Researchers at MIT’s Singapore-based research venture, the Singapore-MIT Alliance for Research and Technology (SMART) interdisciplinary Research Group on Future Urban Transportation, have created a synthetic framework called Theory-based Residual Neural Network (TB-RESNET), It combines discrete choice model (DCM) and deep neural network (DNN) (also known as deep learning) to improve individual decision analysis used in travel behavior research.

Their paper, [“Theory-based residual neural networks: A synergy of discrete ‘choice models and deep neural networks “] (www.sciencedirect.com/science/art… , recently published in Transportation Research: Part B, SMART researchers explain the TB-RESNet framework they developed and demonstrate the advantages of combining DCMs and DNNs methods, demonstrating that they are highly complementary.

As machine learning is increasingly used in transportation, DCMs and DNNs, two unrelated research concepts, have long been seen as competing research methods.

By combining these two important research paradigms, TB-RESNet leverages the simplicity of DCMs and the expressive power of DNNs to produce richer findings and more accurate predictions for individual decision analysis, which is important for improving travel behavior research. The TB-RESNet framework developed is more predictive, interpretable and robust than DCMs or DNNs, and its results are consistent across a wide range of data sets.

Accurate and effective analysis of individual decisions in the everyday context is crucial for mobile companies, governments and policy makers seeking to optimise transport networks and address transport challenges, especially in cities. Tb-resnet will eliminate existing difficulties faced by DCMs and DNNs and allow stakeholders to take a holistic, unified view of transportation planning.

“Improving insights into how travelers make decisions about how to travel, where to go, when to leave, and planning activities is critical to urban transportation planning by governments and transportation companies around the world,” said Shenhao Wang, a postdoctoral fellow and lead author at MIT’s Urban Transportation Lab. I look forward to further developing TB-ResNet and its applications in transportation planning, as it is already recognized by the transportation research community.”

“Our Future Urban Transport research team is focused on developing new paradigms and innovating future urban transport systems in Singapore and elsewhere,” said Jin Hua Zhao, SMART FM principal investigator and associate professor in MIT’s Department of Urban Studies and Planning. “This new TB-RESNet framework is an important milestone that can enrich our investigation into the impact of decision-making patterns on urban development.”

Tb-resnet can also be widely used to understand the individual decision cases illustrated in this study, whether it is about travel, spending, or voting.

In this study, the TB-RESNet framework was tested in three scenarios. First, the researchers used it to predict travel pattern decisions between traffic, driving, self-driving cars, walking and cycling, all of which are major travel patterns in urban environments. Second, they assessed risk choices and preferences when uncertain monetary returns were involved. Examples of this include insurance, financial investments, and voting decisions.

Finally, they looked at timing, measuring the tradeoff between current and future monetary rewards. A typical example is in transportation development, where shareholders analyze infrastructure investments with large down payments and long-term benefits to make decisions.

The study was conducted by SMART and supported by the National Research Foundation of Singapore (NRF) under its Research excellence and Technology Enterprise Park (CREATE) programme.

The Future Urban Transport Research Group leverages new technologies and institutional innovations to create next-generation urban transport systems that improve accessibility, equity, safety and environmental performance for citizens and businesses in Singapore and other metropolises around the world. FM is supported by the CREATE Programme of the Singapore National Foundation.

SMART was founded in 2007 as a collaboration between MIT and Singapore’s National Research Foundation. SMART acts as a knowledge and innovation hub for research interaction between MIT and Singapore, conducting cutting-edge research projects in areas of interest to both Singapore and MIT. SMART currently includes an innovation center and five interdisciplinary research groups. Antimicrobial properties, key analytics for personalized medicine, disruptive and sustainable technologies for agricultural precision, frequency modulation and low-power subsystems.

Original link: news.mit.edu/2021/smart-…