Page 156«..1020..155156157158..170180..»

Bruce Gordon: Idealism, Quantum Mechanics, and the Fundamentality of Mind – Walter Bradley Center for Natural and Artificial Intelligence

Michael Egnor May 23, 2024 1 Michael Egnor May 23, 2024 1

Does quantum mechanics, properly understood, point to the fundamentality of mind in the universe? In this episode, Michael Egnor concludes a conversation with philosopher of physics Bruce Gordon about the relationship between idealism and quantum mechanics. Gordon argues that quantum mechanics points to mind as the fundamental unit of the universe, as it is irreducibly probabilistic and exhibits non-local phenomena. He dismisses interpretations such as Bohmian mechanics and Everett’s many-worlds hypothesis as flawed and suggests that a theistic metaphysic can provide a coherent explanation for the probabilistic nature of quantum mechanics. Gordon also discusses the compatibility of free will with determinism and the nature of God’s freedom. Finally, he challenges the notion of the existence of electrons as physical entities and suggests that they do not exist in our experiential reality. Tune in for the conclusion to a fascinating conversation!

Read the original here:

Bruce Gordon: Idealism, Quantum Mechanics, and the Fundamentality of Mind - Walter Bradley Center for Natural and Artificial Intelligence

Read More..

Quantum Leap: Atom Interference and a Breakthrough in Boson Sampling – SciTechDaily

La Nia Sea Surface Height, December 1, 2021

This coupling of the atmosphere and ocean alters atmospheric circulation and jet streams in ways that intensify rainfall in some regions and bring drought to others.

For the second year in a row, the cooler sister to El Nio showed up at the winter party in the Eastern Pacific. La Nia is expected to stick around until at least spring 2022 in the Northern Hemisphere.

Part of the El Nio-Southern Oscillation cycle, La Nia appears when energized easterly trade winds intensify the upwelling of cooler water from the depths of the eastern tropical Pacific, causing a large-scale cooling of the eastern and central Pacific ocean surface near the Equator. These stronger than usual trade winds also push the warm equatorial surface waters westward toward Asia and Australia. This dramatic cooling of the oceans surface layers then affects the atmosphere by modifying the moisture content across the Pacific. This La Nia coupling of the atmosphere and ocean alters global atmospheric circulation and can cause shifts in the path of mid-latitude jet streams in ways that intensify rainfall in some regions and bring drought to others.

In the western Pacific, rainfall can increase dramatically over Indonesia and Australia during La Nia. Clouds and rainfall become more sporadic over the central and eastern Pacific Ocean, which can lead to dry conditions in Brazil, Argentina, and other parts of South America and wetter conditions over Central America. In North America, cooler and stormier conditions often set in across the Pacific Northwest, while weather typically becomes warmer and drier across the southern United States and northern Mexico. (These and other trends are reflected in the map lower in this story.)

The image above shows conditions across the central and eastern Pacific Ocean as observed from November 26 to December 5, 2021, by the Sentinel-6 Michael Freilich satellite and analyzed by scientists at NASAs Jet Propulsion Laboratory (JPL). The globe depicts sea surface height anomalies. Shades of blue indicate sea levels that were lower than average; normal sea-level conditions appear white; and reds indicate areas where the ocean stood higher than normal. The expansion and contraction of the ocean surface is a good proxy for temperatures because warmer water expands to fill more volume, while cooler water contracts.

This moderate strength La Nia can be seen in the Sentinel-6 data as an area of lower-than-normal sea level along and below the Equator in the central and eastern Pacific, said Josh Willis, a climate scientist and oceanographer at JPL. He noted that the deep trough (blue) above the Equator is not the La Nia water mass; it is a shift in the North Equatorial Counter Current, which tends to strengthen during La Nia events.

December 1, 2021

This La Nia probably means bad news for the American Southwest, which should see lower than normal rainfall this winter, Willis said. This La Nia may not be a whopper, but its still an unwelcome sign for an area already deep into a drought.

The La Nia event that started in late 2020 fits into a larger climate pattern that has been going on for nearly two decadesa cool (negative) phase of the Pacific Decadal Oscillation (PDO). During most of the 1980s and 1990s, the Pacific was locked in a PDO warm phase, which coincided with several strong El Nio events. But since 1999, a cool phase has dominated. The long-term drought in the American Southwest coincides with this trend, Willis noted.

In a report released on December 9, 2021, the NOAA Climate Prediction Center noted that sea surface temperatures in November in the eastern tropical Pacific ranged from 0.7 to 1.2 degrees Celsius (1.26 to 2.16 degrees Fahrenheit) below the long-term average and 0.9C (1.62F) below average in the Nio 3.4 region of the tropical Pacific (from 170 to 120 West longitude). Forecasters predicted La Nia conditions would persist through Northern Hemisphere winter, with a 60 percent chance that the ocean would transition back to neutral conditions during the April through June period.

This La Nia is the first to be observed by Sentinel-6 Michael Freilich, which was launched in November 2020. The new satellite is giving us a great picture of this La Nia, Willis said. With the public release of the missions climate-quality data, we are now in a position where Sentinel-6 Michael Freilich can soon take over the climate record of sea level rise, which goes all the way back to the early 1990s.

Engineers and scientists have spent the past year calibrating and analyzing data from the new satellite against the existing Jason-3 mission. The team is ensuring that the new, more advanced data correlate properly with long-term records. New, high-resolution Sentinel-6 Michael Freilich data sets were released at the end of November 2020.

NASA Earth Observatory images by Joshua Stevens, using modified Copernicus Sentinel data (2021) processed by the European Space Agency courtesy of Josh Willis/NASA/JPL-Caltech, and information adapted from the Famine Early Warning Systems Network.

Here is the original post:

Quantum Leap: Atom Interference and a Breakthrough in Boson Sampling - SciTechDaily

Read More..

Why are computational chemists making up their data? – Chemistry World

For scientists, faking or making up data has obvious connotations and, thanks to some high-profile cases of scientific misconduct, theyre generally not positive ones. Chemists may, for example, be aware of a 2022 case in which a respected journal retracted two papers by a Japanese chemistry group that were found to contain manipulated or fabricated data. Or the case of Beng Sezen, the Columbia University chemist who, during the 2000s, falsified, fabricated and plagiarised data to get her work on chemical bonding published including fixing her NMR spectra with correcting fluid.

Synthetic data, unlike dishonestly made-up data, is created in a systematic way for legitimate reasons, however, usually by a machine and for a variety of reasons. Synthetic data is familiar to machine learning experts, and increasingly to computational chemists, but relatively unknown to the wider chemistry community, as Keith Butler, a materials researcher who works with machine learning methods at University College London in the UK, acknowledges.

I dont think very many people at all, in chemistry, would refer to synthetic data, he says, adding that its probably a confusing term for chemists not just because of the apparent associations with scientific misconduct, but because synthetic in chemistry has another important meaning, relating to the way chemical compounds are made. But why are chemists making up their data, how is it any different to the simulations theyve been making for decades and what are the implications?

In areas like health and finance, synthetic data is often used to replace real data from real people due to privacy concerns, or to deal with issues of imbalance, such as when people from certain ethnic groups arent well-represented. While researchers might like to use peoples personal medical records, for example, to inform their understanding of a new disease, that data is difficult to access in the quantities or levels of detail that would be most useful. Here, the creation of synthetic data, which mirrors the statistical features of the real-world data, offers a solution. Youre not taking a whole dataset and just masking it, explains Benjamin Jacobsen, a sociologist at the University of York in the UK, whose work focuses partly on the use of synthetic data. The promise of synthetic data is that you can train a model to understand the overall distribution of the particular dataset. In this way, synthetic data draws from real-world sources, but cant be traced back to real individuals. In the chemical sciences, though, synthetic data relates more to the behaviour of molecules than people and so as Butler notes, its used for a very different reason.

Synthetic data is generated by algorithms and for algorithms

In some ways, legitimately made-up data is nothing very new in chemistry. Areas such as materials and drug discovery, for example, have long used whats referred to as simulated or calculated data to expand the chemical space for exploration the data might describe predicted properties of new materials or potential drug compounds. Whats different now is that made-up data, whether its considered synthetic, simulated or calculated, is being used in combination with machine learning models. These AI models are algorithms capable of making sense of vast quantities of data, which could be from real experiments or made up (or a combination of both). They learn patterns in the data and use them to make classifications and predictions to provide valuable insights to humans whether or not its clear how the machines have delivered them.

Synthetic data is not just an input for AI models its an output from them too. Jacobsen, in fact, previously defined synthetic data as data generated by algorithms and for algorithms, although chemists may not be concerned with having such a strict definition. Some of the techniques commonly used to create it are related to those used in making deepfakes. In the same way that deepfakers might ask their machines to generate realistic-looking faces and speech, chemists might prompt theirs to generate realistic-looking chemical structures.

