In these frameworks, templates are constructed using an iterative process of template estimation and alignment, which is often computationally very expensive. Due in part to this shortcoming, most methods compute asingle template for the entire population of images, or a few templates for specific sub-groups of the data.

We introduce HyperMorph, a framework that facilitates efficient hyperparameter tuning in learning-based deformable image registration. Classical registration algorithms perform an iterative pair-wise optimization to compute a deformation field that aligns two images. Recent learning-based approaches leverage large image datasets to learn a function that rapidly estimates a deformation for a given image pair. In both strategies, the accuracy of the resulting spatial correspondences is strongly influenced by the choice of certain hyperparameter values. However, an effective hyperparameter search consumes substantial time and human effort as it often involves training multiple models for different fixed hyperparameter values and may lead to suboptimal registration. We propose an amortized hyperparameter learning strategy to alleviate this burden by learning the impact of hyperparameters on deformation fields. We design a meta network, or hypernetwork, that predicts the parameters of a registration network for input hyperparameters, thereby comprising a single model that generates the optimal deformation field corresponding to given hyperparameter values. This strategy enables fast, high-resolution hyperparameter search at test-time, reducing the inefficiency of traditional approaches while increasing flexibility. We also demonstrate additional benefits of HyperMorph, including enhanced robustness to model initialization and the ability to rapidly identify optimal hyperparameter values specific to a dataset, image contrast, task, or even anatomical region, all without the need to retrain models. We make our code publicly available at http://hypermorph.voxelmorph.net.

Davis Blalock, a computer science graduate student at the Massachusetts Institute of Technology (MIT) told Science magazine that some of the gains may not exist at all. Blalock and his mates compared dozens of approaches to improving neural networks--software architectures that loosely mimic the brain and found that it wasn't obvious what the state of the art even was. The researchers evaluated 81 pruning algorithms, programs that make neural networks more efficient by trimming unneeded connections. All claimed superiority in slightly different ways. But they were rarely compared properly--and when the researchers tried to evaluate them side by side, there was no clear evidence of performance improvements over a 10 year period.

Artificial intelligence (AI) just seems to get smarter and smarter. Each iPhone learns your face, voice, and habits better than the last, and the threats AI poses to privacy and jobs continue to grow. The surge reflects faster chips, more data, and better algorithms. But some of the improvement comes from tweaks rather than the core innovations their inventors claim--and some of the gains may not exist at all, says Davis Blalock, a computer science graduate student at the Massachusetts Institute of Technology (MIT). Blalock and his colleagues compared dozens of approaches to improving neural networks--software architectures that loosely mimic the brain.

Artificial intelligence (AI) just seems to get smarter and smarter. Each iPhone learns your face, voice, and habits better than the last, and the threats AI poses to privacy and jobs continue to grow. The surge reflects faster chips, more data, and better algorithms. But some of the improvement comes from tweaks rather than the core innovations their inventors claim--and some of the gains may not exist at all, says Davis Blalock, a computer science graduate student at the Massachusetts Institute of Technology (MIT). Blalock and his colleagues compared dozens of approaches to improving neural networks--software architectures that loosely mimic the brain.

Jensen Huang, the billionaire CEO of Nvidia, has made a fortune by supplying the hardware used for artificial-intelligence algorithms. He's now betting that AI is about to become an indispensable part of medicine. In the early 1990s, Huang recognized that the limitations of general-purpose computer chips and the rise of computer gaming would be likely to increase demand for specialized graphics processors. During the late '90s and 2000s, the company he cofounded found huge success making high-end graphics chips for gamers. More recently Huang and Nvidia have ridden a different technology wave, supplying the hardware used to train and run the deep-learning algorithms that have been key to a recent renaissance in artificial intelligence.

Researchers at MIT's Computer Science and Artificial Intelligence Laboratory have developed a new algorithm that can accurately measure the heart rates of people depicted in ordinary digital video by analyzing imperceptibly small head movements that accompany the rush of blood caused by the heart's contractions. In tests, the algorithm gave pulse measurements that were consistently within a few beats per minute of those produced by electrocardiograms (EKGs). It was also able to provide useful estimates of the time intervals between beats, a measurement that can be used to identify patients at risk for cardiac events. Guha Balakrishnan, a graduate student in MIT's Department of Electrical Engineering and Computer Science, and his two advisors -- John Guttag, the Dugald C. Jackson Professor of Electrical Engineering and Computer Science and director of MIT's Data-Driven Medicine Group, and professor of computer science and engineering Fredo Durand -- describe the new algorithm in a paper appearing this summer at the Institute of Electrical and Electronics Engineers' Computer Vision and Pattern Recognition conference. A video-based pulse-measurement system could be useful for monitoring newborns or the elderly, whose sensitive skin could be damaged by frequent attachment and removal of EKG leads.

Machine learning has proven it can beat traditional human techniques in healthcare for some time now, yet it remains limited in use in the healthcare industry. But that may be about to change. "Machine learning is changing everything -- except maybe healthcare," MIT professor John Guttag said here at the Big Data and Healthcare Analytics Forum on Oct. 24. While machine learning drives products and services such as Google Maps, many websites' tracking of shopping habits and presenting options, banking, credit card companies and others, healthcare providers have done much less with the existing technologies. "There's lots of talk, but very little action, very little progress in healthcare," Guttag said.

There are two main approaches to machine learning – supervised and unsupervised – and each has specific applications in the context of healthcare. And even though their impact has not yet sent shockwaves through the industry, the potential of each is enormous, according to John Guttag, head of the Data Driven Inference Group at MIT's Computer Science and Artificial Intelligence Laboratory. At its basic level, machine learning involves looking at data, and from that data finding information that is not readily visible. Example: Applying machine learning to data about patients infected with Zika or another virus and using what we can learn about what happens to those people to inform care decisions regarding the best ways to treat people who get infected in the future. "Typically we use machine learning to build inference tools, where we find patterns in existing data that allow us – when presented with new data – to infer something interesting about that data," said Guttag.

There are two main approaches to machine learning – supervised and unsupervised – and each has specific applications in the context of healthcare. And even though their impact has not yet sent shockwaves through the industry, the potential of each is enormous, according to John Guttag, head of the Data Driven Inference Group at MIT's Computer Science and Artificial Intelligence Laboratory. At its basic level, machine learning involves looking at data, and from that data finding information that is not readily visible. Example: Applying machine learning to data about patients infected with Zika or another virus and using what we can learn about what happens to those people to inform care decisions regarding the best ways to treat people who get infected in the future. "Typically we use machine learning to build inference tools, where we find patterns in existing data that allow us – when presented with new data – to infer something interesting about that data," said Guttag.