Machine learning predicts mechanical properties of porous materials -- Department of Chemical Engineering and Biotechnology


Researchers from our Adsorption and Advanced Materials Group have used machine learning techniques to accurately predict the mechanical properties of metal organic frameworks (MOFs), materials which could be used to extract water from the air in the desert, store dangerous gases or power hydrogen-based cars. The researchers used their algorithm to predict the properties of more than 3000 existing MOFs, as well as MOFs which are yet to be synthesised in the laboratory. The results, published in the inaugural edition of the Cell Press journal Matter, could be used to significantly speed up the way materials are characterised and designed at the molecular scale. MOFs are self-assembling 3D compounds made of metallic and organic atoms connected together. Like plastics, they are highly versatile, and can be customised into millions of different combinations.

Algorithm accurately predicts mechanical properties of existing and theoretical MOFs


A machine learning algorithm that can predict the mechanical properties of metal–organic frameworks (MOFs) offers a way to overcome these highly varied and versatile materials' achilles heel – their instability.1 The team behind this work hope that this computational tool will speed up acceptance of these materials by industry. MOFs are a type of crystalline coordination polymers that form porous structures by combining metal clusters and organic ligands. 'Their "building block" nature allows chemists to easily tune their syntheses to tailor the pore size and surface chemistry for a specific application,' explains David Fairén-Jiménez at the University of Cambridge, UK. 'However, if you wish to use MOFs in real life, you need to shape them into pellets, and this densification may destroy their porosity, thus their functionality.'

Interpretable multiclass classification by MDL-based rule lists Artificial Intelligence

Interpretable classifiers have recently witnessed an increase in attention from the data mining community because they are inherently easier to understand and explain than their more complex counterparts. Examples of interpretable classification models include decision trees, rule sets, and rule lists. Learning such models often involves optimizing hyperparameters, which typically requires substantial amounts of data and may result in relatively large models. In this paper, we consider the problem of learning compact yet accurate probabilistic rule lists for multiclass classification. Specifically, we propose a novel formalization based on probabilistic rule lists and the minimum description length (MDL) principle. This results in virtually parameter-free model selection that naturally allows to trade-off model complexity with goodness of fit, by which overfitting and the need for hyperparameter tuning are effectively avoided. Finally, we introduce the Classy algorithm, which greedily finds rule lists according to the proposed criterion. We empirically demonstrate that Classy selects small probabilistic rule lists that outperform state-of-the-art classifiers when it comes to the combination of predictive performance and interpretability. We show that Classy is insensitive to its only parameter, i.e., the candidate set, and that compression on the training set correlates with classification performance, validating our MDL-based selection criterion.

What would it take for renewably powered electrosynthesis to displace petrochemical processes?


Plants that grow in the ground make all their carbon-based infrastructure from carbon dioxide (CO2). By contrast, plants built by chemists use petroleum and natural gas as their carbon feedstock. In a review, De Luna et al. explore the prospective challenges and opportunities for manufacturing commodity chemicals such as ethylene and alcohols by direct electrochemical reduction of CO2. They estimate that production costs would be competitive with fossil technologies if renewable electricity costs drop below 4 cents per kilowatt-hour and electrical-to-chemical conversion efficiencies reach 60%. As the world continues to transition toward carbon emissions–free energy technologies, there remains a need to also reduce the carbon emissions of the chemical production industry. Today many of the world's chemicals are produced from fossil fuel–derived feedstocks. Electrochemical conversion of carbon dioxide (CO2) into chemical feedstocks offers a way to turn waste emissions into valuable products, closing the carbon loop. When coupled to renewable sources of electricity, these products can be made with a net negative carbon emissions footprint, helping to sequester CO2 into usable goods. Research and development into electrocatalytic materials for CO2 reduction has intensified in recent years, with advances in selectivity, efficiency, and reaction rate progressing toward practical implementation. A variety of chemical products can be made from CO2, such as alcohols, oxygenates, synthesis gas (syngas), and olefins--staples in the global chemical industry. Because these products are produced at substantial scale, a switch to renewably powered production could result in a substantial carbon emissions reduction impact. The advancement of electrochemical technology to convert electrons generated from renewable power into stable chemical form also represents one avenue to long-term (e.g., seasonal) storage of energy. The science of electrocatalytic CO2 reduction continues to progress, with priority given to the need to pinpoint more accurately the targets for practical application, the economics of chemical products, and barriers to market entry.