Another option for generating synthetic data is large language modelling the basis of generative AI tools like ChatGPT. This is an approach Butlers team recently used to build an app that can produce a hypothetical crystal structure for a compound based on a chemical formula, which is typed in as a prompt (like a question in a chat). More advanced versions of such tools, which would know more of the rules of chemistry, could prove invaluable in the search for new materials. If you could prompt [it] by saying produce me a plausible chemical structure that absorbs light well and is made only from Earth-abundant elements, thats actually interesting, says Butler. The problem being that you want to make sure that its reasonable and viable so that you dont start telling people things that are impossible to make.

Applications for synthetic data in the chemical sciences abound. However, for those who are less well-acquainted with machine learning, it can be hard to get a grip on what synthetic data is and how it can be so useful to chemists when its not actually real. Here, an example may help.

Most students of chemistry learn, fairly early on, to sketch out a chemical structure by hand, but transferring that structure from a piece of paper to a computer could be a tedious task. It would be useful if there was a quick way of doing it say, by taking a picture of the sketch on your phone. At Stanford University in California, US, Todd Martinez and his team set about tackling this problem, their idea being to teach a machine learning model how to recognise hand-drawn chemical structures so that they could be quickly converted into virtual versions. However, to do this using real data, they would have needed to train their data-hungry model with a vast dataset of hand-drawn structures. As Martinez notes, even someone who can draw really fast is only going to be able to churn out a handful a minute. We did try, he recalls. We got 30 people together and spent hours just doing this, but we were only able to get 1000 structures or something. You need this data in the hundreds of thousands.

So, instead, they developed a process for roughing-up half a million clean, artificially generated chemical structures made with well-known software called RDKit and sticking them onto backgrounds to simulate photographs of hand-drawn structures. Having ingested this synthetic data, combined with a much smaller sample of real hand-drawn structures, their machine learning approach was able to correctly identify hand-drawn hydrocarbon structures 70% of the time compared to never when trained just with their limited hand-drawn data (and only 56% of the time with the clean RDKit structures). More recently, they developed an app that turns hand-drawn structures directly into 3D visualisations of molecules on a mobile device.

Had it been fed half a million genuine, hand-drawn structures, Martinezs model might have performed even better, but the real data just wasnt available. So in this case, using synthetic data was a solution to the problem of data sparsity a problem described as one of the main barriers to the adoption of AI in the chemical sciences. As Martinez puts it, There are lots of problems in chemistry actually, most problems, I would say where there is insufficient data to really apply machine learning methods the way that practitioners would like to. The cost savings to be made by using synthetic data, he adds, could be much larger than in his molecular recognition example, because elsewhere this made-up data wouldnt just be replacing data that it takes someone a few seconds to draw, but data from expensive, real-life experiments.

Theres certainly no shortage of such experimental data at the Rutherford Appleton Laboratory in Oxfordshire. Here, petabytes and petabytes of data are produced, Butler says, by real-life experiments in which materials are bombarded with subatomic particles in order to probe their structure. But Butlers collaborators at the facility face a different problem: a massive surplus of unlabelled data data that is effectively meaningless to many machine learning models. To understand why, think back to the previous example, where each hand-drawn chemical structure would need a label attached to teach the machine learning model how to recognise it. Without labels, its harder for the AI to learn anything. The problem is that its really expensive to label that kind of experimental data, Butler says. Synthetic data, though, can be generated ready-labelled and then used as training data to help machine learning models learn to interpret the masses of real data produced by the laboratorys neutron-scattering experiments a line of thinking Butler and co-workers explored in a 2021 study.

In the study, they trained their models with thousands of synthetic spectra images in the same format as those they got from real neutron scattering experiments, but that were created via theoretical calculations. What they realised, though, was that when it came to interpreting real spectra, the models trained on simulated data just werent very good, because they werent used to all the imperfections that exist in real-world experimental spectra. Like the fake, hand-drawn chemical structures made by Martinezs team, they needed roughing up. As a solution, Butlers team came up with a way to add experimental artefacts including noise to the clean synthetic data. He describes it as akin to a filter you might apply to selfie, except instead of giving a photo of your face the style of a Van Gogh painting, it gives a perfect, simulated spectrum the style of a messier one from a real experiment. This restyling of synthetic data could be useful more broadly than in neutron scattering experiments, according to Butler.

Another area in which synthetic data could have an important impact is in combination with AI approaches in drug discovery. Though, as in other fields, the terminology could be a little off-putting. People are a bit shy to accept synthetic data [perhaps because] it sounds like its lower quality than real data, says Ulrich Zachariae, a drugs researcher at the University of Dundee in the UK. Zachariaes recent work has focused on searching for new compounds to target antibiotic-resistant Gram-negative bacteria like Pseudomonas aeruginosa, which causes life-threatening infections in hospital. One of the issues slowing down the search is that these bugs outer shells are virtually impenetrable, and while machine learning models could help make useful predictions about which compounds might work, theres a lack of data on bacterial permeability to feed the models.

That dataset provided us with enough input data that we could understand what was going on

To start tackling the permeability problem, Zachariaes team first constructed a model with what data they had data that came from existing antibiotics and used it predict whether other compounds would be good permeators, or not. This worked well, but didnt explain anything about why one compound was better than another. The researchers then wanted to probe the effects of small differences in chemical structure on permeability , but this required lots more data. So, to generate more, they used their own machine learning model to predict the properties of hundreds of thousands of (real) compounds for which there was no experimental data on permeability creating a huge new synthetic dataset. That dataset provided us with enough input data for [the additional analysis] that we could understand what was going on and why these compounds were good permeators or not, Zachariae explains. They were then able to suggest chemical features, including amine, thiophene and halide groups, that medicinal chemists should look out for in their hunt for new Gram-negative antibiotics.

For Martinez, understanding why machine learning models make the predictions they do is another key motivation for using synthetic data. The internal workings of AI models can be difficult to unravel theyre widely referred to as black boxes but Martinez says he thinks of synthetic data as a tool to sharpen a particular model that is producing that data, or to understand its essence in a simpler form. I think you can see examples of people groping towards this, but I dont know that its clearly stated, he muses. Martinezs interests lie mainly in quantum chemistry, where more traditional computational models are used to solve theoretical problems. Addressing the same problems with machine learning may be a way to get to solutions faster while also with the help of synthetic data getting to the heart of what the AI models have learned. In this way, chemists may be able to improve their more traditional models.

But what are the risks of using data that isnt real? Its hard to answer this question at this point. In other fields, the risks often relate to people whose data was absorbed to generate synthetic data, or who are affected by the decisions it is used to make. As Jacobsen notes, the risks are going to vary depending on the application area. Chemists have to delineate exactly How do we frame what risk is in this context? he says.

In the drug discovery space, Zachariae wonders if there is any more risk associated with artificially generated data than with simulated data used in the past. From a purely scientific perspective, I dont see how its any different from previous cycles of predictions, where we just used physical models, he says, adding that any hit molecule identified using AI and synthetic data would still have to go through rigorous safety testing.

The risk is that we lose credibility if our predictions dont match up with reality

Martinez, though, sees a potential problem in areas of theoretical chemistry where there is no real or experimental data on which to train machine learning models because that data can only be arrived at by computational means. Here, synthetic data may effectively be the norm, although the magic words often arent mentioned, he says, because chemists arent familiar with them. In quantum chemistry, for example, a molecules geometry can be used to compute its energy and now machine learning models are trained, based on existing theory, to take in geometries and spit out energies, just in faster and cheaper ways. However, since traditional methods are more accurate for smaller molecules than bigger ones and theres no way of experimentally checking the results, machine learning algorithms trained to spit out energies for big molecules could be doing a poor job which could be a concern if millions of data points are being generated. The interesting point about synthetic data in this context is that these issues do not seem to always be at the forefront of the communitys thinking, Martinez says. This seems to be because the synthetic nature of the data implies tight control over the training dataset and this can give a false confidence in data correctness and coverage.

In materials discovery, similar concerns surround the use of machine learning to predict stable chemical structures as Google DeepMind researchers have done in chemical spaces where the existing theory was not necessarily that accurate. The risk, says Butler, is that we lose credibility (and funding) if the properties of predicted materials dont match up with reality. So, while making up data may mean something different these days, its worth remembering that there could still be a lot at stake if its not done well.

Hayley Bennett is a science writer based in Bristol, UK

View original post here:
Why are computational chemists making up their data? - Chemistry World

Read More..

Inexpensive microplastic monitoring through porous materials and machine learning – EurekAlert

image:

Inexpensive microplastic monitoring through porous materials and machine learning

Credit: Reiko Matsushita

Optical analysis and machine learning techniques can now readily detect microplastics in marine and freshwater environments using inexpensive porous metal substrates. Details of the method, developed by researchers at Nagoya University with collaborators at the National Institute for Materials Sciences in Japan and others, are published in the journal Nature Communications.

