile waste, artificial neural networks – mimicking human brain’s neural networks – consume enormous amounts of energy.

The exhibition brings together the diverse fields of Aalto University from natural sciences, technology, and architecture to design and arts. The scientists, artists, and students behind the exhibition come from diverse backgrounds; some of them are “pure” AI researchers, while others use AI tools in their scientific or artistic work.

According to Halmetoja, “if only certain types of people take part in discussions about AI, we will end up designing discriminatory technology. Therefore, we need to make sure that everyone in the society participates in these discussions”

Entrance to the Connecting the Dots exhibition is free and the event is open during Dipoli’s normal opening hours from November 27 until January 15. For larger groups, such as school groups, the minimum age recommendation is 12 years old. If you want to visit the exhibition with a group of people, please contact Saara Halmetoja, the exhibition coordinator, in advance.

Connecting the Dots exhibition - Dipoli, Otakaari 24, 01250 Espoo, Finland - November 27, 2019 - January 15, 2020

Further information
Saara Halmetoja
Exhibition Coordinator
Aalto Digi Platform
Phone +358 50 5720730
saara.halmetoja@aalto.fi