Phys.org December 23, 2024 Koopmans spectral functionals enable the prediction of spectral properties with state-of-the-art accuracy which relies on capturing the effects of electronic screening through scalar, orbital-dependent parameters. The parameters must be computed for every calculation, making Koopmans spectral functionals more expensive. Researchers in Switzerland developed a machine-learning model that can predict these screening parameters directly from orbital densities calculated at the density-functional theory (DFT) level. In two cases they showed that using the screening parameters predicted by this mode led to orbital energies that differ by less than 20 meV on average. This approach substantially reduced the run time […]
Tag Archives: Machine learning
Machine learning unlocks secrets to advanced alloys
MIT News July 18, 2024 The tendency of certain chemical motifs to be more common than others is known as chemical short-range order (SRO), and it has received substantial consideration in alloys with multiple chemical elements present in large concentrations due to their extreme configurational complexity. SRO renders solid solutions “slightly less random than completely random,” but not easily quantifiable due to the sheer number of possible chemical motifs and their subtle spatial distribution on the lattice. Researchers at MIT presented a multiscale method to predict and quantify the SRO state of an alloy with atomic resolution, incorporating machine learning […]
Training machines to learn more like humans do
MIT News May 9, 2023 Unlike humans, computer vision models don’t typically exhibit perceptual straightness, so they learn to represent visual information in a highly unpredictable way. But if machine-learning models had this ability, it might enable them to better estimate how objects or people will move. Researchers at MIT explored the relationship between network architecture, differing types of robustness, biologically-inspired filtering mechanisms, and representational straightness in response to time-varying input; they identified strengths and limitations of straightness as a useful way of evaluating neural network representations. They found that adversarial training leads to straighter representations in both convolutional neural […]
A method for designing neural networks optimally suited for certain tasks
MIT News March 30, 2023 While neural networks are used for classification tasks across domains, a long-standing open problem in machine learning is determining whether neural networks trained using standard procedures are consistent for classification, i.e., whether such models minimize the probability of misclassification for arbitrary data distributions. A team of researchers in the US (MIT, UC San Diego) identified, constructed and analyzed infinitely wide networks that were also infinitely deep. Using the recent connection between infinitely wide neural networks and neural tangent kernels, they provided explicit activation functions that could be used to construct networks that achieve consistency. They […]
Using machine learning to infer rules for designing complex mechanical metamaterials
Phys.org November 23, 2022 Combinatorial problems arising in puzzles, origami, and (meta) material design have rare sets of solutions, which define complex and sharply delineated boundaries in configuration space. The boundaries are difficult to capture with conventional statistical and numerical methods. Researchers in the Netherlands have shown that convolutional neural networks can learn to recognize these boundaries for combinatorial mechanical metamaterials, down to the finest detail, despite using heavily undersampled training sets, and can successfully generalize. According to the researchers even if machine learning is typically a “black box” approach, it can still be very valuable for exploring the design […]
Study shows widely used machine learning methods don’t work as claimed
EurekAlert March 16, 2020 A widely used algorithmic technique for modeling complex networks is to construct a low-dimensional Euclidean embedding of the vertices of the network, where proximity of vertices is interpreted as the likelihood of an edge. A team of researchers in the US (UC Santa Cruz, Google, Stanford University) focused on low degree and large clustering coefficients, which have been widely established to be empirically true for real-world networks. They demonstrated mathematically that significant structural aspects of complex networks are lost in this embedding process and confirmed this result empirically by testing various embedding techniques on different kinds […]
Developing a digital twin
Eurekalert December 5, 2019 A team of researchers in the US ( UT Austin, MIT, industry) is working on Dynamic Data-Driven Application Systems (DDDAS), a project sponsored by the US Air Force, to develop a predictive digital twin for a custom-built UAV. The twin represents each component of the UAV, as well as its integrated whole. The twin also ingests on-board sensor data from the vehicle and integrates that information with the model to create real-time predictions of the health of the vehicle. They paired computational modeling is paired with machine learning to produce predictions that are reliable, and explainable. […]
Novel math could bring machine learning to the next level
EurekAlert September 2, 2019 Before the neural network can begin to perform facial recognition, it is typically necessary to present it with thousands of faces. Much of these machines have been increasingly successful at pattern recognition but what goes on inside them as they learn their task is unknown. An international team of researchers (Portugal, Italy) has shown that artificial vision machines can learn to recognize complex images faster by using topological data analysis which was developed 25 years ago. Current neural networks are not good at topology. The team mathematically describe how to enforce certain symmetries, and this provides […]
Here are 10 ways AI could help fight climate change
MIT Technology Review June 20, 2019 A team of researchers in the US led by the University of Pennsylvania has developed a road map suggesting how machine learning can help save our planet and humanity from imminent peril. 10 of the “high leverage” recommendations from the report are: Improve predictions of how much electricity we need, Discover new materials, Optimize how freight is routed, Lower barriers to electric-vehicle adoption, Help make buildings more efficient, Create better estimates of how much energy we are consuming, Optimize supply chains, Make precision agriculture possible at scale, Improve deforestation tracking, Nudge consumers to change […]
Best of arXiv.org for AI, Machine Learning, and Deep Learning – January 2019
Inside Big Data February 20, 2019 The articles are academic research papers, typically geared toward graduate students, post docs, and seasoned professionals. Articles are listed in no particular with a brief overview – Hard-Exploration Problems , Deep Neural Network Approximation for Custom Hardware: Where We’ve Been, Where We’re Going , Generating Textual Adversarial Examples for Deep Learning Models: A Survey , Revisiting Self-Supervised Visual Representation Learning , Self-Driving Cars: A Survey read more.