Detecting and identifying microplastics in water samples is essential for environmental monitoring but is challenging due in part to the structural similarity of microplastics with natural organic compounds derived from biofilms, algae, and decaying organic matter. Existing detection methods generally require complex separation techniques that are time-consuming and costly.

Our new method can simultaneously separate and measure the abundance of six key types of microplastics - polystyrene, polyethylene, polymethylmethacrylate, polytetrafluoroethylene, nylon and polyethylene terephthalate, says Dr. Olga Guselnikova of the National Institute for Materials Science (NIMS).

The system uses a porous metal foam to capture microplastics from solution and detect them optically using a process called surface-enhanced Raman spectroscopy (SERS). The SERS data obtained is highly complex, explains Dr. Joel Henzie of NIMS, but it contains discernible patterns that can be interpreted using modern machine learning techniques.

To analyse the data, the team created a neural network computer algorithm called SpecATNet. This algorithm learns how to interpret the patterns in the optical measurements to identify the target microplastics more quickly and with higher accuracy than traditional methods.

Our procedure holds immense potential for monitoring microplastics in samples obtained directly from the environment, with no pretreatment required, while being unaffected by possible contaminants that could interfere with other methods, says Professor Yusuke Yamauchi of Nagoya University.

The researchers hope their innovation will greatly assist society in evaluating the significance of microplastic pollution on public health and the health of all organisms in marine and freshwater environments. By creating inexpensive microplastic sensors and open-source algorithms to interpret data, they hope to enable the rapid detection of microplastics, even in resource-limited labs.

Currently, materials required for the new system bring cost savings of 90 to 95% compared to commercially available alternatives. The group plans to drive the cost of these sensors down even further and make the methods simple to replicate without the need for expensive facilities. In addition, the researchers hope to expand the capability of the SpecATNet neural network to detect a broader range of microplastics and even accept different kinds of spectroscopic data in addition to SERS data.

Nature Communications

28-May-2024

Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.

Go here to read the rest:
Inexpensive microplastic monitoring through porous materials and machine learning - EurekAlert

Read More..

Deep learning algorithm-enabled sediment characterization techniques to determination of water saturation for tight … – Nature.com

The aim of this research is to develop precise and dependable machine learning models for the prediction of SW (Water Saturation) using three DL and three SL techniques: LSTM, GRU, RNN, SVM, KNN and DT. These models were trained on an extensive dataset comprising various types of log data. The findings of our investigation illustrate the efficacy of data-driven machine learning models in SW prediction, underscoring their potential for a wide range of practical applications.

When evaluating and comparing algorithms, researchers must take into account several crucial factors. Accuracy and disparities in prediction are among the most significant considerations. To evaluate these factors, researchers can utilize various criteria, including Eqs.16. The Mean Percentage Error (MPE) calculates the average difference between predicted and actual values as a percentage, while the Absolute Mean Percentage Error (AMPE) measures the absolute difference between them. Additionally, the Standard Deviation (SD) determines the variability of data points around the mean. Moreover, the Mean Squared Error (MSE) and Root Mean Squared Error (RMSE) quantify the mean and root mean squared differences between predicted and actual values, respectively. Lastly, the R2 metric assesses the fraction of diversity in the reliant variable that can be accounted for by the autonomous variable.

$$text{MPE}=frac{{{sum }_{text{i}=1}^{text{n}}(frac{{SWE}_{(text{Meas}.)}-{SWE}_{(text{Pred}.)}}{{SWE}_{(text{Meas}.)}}text{x }100)}_{text{i}}}{text{n}}$$

(1)

$$text{AMPE}=frac{{sum }_{text{i}=1}^{text{n}}left|{(frac{{SWE}_{(text{Meas}.)}-{SWE}_{(text{Pred}.)}}{{SWE}_{(text{Meas}.)}}text{x }100)}_{text{i}}right|}{text{n}}$$

(2)

$$text{SD}=sqrt{frac{{sum }_{text{i}=1}^{text{n}}{({left(frac{1}{text{n}}sum_{text{i}=1}^{text{n}}left({{SWE}_{text{Meas}.}}_{text{i}}-{{SWE}_{text{Pred}.}}_{text{i}}right)right)}_{text{i}}-(frac{1}{text{n}}sum_{text{i}=1}^{text{n}}left({{SWE}_{text{Meas}.}}_{text{i}}-{{SWE}_{text{Pred}.}}_{text{i}}right))text{imean})}^{2}}{text{n}-1}}$$

(3)

$$text{MSE}=frac{1}{text{n}}sum_{text{i}=1}^{text{n}}{left({{SWE}_{text{Meas}.}}_{text{i}}-{{SWE}_{text{Pred}.}}_{text{i}}right)}^{2}$$

(4)

$$text{RMSE}=sqrt{frac{1}{text{n}}sum_{text{i}=1}^{text{n}}{left({{SWE}_{text{Meas}.}}_{text{i}}-{{SWE}_{text{Pred}.}}_{text{i}}right)}^{2}}$$

(5)

$${text{R}}^{2}=1-frac{sum_{text{i}=1}^{text{N}}{({{SWE}_{text{Pred}.}}_{text{i}}-{{SWE}_{text{Meas}.}}_{text{i}})}^{2}}{sum_{text{i}=1}^{text{N}}{({{SWE}_{text{Pred}.}}_{text{i}}-frac{{sum }_{text{I}=1}^{text{n}}{{SWE}_{text{Meas}.}}_{text{i}}}{text{n}})}^{2}}$$

(6)

In order to forecast SW, three DL and three SL techniques: LSTM, GRU, RNN, SVM, KNN and DT, were used in this study. Each algorithm underwent individual training and testing processes, followed by independent experiments. To ensure the accuracy of the predictions, the dataset was carefully divided into three subsets. The training subset accounted for 70% of the data records, while 30% was allocated for independent testing.

Choosing the most suitable algorithm for a specific task is a crucial undertaking within the realm of data analysis and machine learning. Therefore, this research aimed to assess and compare the performance of multiple LSTM, GRU, and RNN algorithms in predicting SW. The outcomes of these algorithms, utilizing the train data values, as well as the test, have been meticulously documented and presented in Table 2. By analyzing the results, researchers can gain insights into the effectiveness of each algorithm and make informed decisions about their implementation in practical applications.

The results from the test data are presented in Table 2, highlighting the excellent performance of the RMSE, MPE and AMPE metrics for the GRU algorithm, with values of 0.0198,0.1492 and 2.0320, respectively. Similarly, for the LSTM algorithm, the corresponding values are 0.0284,0.1388 and 3.1136, while for the RNN algorithm, they are 0.0399,0.0201 and 4.0613, respectively. For SVM, KNN and DT these metrics are includes: 0.0599,0.1664 and 6.1642; 0.7873, 0.0997 and 7.4575; 0.7289,0.1758 and 8.1936. The results show the GRU model has high accuracy than other algorithms.

The R2 parameter is a crucial statistical measure for evaluating and comparing different models. It assesses the adequacy of a model by quantifying the amount of variation in the outcome variable that can be clarified by the explanatory variables. In this study, Fig.5 illustrates cross plots for predicting SW values based on the train and test data, demonstrating significantly higher prediction accuracy compared to the other evaluated models. Additionally, Fig.5 confirms that the RGU model exhibits superior prediction accuracy compared to the LSTM and RNN models.

Cross plot for predicting SW using three DL algorithms such as RGU, LSTM and RNN for test data.

To assess the precision of the GRU model, the results presented in Table 2 and Fig.5 were carefully analyzed for the train and test data. The analysis revealed that the GRU algorithm achieved low errors for SW, with RMSE values of 0.0198 and R-square values of 0.9973. The R2 values provided serve as quantitative metrics assessing the predictive prowess of ML models. The R2, denoting the coefficient of determination, gauges the extent to which the variance in the dependent variable can be foreseen from the independent variable(s), essentially showcasing how well the model aligns with observed data points. The R2 values for the GRU, LSTM, RNN, SVM, KNN and DT models stand at 0.9973, 0.9725, 0.9701, 0.8050, 0.7873 and 0.7289 respectively, reflecting their respective accuracy and reliability in predicting SW levels. Figure5 shows the cross plot for predicting SW using three DL algorithms such as RGU, LSTM, and RNN for test data. The GRU model's notably high R2 of 0.9973 underscores its exceptional correlation between predicted and observed SW values, implying that nearly 99.73% of SW data variance can be elucidated by its predictions, showcasing its precision and reliability in SW prediction tasks. Comparatively, the LSTM and RNN models, with R2 values of 0.9725 and 0.9701 respectively, also exhibit strong predictive capabilities, albeit slightly lower than the GRU model. These findings underscore the GRU model's superiority in SW prediction, attributed to its adeptness in capturing intricate temporal dependencies within SW data, thereby yielding more accurate predictions.