NTT to launch trial of farming support service with drones and AI tech in Fukushima

The Japan Times

Nippon Telegraph and Telephone Corp. (NTT) said Thursday it will launch a trial for a farming support service using drones and artificial intelligence technology, with a goal of commercializing the service in Japan and other Asian countries. The new system, which connects drones with GPS satellites, is anticipated to help the farm industry in the nation amid a serious labor shortage. NTT aims to raise crop output by up to 30 percent through the new service. The telecommunications giant will conduct the trial service on 8 hectares of a rice field in Fukushima Prefecture from later this month to March 2021. It aims to launch the service on a commercial basis in Japan in two years.

The Why's and how's of Machine Learning


The knowledge is the output of learning through the inseparable combination of theory and practice. It's what remains in one's experience from all the data which got shaped into what we call information. This process can be noticed throughout the different stages of our lives and it's never limited to the academic journey. What I'm aiming to express is that machine learning is nothing but a human logic tailored for more complex problems that surely require more computational capabilities. The last quote represents the nature knowledge acquiring process which, as you may notice, is similar to CRISP-DM Methodology which I detailed in a previous article and which is essential to succeed in your data mining project.

EmbraceNet: A robust deep learning architecture for multimodal classification Machine Learning

Classification using multimodal data arises in many machine learning applications. It is crucial not only to model cross-modal relationship effectively but also to ensure robustness against loss of part of data or modalities. In this paper, we propose a novel deep learning-based multimodal fusion architecture for classification tasks, which guarantees compatibility with any kind of learning models, deals with cross-modal information carefully, and prevents performance degradation due to partial absence of data. We employ two datasets for multimodal classification tasks, build models based on our architecture and other state-of-the-art models, and analyze their performance on various situations. The results show that our architecture outperforms the other multimodal fusion architectures when some parts of data are not available.

Lunar lander crashes into moon. Did an archive of earth books survive?


A team of experts is trying to ascertain whether a special collection of digitized books sent into space on a disc archive is intact after an unmanned moon landing went off course last week. Israel company SpaceIL was attempting to become the first private company to successfully touch down on the moon, but its Beresheet spacecraft lost power on descent and crashed into the surface. A finalist for Google's Lunar Xprize, SpaceIL helped Israel become just the fourth country to put an object on the moon, after the U.S., the former Soviet Union, and China. The mission carried a price tag of $100M, relative pennies compared to previous lander attempts. Deep space mine: Luxembourg's robot experts have their sights on asteroid mining The tiny European state of Luxembourg has bold ambitions when it comes to its place in future asteroid-mining operations.

Material Segmentation of Multi-View Satellite Imagery Artificial Intelligence

Material recognition methods use image context and local cues for pixel-wise classification. In many cases only a single image is available to make a material prediction. Image sequences, routinely acquired in applications such as mutliview stereo, can provide a sampling of the underlying reflectance functions that reveal pixel-level material attributes. We investigate multi-view material segmentation using two datasets generated for building material segmentation and scene material segmentation from the SpaceNet Challenge satellite image dataset. In this paper, we explore the impact of multi-angle reflectance information by introducing the \textit{reflectance residual encoding}, which captures both the multi-angle and multispectral information present in our datasets. The residuals are computed by differencing the sparse-sampled reflectance function with a dictionary of pre-defined dense-sampled reflectance functions. Our proposed reflectance residual features improves material segmentation performance when integrated into pixel-wise and semantic segmentation architectures. At test time, predictions from individual segmentations are combined through softmax fusion and refined by building segment voting. We demonstrate robust and accurate pixelwise segmentation results using the proposed material segmentation pipeline.

How do we design "good" AI?


"AI is freeing up time, creativity, and human capital, essentially letting people work more like humans and less like robots." It's true - but how do we build "good" AI? In 2015, Professor Liu Jing and his team in Beijing discovered how to make programmable liquid metal invoking images from The Terminator. "The [metal] device is made from a drop of metal alloy consisting mostly of gallium, which is a liquid at just under 30 degrees Celsius. Last year they discovered that an applied electrical current causes the gallium alloy to drastically alter its shape. Changing the voltage applied to the metal allowed it to'shape-shift' into different formations. When the current was switched off, the metal returned to its original drop shape."