Figure6 provides a visual representation of the calculation error for the test data, illustrating the error distribution for predicting SW using three DL algorithms (GRU, LSTM, and RNN). The plotted coordinates in the figure depict the error range for each algorithm. For the GRU algorithm, the error range is observed to be between0.0103 and 0.0727. This indicates that the predictions made by the GRU model for the test data exhibit a relatively small deviation from the actual SW values within this range. In contrast, the LSTM algorithm demonstrates a slightly wider error range, ranging from0.146 to 0.215. This suggests that the predictions generated by the LSTM model for the test data exhibit a somewhat higher variability and may deviate from the actual SW values within this broader range.

Error points for predicting SW using three DL and three SL algorithms such as GRU, LSTM, RNN, SVM, KNN and DT for test data.

Similarly, the RNN algorithm exhibits an error range between0.222 and 0.283. This indicates that the predictions made by the RNN model for the test data show a larger spread and have the potential to deviate more significantly from the actual SW values within this range. By visually comparing the error ranges for the three DL algorithms, it becomes apparent that the GRU algorithm achieves a narrower range and thus demonstrates better precision and accuracy in predicting SW for the test data. Conversely, the LSTM and RNN algorithms exhibit broader error ranges, indicating a higher degree of variability in their predictions for the same dataset. These findings further support the conclusion that the GRU algorithm outperforms the LSTM and RNN algorithms in terms of SW prediction accuracy, as it consistently produces predictions with smaller errors and tighter error bounds.

Figure7 presents an error histogram plot, depicting the prediction errors for SW using three DL and three SL algorithms such as GRU, LSTM, RNN, SVM, KNN and DT. Each histogram represents the distribution of prediction errors for each algorithm, displaying a normal distribution centered around zero with a relatively narrow spread and no noticeable positive or negative bias. This plot enables a comprehensive analysis of the algorithms' performance and aids in determining the best algorithm with a normal error distribution. Upon careful investigation, it becomes evident that the GRU algorithm exhibits a superior normal distribution of data compared to the other algorithms. The GRU algorithm's performance is characterized by a more accurate standard deviation and a narrower spread of prediction errors. This indicates that the GRU algorithm consistently produces more precise and reliable predictions for SW. By comparing the results presented in Table 2 and analyzing the error histogram plot in Fig.7, we can conclude that the performance accuracy of the algorithms can be ranked as follows: GRU>LSTM>RNN>SVM>KNN>DT.

Histogram plot for SW prediction using three DL and three SL algorithms such as GRU, LSTM, RNN, SVM, KNN and DT.

Figure8 illustrates the error rate of the three DL and three SL algorithms such as GRU, LSTM, RNN, SVM, KNN and DT as a function of iteration for SW prediction. The findings of this study indicate that the GRU and LSTM algorithms initially exhibit higher error values that progressively decrease over time. However, this pattern is not observed in the RNN algorithm. Upon analyzing the figure, it becomes evident that the LSTM algorithm achieves higher accuracy than the other algorithms at the beginning of the iteration. At the threshold of 10 iterations, the LSTM algorithm surpasses the GRU algorithm with a lower error value. However, in the subsequent iterations, specifically at iteration 31, the GRU algorithm outperforms the LSTM algorithm with superior performance accuracy. In contrast, the RNN algorithm shows a consistent decrease in performance accuracy from the start to the end of the iterations, without displaying significant fluctuations. When focusing on the zoomed-in portion of the figure, specifically repetitions 85100, the ongoing performance trends of these algorithms become more apparent. It is evident from the analysis that the GRU algorithm consistently outperforms the other algorithms in terms of performance accuracy. The LSTM algorithm follows, with a decrease in accuracy over the iterations. On the other hand, the RNN algorithm exhibits a declining performance accuracy without any notable changes or fluctuations. These findings emphasize the superiority of the GRU algorithm in terms of performance accuracy when compared to the LSTM and RNN algorithms. The GRU algorithm consistently maintains a higher level of accuracy throughout the iterations, while the LSTM and RNN algorithms experience fluctuations and decreasing accuracy over time.

Iteration plot for SW prediction using three DL and three SL algorithms such as GRU, LSTM, RNN, SVM, KNN and DT.

Pearson's coefficient (R) is a widely used method for assessing the relative importance of input-independent variables compared to output-dependent variables, such as SWL. The coefficient ranges between1 and+1 and represents the strength and direction of the correlation. A value of+1 indicates a strong positive correlation, -1 indicates a strong negative correlation, and a value close to 0 indicates no correlation. Equation7 illustrates the calculation of Pearson's correlation coefficient, which is a statistical measure of the linear relationship between two variables. It allows researchers to quantify the extent to which changes in one variable are associated with changes in another variable. By applying Pearson's coefficient, researchers can determine the level of influence that input-independent variables have on the output-dependent variable, SWL.

$$R=frac{sum_{i=1}^{n}({Z}_{i}-overline{Z })({Q}_{i}-overline{Q })}{sqrt{{sum }_{i=1}^{n}{({Z}_{i}-overline{Z })}^{2}}sqrt{{sum }_{i=1}^{n}{({Q}_{i}-overline{Q })}^{2}}}$$

(7)

A coefficient of+1 indicates a perfect positive correlation, suggesting that the input-independent variables have the greatest positive impact on the output-dependent variable. Conversely, a coefficient of1 represents a perfect negative correlation, indicating that the input-independent variables have the greatest absolute impact on the output-dependent variable. When the coefficient is close to 0, it suggests that there is no significant correlation between the variables, indicating that changes in the input-independent variables do not have a substantial effect on the output-dependent variable. Pearson's correlation coefficient is a valuable tool for assessing the relationship between variables and understanding their impact. It provides researchers with a quantitative measure to determine the relative importance of input-independent variables compared to the output-dependent variable, SWL.

By heat map shows Fig.9, a comparison of Pearson correlation coefficients can be made to gain insights into the relationship between input variables and SW. The results reveal several significant correlations between the variables. Negative correlations are observed with URAN and DEPTH, indicating an inverse relationship with SW. This suggests that higher values of URAN and DEPTH are associated with lower SW values. On the other hand, positive correlations are observed with CGR, DT, NPHI, POTA, THOR, and PEF. These variables show a direct relationship with SW, meaning that higher values of CGR, DT, NPHI, POTA, THOR, and PEF are associated with higher SW values. The comparison of Pearson correlation coefficients provides valuable insights into the relationship between input variables and SW.

Heat map plot for SW prediction using three DL and three SL algorithms such as GRU, LSTM, RNN, SVM, KNN and DT.

These findings can be utilized to develop predictive models of SW based on the input variables. By incorporating the correlations into the models, researchers can enhance their accuracy and reliability in predicting SW values. The expression of the relationships between the input variables and SW in the form of Eq.8 allows for quantitative analysis of the data. This equation provides a mathematical representation of the correlations, enabling researchers to quantitatively evaluate the impact of the input variables on SW.

$$SWE=propto left(text{CGR},text{ DT},text{ NPHI},text{ POTA},text{ THOR},text{ PEF}right) and SWE=propto frac{1}{left(URAN, DEPTHright)}$$

(8)

Read more here:
Deep learning algorithm-enabled sediment characterization techniques to determination of water saturation for tight ... - Nature.com

Read More..

AI is helping NASA pinpoint the most violent explosions in the universe – Earth.com

The dawn of artificial intelligence (AI) has ushered in a transformative era, promising to reshape every facet of our lives.Now, AI has moved off-world, helping NASA scientists unlock the secrets of the cosmos, including the location of gamma ray bursts (GRBs).

This exciting intersection of technology and astronomy is now a reality,thanks to recent research led by Maria Dainotti,a visiting professor at UNLVs Nevada Center for Astrophysics.

Dainotti and her teams have harnessed the power of AI to delve into the enigmatic world of gamma-ray bursts (GRBs) the most luminous and violent explosions in the universe.

By leveraging machine learning models,they have achieved unprecedented precision in measuring the distances of these cosmic behemoths.This breakthrough has far-reaching implications for our understanding of the universes evolution and the life cycles of stars.

Imagine an explosion so powerful that it releases as much energy in a few seconds as our sun does in its entire lifetime. Thats the awe-inspiring power of a GRB.

These cataclysmic events are visible across vast cosmic distances,even at the very edge of the observable universe.Their brilliance makes them invaluable tools for astronomers seeking to study the most ancient and distant stars.

But theres a catch: current observational technology limits our ability to gather the necessary data for calculating the distances of all known GRBs.This is where Dainottis ingenious approach comes in.

By combining gamma ray bursts data from NASAs Neil Gehrels Swift Observatory with multiple machine learning models,she has overcome these limitations, enabling far more accurate distance estimations.

This research pushes forward the frontier in both gamma-ray astronomy and machine learning, says Dainotti.

Follow-up research and innovation will help us achieve even more reliable results and enable us to answer some of the most pressing cosmological questions,including the earliest processes of our universe and how it has evolved over time.

One of Dainottis studies focused on using machine learning methods to determine the distances of GRBs observed by the Swift UltraViolet/Optical Telescope (UVOT) and ground-based telescopes.

The results were remarkably precise,allowing researchers to estimate the number of GRBs in a given volume and time with astonishing accuracy.

In another study,Dainotti and her collaborators employed a powerful machine learning technique called Superlearner to measure GRB distances using data from NASAs Swift X-ray Telescope.

Superlearner combines multiple algorithms, assigning weights based on their predictive power.This innovative approach significantly improved the reliability of distance estimations for a large sample of GRBs.

But the discoveries didnt stop there.A third study,led by Stanford University astrophysicist Vah Petrosian and Dainotti, used Swift X-ray data to shed light on the puzzling question of how GRBs are formed.

Their findings challenged conventional wisdom, suggesting that some long GRBs may originate from the fusion of incredibly dense objects like neutron stars, rather than the collapse of massive stars.

This opens the possibility that long GRBs at small distances may be generated not by a collapse of massive stars but rather by the fusion of very dense objects like neutron stars, explains Petrosian.

The implications of this research are vast and profound.By enhancing our ability to measure GRB distances,scientists can gain deeper insights into the evolution of stars and the overall structure of the universe.

The discovery of alternative formation mechanisms for long GRBs opens up new avenues for investigation into the most extreme cosmic phenomena.

Dainotti and her colleagues are now working to make their machine learning tools accessible to the wider scientific community through an interactive web application.

This will empower astronomers worldwide to build upon their groundbreaking work and further our understanding of the cosmos.

The marriage of AI and astronomy has ushered in a new era of cosmic exploration.With each new discovery,we inch closer to unraveling the mysteries of the universe and our place within it.The future of astronomy is bright,and AI is illuminating the path forward.

The study is published in the journal The Astrophysical Journal Letters.

Like what you read? Subscribe to our newsletter for engaging articles, exclusive content, and the latest updates.

Check us out on EarthSnap, a free app brought to you by Eric Ralls and Earth.com.

Link:
AI is helping NASA pinpoint the most violent explosions in the universe - Earth.com

Read More..

Automated Machine Learning (AutoML) Market Size to Exhibit a CAGR of 44.6% By 2028 – openPR

Automated Machine Learning (AutoML) Market

Download Report Brochure @ https://www.marketsandmarkets.com/pdfdownloadNew.asp?id=193686230

The AutoML market has been expanding rapidly in recent years, driven by the increasing demand for machine learning solutions across a variety of industries. AutoML tools offer a range of functionalities, such as automating feature engineering, hyperparameter tuning, model selection, and deployment. This allows data scientists, engineers, and businesses to build and deploy high-quality machine learning models much faster and with less expertise required.

Healthcare & Lifesciences to account for higher CAGR during the forecast period

The AutoML market for healthcare is categorized into various applications, such as anomaly detection, disease diagnosis, drug discovery, chatbot and virtual assistance and others (clinical trial analysis and electronic health record (EHR) analysis). In the healthcare and life sciences industry, AutoML can help automate various tasks such as disease diagnosis, drug discovery, and patient care. AutoML can be used to analyze large volumes of medical data, such as electronic health records, medical images, and genomic data, to identify patterns and make predictions. This can help healthcare professionals make more accurate diagnoses, identify potential treatments, and improve patient outcomes. AutoML can also be used in drug discovery to identify potential drug candidates and optimize drug development processes. By analyzing molecular structures, genetic data, and other factors, AutoML can help identify potential drug targets and optimize drug efficacy and safety. AutoML can also be used to monitor patient progress and adjust treatment plans as needed. The implementation of AutoML in healthcare and life sciences should be done with caution and consideration for ethical and regulatory concerns.

Services Segment to account for higher CAGR during the forecast period

The market for Automated Machine Learning is bifurcated based on offering into solution and services. The CAGR of services is estimated to be highest during the forecast period. AutoML services allow users to automate various tasks involved in building and deploying machine learning models, such as feature engineering, hyperparameter tuning, model selection, and deployment. These services are designed to make it easier for businesses and individuals to leverage the power of machine learning without requiring extensive knowledge or expertise in the field.

Asia Pacific to exhibit the highest CAGR during the forecast period

The CAGR of Asia Pacific is estimated to be highest during the forecast period. Automated machine learning is rapidly growing in Asia Pacific, which includes China, India, Japan, South Korea, ASEAN, and ANZ (Australia and New Zealand). In recent years, there has been significant growth in the adoption of both AutoML and machine learning across various industries in Asia Pacific, driven by the region's large and diverse datasets, as well as the need for faster and more efficient decision-making. Many companies in the region are also investing in the development of AutoML platforms and tools to help accelerate the adoption of AI and machine learning. To support the adoption of AutoML and machine learning, governments and organizations in the Asia Pacific region are investing in infrastructure and programs to promote innovation, education, and collaboration.

Get More Info About Automated Machine Learning Industry - https://www.marketsandmarkets.com/Market-Reports/automated-machine-learning-market-193686230.html

Top Companies in Automated Machine Learning :

Major vendors in the global Automated Machine Learning market are IBM (US), Oracle (US), Microsoft (US), ServiceNow (US), Google (US), Baidu (China), AWS (US), Alteryx (US), Salesforce (US), Altair (US), Teradata (US), H2O.ai (US), DataRobot (US), BigML (US), Databricks (US), Dataiku (France), Alibaba Cloud (China), Appier (Taiwan), Squark (US), Aible (US), Datafold (US), Boost.ai (Norway), Tazi.ai (US), Akkio (US), Valohai (Finland), dotData (US), Qlik (US), Mathworks (US), HPE (US), and SparkCognition (US).

Contact: Mr. Aashish Mehra MarketsandMarkets INC.

630 Dundee Road

Suite 430

Northbrook, IL 60062

USA: +1-888-600-6441

Email: sales@marketsandmarkets.com

About MarketsandMarkets

MarketsandMarkets is a blue ocean alternative in growth consulting and program management, leveraging a man-machine offering to drive supernormal growth for progressive organizations in the B2B space. We have the widest lens on emerging technologies, making us proficient in co-creating supernormal growth for clients.

The B2B economy is witnessing the emergence of $25 trillion of new revenue streams that are substituting existing revenue streams in this decade alone. We work with clients on growth programs, helping them monetize this $25 trillion opportunity through our service lines - TAM Expansion, Go-to-Market (GTM) Strategy to Execution, Market Share Gain, Account Enablement, and Thought Leadership Marketing.

Built on the 'GIVE Growth' principle, we work with several Forbes Global 2000 B2B companies - helping them stay relevant in a disruptive ecosystem. Our insights and strategies are molded by our industry experts, cutting-edge AI-powered Market Intelligence Cloud, and years of research. The KnowledgeStore (our Market Intelligence Cloud) integrates our research, facilitates an analysis of interconnections through a set of applications, helping clients look at the entire ecosystem and understand the revenue shifts happening in their industry.

To find out more, visit http://www.MarketsandMarkets.com or follow us on Twitter, LinkedIn and Facebook.

This release was published on openPR.

Original post:
Automated Machine Learning (AutoML) Market Size to Exhibit a CAGR of 44.6% By 2028 - openPR

Read More..

From density functional theory to machine learning predictive models for electrical properties of spinel oxides … – Nature.com

In the following section, we will give details about the methodologies that make up our proposed workflow for predicting the conductivity and band gap of spinel oxide materials: DFT for band structure calculation, principal layer, Greens functions, Landauer formalism for electric conductivity, and machine learning and band structure fitting to tight binding Hamiltonian.

All DFT calculations were done using the Vienna ab-initio simulation package (VASP)25 with the PerdewBurkeErnzerhof (PBE) functional26. We align the spins in a ferrimagnetic arrangement with Nel collinear configuration, where the magnetic moment of tetrahedral cations is opposing that of octahedral cations. Following convergence tests, the calculations for the cubic cells utilized a 600 eV plane wave energy cut-off and a 333 k-point mesh, centered at the -point. The self-consistent electronic minimization employed a 0.1meV threshold and ionic relaxation had a force threshold of 0.01eV/. The band structure was calculated using k-points along high symmetry lines, selected based on the implementation in the pymatgen27 library as described in ref28. The k-point file is available in the SI. The Hubbard U values applied to Mn, Co, Fe, Ni, and O atoms were 3.9, 3.5, 4.5, 5.5, and 0 eV, respectively, where all U values were used based on previous literature works29,30,31,32,33,34,35. The unit cell of all the spinels in the dataset consists of 56 atoms, 32 of which are oxygen, 16 of which are transition metal ions in tetrahedral sites and 8 transition metal ions in octahedral sites (Fig.8). The dataset contains 190 different combinations of this cell as explained later in the results.

An example of normal spinel oxide unit cell geometry (Co16Ni8O32). The red spheres are oxygen ions, gray are nickel ions in tetrahedral sites and blue are cobalt ions in octahedral sites.

For calculating the current, we followed the well-established approach of LandauerBttiker for quantum transport36,37. In this formalism, we postulate that there is a potential between two electron reservoirs that possess an equilibrium distribution and has dissimilar chemical potentials. These reservoirs act as sources and drains of electrons for the left and right electrodes, respectively. The device, which we intend to evaluate its current, constitutes the scattering region located between the electrodes. The variance in the electrochemical potentials of the reservoirs represents the applied bias on the scattering region. The LandauerBttiker is

$$I=frac{2e}{h}underset{-infty }{overset{infty }{int }}Tleft(Eright)left[fleft(E-{upmu }_{L}right)-fleft(E-{upmu }_{R}right)right]dE$$

(1)

where (fleft(E-{upmu }_{L}right)) and (fleft(E-{upmu }_{R}right)) are the FermiDirac electron distributions with ({upmu }_{L}) and ({upmu }_{R}) the chemical potentials of the left and right electrodes, respectively. (Tleft(Eright)) is the transmission function of an electron from left to right electrodes through the scattering region. In this setup the system is divided into principal layers that interact only with the nearest neighbors layer and each layer is described by a tight-binding Hamiltonian, ({H}_{i}), where i is the layer number. Since each electrode is semi-infinite, this approach introduces an infinite problem. Fortunately, this problem is solved by using Greens function. The scattering region which must interest us is a finite region that can be obtained by

$${{varvec{G}}}_{{varvec{S}}}={left(left(E+ieta right){{varvec{S}}}_{{varvec{S}}}-{{varvec{H}}}_{S}-{boldsymbol{Sigma }}_{L}-{boldsymbol{Sigma }}_{R}right)}^{-1}$$

(2)

where E is the energy of the system and (eta) is a small number. ({{varvec{S}}}_{{varvec{S}}}) is the overlap matrix of the scattering region The overlap matrix terms are defined as the overlap integral between associate vectors of the Hamiltonian basis. ({{varvec{H}}}_{{varvec{S}}}) is a block matrix representing the scattering region, and the effect of the semi-infinite electrodes is contained through the finite self-energy matrices ({{varvec{Sigma}}}_{{varvec{L}}backslash {varvec{R}}}) of electrode left (L) and right (R) electrodes.

The transmission formula can be obtained from the LandauerBttiker formalism and quantum scattering theory36,37,38,39, and it is given by

$${mathbf{T}}left( {mathbf{E}} right) = {text{Tr}}left{ {{{varvec{Gamma}}}_{{mathbf{L}}} {mathbf{G}}_{{mathbf{S}}}^{dag } {{varvec{Gamma}}}_{{mathbf{R}}} {mathbf{G}}_{{mathbf{S}}} } right}$$

(3)

where ({Gamma }_{Lbackslash R}) is called the broadening matrix and it is given by the expression

$${{varvec{Gamma}}}_{{{mathbf{L}}backslash {mathbf{R}}}} = {text{i}}left( {{{varvec{Sigma}}}_{{{varvec{L}}backslash {varvec{R}}}} -{varvec{varSigma}}_{{{varvec{L}}backslash {varvec{R}}}}^{dag } } right)$$

(4)

A technique for deriving finite self-energy matrices is available in the literature40,41. This procedure results in the following terms for the left electrodes self-energy.

$$hat{user2{Sigma }}_{{varvec{L}}} = {varvec{H}}_{{user2{L^{prime}},{varvec{L}}}}^{dag } {varvec{g}}_{{varvec{L}}} {varvec{H}}_{{user2{L^{prime}},{varvec{L}}}}^{{}}$$

(5)

where ({{varvec{H}}}_{{{varvec{L}}}^{boldsymbol{^{prime}}},{varvec{L}}}) is the coupling Hamiltonian between the scattering region and the left electrode. And ({{varvec{g}}}_{{varvec{L}}}) is given by a recursive formula,

$$g_{L}^{{left( {n + 1} right)}} = left[ {{varvec{H}}_{{varvec{L}}} - {varvec{H}}_{{user2{L^{prime}},{varvec{L}}}}^{dag } {varvec{g}}_{{varvec{L}}}^{{left( {varvec{n}} right)}} {varvec{H}}_{{user2{L^{prime}},{varvec{L}}}}^{{}} } right]^{ - 1}$$

(6)

The initial guess is ({{varvec{g}}}_{{varvec{L}}}^{left({varvec{n}}right)}={{varvec{H}}}_{{{varvec{L}}}^{boldsymbol{^{prime}}},{varvec{L}}}) and the iterative procedure continues until ({left({{varvec{g}}}_{{varvec{L}}}^{left({varvec{n}}-1right)}-{{varvec{g}}}_{{varvec{L}}}^{left({varvec{n}}right)}right)}_{ij}^{2}le {updelta }^{2}), where (updelta) is a small number in the order of 11015, this convergence criterion is achieved within few iterations. A similar procedure is done for the right self-energy (for symmetric electrodes and bias ({{varvec{g}}}_{{varvec{L}}}={{varvec{g}}}_{{varvec{R}}})). ({boldsymbol{Sigma }}_{{varvec{L}}}) takes the size of the scattering Hamiltonian and has a block element only in the top left of the left self-energy matrix and in the bottom right of the right self-energy,

$${boldsymbol{Sigma }}_{{varvec{L}}}=left(begin{array}{ccccc}{widehat{Sigma }}_{L}& 0& cdots & 0& 0\ 0& 0& 0& cdots & 0\ vdots & 0& 0& cdots & vdots \ 0& vdots & cdots & ddots & 0\ 0& 0& cdots & 0& 0end{array}right)$$

(7)

Our system is divided into the so-called principal layers42,43,44 where each layer consists of four unit cells of spinel oxide as illustrated in Fig.9. Using only four unit cells is sufficient to capture all the necessary information from the TB Hamiltonians that were adjusted to fit the DFT band structure. The layers are interconnected by a coupling Hamiltonian, effectively forming an endless wire. The Hamiltonian of a single layer of our system was defined as follows

$$H_{layer} = left( {begin{array}{*{20}c} {{varvec{H}}_{00} } & {{varvec{H}}_{010} } & {{varvec{H}}_{001} } & 0 \ {{varvec{H}}_{010}^{dag } } & {{varvec{H}}_{00} } & 0 & {{varvec{H}}_{001} } \ {{varvec{H}}_{001}^{dag } } & 0 & {{varvec{H}}_{00} } & {{varvec{H}}_{010} } \ 0 & {{varvec{H}}_{001}^{dag } } & {{varvec{H}}_{010}^{dag } } & {{varvec{H}}_{00} } \ end{array} } right)$$

(8)

where the elements in the Hamiltonian matrix are blocks of matrices. The blocks along the diagonal correspond to a tight-binding Hamiltonian (the tight-binding Hamiltonians defined in the next section) of each of the four unit cells within a single layer of spinel oxide. The block matrices along the off-diagonal represent the tight-binding Hamiltonian matrix elements that connect each unit cell with its neighboring cells within the same layer. The subscripts used for H indicate the spatial directions between a unit cell and its adjacent neighbors. The block matrix below defines the coupling between each layer and its neighboring layers,

A schematic representation of a wire built from principal layers of four unit cells each. The layers are coupled in the 100 direction.

$${H}_{layerCoupling}=left(begin{array}{cccc}{{varvec{H}}}_{100}& 0& 0& 0\ 0& {{varvec{H}}}_{100}& 0& 0\ 0& 0& {{varvec{H}}}_{100}& 0\ 0& 0& 0& {{varvec{H}}}_{100}end{array}right)$$

(9)

To maintain a manageable size for the Hamiltonians and take advantage of the electrodes being made of the same material, we opted to use only two layers to represent our material in the scattering region. This approach essentially yields an infinite material through which the current can flow. The Hamiltonian for the scattering region is expressed as follows:

$$H_{s} = left( {begin{array}{*{20}c} {{varvec{H}}_{{{varvec{layer}}}} } & {{varvec{H}}_{{{varvec{layerCoupling}}}} } \ {{varvec{H}}_{{{varvec{layerCoupling}}}}^{dag } } & {{varvec{H}}_{{{varvec{layer}}}} } \ end{array} } right)$$

(10)

The size of the scattering Hamiltonian is composed of 22 blocks representing two layers and their connection, and each block is composed of 44 block matrices that include ({{varvec{H}}}_{00},{{varvec{H}}}_{100},{{varvec{H}}}_{010},{{varvec{H}}}_{001}) , where these matrices have the size of 88 (since in this work we fitted to eight bands). Thus, in total the size of ({H}_{s}) is 6464.

The project emphasizes material property prediction and follows a workflow suited for such a task. The workflow includes the following steps: building a dataset by collecting and cleaning data and dividing it into a train and test set, selecting features, training the model by selecting an appropriate model, tuning hyperparameters, and optimizing it. The model's performance is evaluated on the test set and compared to other models, and finally, the model is used for making predictions.

To prepare for model training and testing, the dataset is first divided into two sets: a training set and a test set. The latter is used to evaluate the model's ability to predict properties on new data that it has not seen during training. It's a mistake to train and test on the same data since the model can overfit the training data and perform poorly on new data. A standard approach is to partition 80% of the dataset for training and reserve the remaining 20% for testing.

To set up a dataset for machine learning (ML) models, it should be divided into features and target variables. Features are independent variables grouped into a vector that serves as input for the ML model. Choosing informative and independent features is crucial for model performance. A single input vector represents a sample, and the order of features must remain consistent. The target variables are the properties to be predicted. For the materials in this project, A and B elements and stoichiometric numbers x and y from the formula AyBy8A16xBxO32 are the most representative features. Oxygen has a constant value and doesn't add information to the feature vector. The A and B atoms are represented by their atomic number values: Mn=25, Fe=26, Co=27, and Ni=28. Features normalization and scaling are important to ensure a proper functioning of some ML algorithms and to speed up gradient descent. In this project, we applied standardization using Scikit-learn module45, which involves removing the mean value and scaling to unit variance.

For the regression problems, which are predicting current from compositions, predicting current from bandwidth and prediction of bandgaps, we have chosen to utilize five different machine learning algorithms for comparison of their performance. For predicting current from composition these algorithms include kernel ridge regression (KRR), support vector regression (SVR), random forests (RF), neural networks (NN), and an ensemble model which takes the average result of all other four algorithms. For the prediction of current from bandwidth and for the bandgap prediction we used the XGboost algorithm46. These algorithms were chosen as they each have unique strengths in handling different types of data and relationships. Specifically, NN were chosen due to their ability to model complex, non-linear relationships, which is often the case in predicting material properties like electronic conductivity and band gaps; SVR was selected for its effectiveness in high-dimensional spaces, which is common in material science data. It also has strong theoretical foundations in statistical learning theory; KRR was chosen for its ability to handle non-linear relationships through the use of kernel functions. It also has the advantage of being less sensitive to outliers; RF was selected for its robustness and ease of use. It performs well with both linear and non-linear data.

ML models are functions defined by parameters. During training, the model optimizes the parameters to fit the data. Hyperparameters, like regularization values or architecture of a neural network, are tuned by the user. To avoid overfitting and ensure models perform well with new predictions, cross-validation procedures are used. The training set is split into k-folds, and the model is trained on k-1 folds and evaluated on the remaining fold (validation set) k times. The average performance is reported, and once hyperparameters are tuned, the model is trained on the whole training set and tested. In evaluating the model, selecting an appropriate metric is crucial. For this work, we have chosen the mean absolute error (MAE) as it provides a natural measurement of the accuracy of a model that predicts physical values such as current. Additionally, we have used R2 to measure the linearity between the predicted values of the models and the true values that were calculated.

Traditional tight-binding fitting schemes entail tuning numerous parameters that represent the atomic structure and do not always yield precise results47. A more common approach is utilizing Wannier functions, obtained by transforming extended Bloch functions from DFT calculations, which derive tight-binding parameters from ab-inito results without the need for fitting48,49. However, this method demands extensive system knowledge and a trial-and-error process due to the numerous parameters involved. Our objective is to rapidly and precisely compute the tight-binding Hamiltonian for approximately two hundred material samples.

Our approach follows the method proposed by Wang et al., where a parametrized tight-binding Hamiltonian is obtained by employing the back-propagation technique to fit the real-space tight-binding matrices to the DFT band structure50. Here we implemented the same approach but with pyTorch51 library instead of TensorFlow52 as done by the original authors. In TB theory the reciprocal space Hamiltonian for a desired K vector is,

$${H}^{TB}(k)={sum }_{R}{e}^{ikcdot R}{H}^{R}$$

(11)

where R is a lattice vector of selected real-space Hamiltonian matrices and ({H}^{R}) is the corresponding TB Hamiltonian. H0 (R=(000)) is the real-space TB Hamiltonian matrix of the unit cell, ({H}^{R}ne 0) (R=(100), (010), (001), etc.) are the Hamiltonian matrices that couple the unit cell Hamiltonian to the adjacent unit cells in the direction of R vectors. The TB band structure is obtained from the eigenvalues ({{varepsilon }^{TB}}_{n,k}) for every desired k vector via,

$${H}_{k}^{TB}{psi }_{n,k}^{TB}={{varepsilon }^{TB}}_{n,k}{psi }_{n,k}^{TB}$$

(12)

where ({{varepsilon }^{TB}}_{n,k}) is the energy of the n-th band at reciprocal vector k. To fit the TB bands to the DFT bands, the mean squared error loss L between the TB and DFT eigenvalues is minimized,

$$L=frac{1}{N}frac{1}{{N}_{k}}{{sum }_{i=1}^{N}{sum }_{k}{left({varepsilon }_{i,k}^{TB}-{varepsilon }_{i,k}^{DFT}right)}^{2}}$$

(13)

The loss is computed for all bands and k-points, where N represents the total number of bands and Nk represents the number of sampled k-points. The TB Hamiltonians parameters (HR) are updated iteratively using the Adam gradient descent algorithm53. The back-propagation procedure is used to calculate the derivative of the loss with respect to HRs. The gradient descent algorithm seeks to minimize the loss by moving in the direction of the steepest descent, which is opposite to the direction of the gradient, according to the general formula,

$${H}_{l+1}^{R}={H}_{l}^{R}-alpha frac{partial L}{partial {H}_{l}^{R}}$$

(14)

The variable "l" represents either an epoch number (epoch defines the number of times the entire data set, in this case all the k points, has worked through the learning algorithm) or a single iteration over all k points. In this context, an epoch is considered complete after the loss function has accumulated all the eigenvalues in the band structure, which occurs after calculating all ({H}^{TB}(k)) values. The derivative of the loss with respect to HR is given with matrix derivative algorithmic differentiation rules54.

The fitting process has a predefined numerical threshold (means squared error<(3cdot {10}^{-5} {text{eV}}^{2}), as defined in Eq.13) for the loss function, which serves as a criterion to terminate the fitting. Additionally, we introduced another criterion that halts the fitting if the loss increases for more than ten epochs due to spikes or jumps in the loss function during training. If the loss was already low, we consider the results of the fitting process. However, if the loss was high, we re-run the process with slightly different initial random weights in the Hamiltonians.

To determine the R vectors for the fitting process, we conducted tests with various vectors, up to 24 in total. After experimentation, we concluded that the three main directions and their opposite directions were sufficient for achieving a highly accurate fitting of the bands. The size of the matrices used in the fitting process is defined by the number of bands being fitted. For instance, for eight bands, the matrix size is 88.

More:
From density functional theory to machine learning predictive models for electrical properties of spinel oxides ... - Nature.com

Read More..

Focus on Humanity in the Age of the AI Revolution – InformationWeek

The arms race for the next AI breakthrough is upon us -- and for good reason. While the news cycles are quick to point to fears around job displacement, bias, security risks, and weaponization, there is reason to be optimistic about this technology if we proceed carefully. We are stepping into the future of an AI-enabled world, and the competitive landscape for AI innovation should have only one bias in mind: a bias towards humanity.

AI should be viewed as a tool for enhancing human capabilities and productivity rather than one that replaces them. AI has the potential to amplify creativity, solve complex problems faster, and tackle tasks that are lengthy, tedious or impossible for humans. By positioning AI as a partner in progress, we can foster an environment where technological advancement serves to elevate human potential. From the perspective of the enterprises and innovators that are building this technology, its important that these tools are built with that idea at the core.

These improvements are lifesaving in certain sectors. In healthcare, AI algorithms are deployed and being refined to allow practitioners to diagnose diseases quickly and accurately, spotting early signs of conditions like cancer, diabetic retinopathy and heart disease from imaging scans that surpass the human eye by itself. AI systems can also speed up drug discovery and development by predicting molecular behavior and identifying promising drug candidates much faster than traditional methods. That significantly reduces the time and cost to bring new medicines to market.

Related:What You Need to Know about AI as a Service

Another example is in education. AI breakthroughs are establishing a more personalized process by adapting content and pacing it to individual learners needs. Machine learning can identify areas where students struggle, provide tailored resources, and enhance engagement through interactive and immersive experiences. This can help teachers create more tailored learning processes for each of their students, including those with special needs, so no one student gets left behind.

This just scratches the surface. AI has the potential to play a role in a variety of industries from manufacturing, supply chain, entertainment, environmental conservation, and much more.

Despite fears surrounding job displacement, history has shown that technological advancements lead to more jobs, not fewer. The advent of AI is no exception. It's true that repetitive tasks that are tedious in nature and easily structured, such as data processing and extraction from documents, will slowly start being automated away. However, these systems will have to be overseen by specialists and experts to ensure everything is running in a user-friendly way.

Related:Is an AI Bubble Inevitable?

By automating these routine tasks, AI frees humans to engage in more creative, strategic, and interpersonal roles. This transition underscores the importance of retraining initiatives to prepare the workforce for the jobs of the future. As the workforce evolves alongside technology, AI will serve more as a job creator rather than a job destroyer, as long as companies invest in upskilling their employees with the necessary capabilities to understand, leverage, implement, and improve technology.

We're already seeing evidence of this phenomenon. Big companies are starting to create new opportunities revolving around machine-learning or Gen AI capabilities. As Aaron Mok of Business Insider points out, job postings on LinkedIn that mention GPT have grown 20-fold from 2022 to 2023. He goes on to point out how Meta, Netflix, Apple and even non-tech companies in the healthcare, education, and legal industries have posted listings for AI-related jobs. Those who learn how to operate this technology effectively will have a growing number of employment opportunities.

Related:Overcoming AIs 5 Biggest Roadblocks

Theres a Pandora's box opening up as it relates to AI as it plays an increasingly significant role in creative and inventive processes, raising questions about authorship and ownership. At what point does a user take what is being given to them from an AI algorithm without challenging it? What are the ethical considerations of a consumer or an employee taking those considerations or recommendations and then parading them as their own work versus the work that's being done by the AI engine?

Addressing these concerns requires a multi-stakeholder approach, including policymakers, legal experts, technologists, and creators, to redefine intellectual property frameworks that recognize the contributions of both humans and AI. This could involve exploring new licensing models, establishing AI as a tool under human direction, establishing regulations and laws, and ensuring that creators are fairly compensated for AI-assisted work.

A human-centric approach should guide AI in all its developments. This involves ensuring transparency, accountability, and fairness in AI systems, with a focus on minimizing biases and increasing accessibility. Its important that AI systems also consider those with disabilities or sensory impairments to make their lives easier rather than creating more walls. Addressing AI bias requires a multifaceted approach, including diversifying datasets, implementing fairness metrics, and incorporating diverse perspectives in AI development teams.

By adopting a proactive, collaborative, and regulation-informed approach, we can harness the benefits of AI while safeguarding against its risks. The arms race of AI is here; but we can ensure that practitioners and thought leaders are always within arms reach of any AI implementation introduced into our daily lives.

Read this article:
Focus on Humanity in the Age of the AI Revolution - InformationWeek

Read More..

A machine learning-based model analysis for serum markers of liver fibrosis in chronic hepatitis B patients | Scientific … – Nature.com

In this multicenter study, we designed a prediction model based on ML to accurately assessment liver fibrosis stages of CHB patients. Compared with traditional statistical models such as APRI or FIB-4, and ML model demonstrated significant improvements and was easy to process, which also suggested the great potential of ML in the field of noninvasive liver fibrosis evaluation. In addition, our study results indicated that ML model provided similar diagnostic efficacy with the reference standard liver biopsy, which may provide a reliable theoretical basis for the further development of simple, easy-to-use and accurate tools for the evaluation of liver fibrosis.

In this study, we used ML methods with the hope of more accurately assessing the staging of liver fibrosis, thereby improving the accuracy of the model. The final results revealed that our model showed superior accuracy compared to traditional serological models such as APRI or FIB-4. It is also significantly higher than the diagnostic efficacy of seventeen noninvasive liver fibrosis models in Chinese patients with hepatitis B mentioned in the study of Li et al.19. In addition, stratification analysis in inflammation subgroups was performed, and the results did show no significant impact on the performance of ML model. These findings suggest that ML model may overcome the influence of inflammation for cirrhosis evaluation, which is likely to be a potential breakthrough in non-invasive diagnosis. This was helped by a new approach to model building that had the following main advantages. First, we compared the performance of models constructed by several ML methods, and then we focused on and validated the DT model because of its better performance and ease of use. In fact, the DT model has been applied to evaluate hepatitis C liver fibrosis and has shown significant performance20. In addition, previous studies mainly used a classification method (logistic regression analysis)21, and features were selected through univariate tests (t tests, Welch tests, etc.) in many patients22,23. However, this method is often overly optimistic, prone to overfitting, and difficult to reproduce. To overcome these problems, we used integration algorithms, including mRMR and GBDT, to remove redundant features to prevent multicollinearity, and we used only high-scoring variables to construct prediction models to avoid overfitting. Second, our model allows patients to be assessed by a single blood draw without the need for additional modalities. This concept is particularly attractive for routine screening of people at high risk of disease development, such as those with advanced or severe liver fibrosis, in primary care settings. These cases which clinically suspected severe liver fibrosis previously required puncture pathology to be confirmed. However, now only need to routine serological examination to judge the probability of severe liver fibrosis, so invasive puncture examination can be avoided. Therefore, it has obvious advantages in terms of cost and prognosis. In addition, our method can be used to construct a similar model visualization to distinguish early liver fibrosis from significant liver fibrosis, and does not require specially trained clinicians, which is more convenient for clinicians in practice and of great value for clinical promotion.

In this study, we also hoped to improve the diagnostic performance of the model by identifying more specific markers and constructing the model based on the combination of known serologically relevant features. We integrated some of the most routine serological markers, in contrast to Zeng et al., who used laboratory markers such as B2-macroglobulin, haptoglobin and apolipoprotein A1, which are not commonly used in most hospitals24. Although these laboratory markers may show higher accuracy than routine serological markers, they are not suitable for practical clinical application. Our results showed of the five conventional serological markers used to construct the ML model, HBV-DNA had the greatest contribution to the model, which is consistent with the recommendation of some guidelines that patients with high HBV-DNA levels should be evaluated for noninvasive liver fibrosis4,25. HBV DNA is the marker for viral replication. For chronic HBV infection, the development of the disease is a dynamic process, and the infection status also exists for a long time. For patients with chronic HBV infection in the indeterminate phase, the results of examination alone may not be able to accurately assess the natural history stage, so dynamic follow-up observation is needed. Studies have shown that HBV DNA levels correlated with significant fibrosis in HBeAg() CHB patients. HBV DNA level could predict liver fibrosis in HBeAg() CHB patients with biopsy indication26,27.

In addition, two coagulation factors including INR and TT were integrated into the model, although the two coagulation factors are closely related in clinical practice28,29, which was may lead to over fitting of the model and overestimate the role of coagulation factors. However, we calculated the VIF value of relevant factors and did not show collinearity. Therefore, we speculate that the contribution of coagulation factors to the model should not be overestimated.

It is well known that distinguishing F0-1 from F2-4 is more challenging in many studies30,31, which is because the heterogeneity of liver fibrosis in patients with F2 liver fibrosis is more serious than that in those with F3 and 4 liver fibrosis, which generally reduces the accuracy of all classification strategies. In fact, our research results confirm that DT model has the lowest accuracy (AUC of 0.891 in training cohort and AUC of 0.876 in Validation cohort) in identifying patients with liver fibrosis grade F2. However, DT model shows high accuracy and excellent stability for each fibrosis grade in two cohorts, especially in identifying liver cirrhosis (F4), which was shows this model could be used to refine phenotypes in large research studies. Our study result also showed that the highest overall recognition rate for patients with liver cirrhosis (F4) was higher than that for patients with other stages of liver fibrosis when the model was used to classify risk prediction in the two cohorts or the whole cohort. These results suggested that our ML model may be part of a more accurate preclinical detection pathway to assess liver cirrhosis and may be used for the screening and treatment of liver cirrhosis in HBV-infected patients in routine clinical environments, although this needs to be validated in prospective studies.

This study has some limitations. First, this study was a retrospective study, which may lead to the simulation of retrospective statistics depending on too many assumptions. Future research should focus on the development of prediction and classification models based on prospective research, which will allow time evolution information to be used to evaluate, modify and reevaluate prediction models. Second, the model itself needs to be further optimized through better engineering and further development through more comprehensive integration of other clinical data to improve the overall performance of the model and achieve a more accurate noninvasive diagnosis of liver fibrosis staging. Finally, our study did not investigate the performance of ML model for classifying patients with CHB of different ethnic populations, which are also worthy of further studies in the future. Of course, in this study, we still emphasize that as conceptual research, it can still provide a certain basis for the real clinical practice in the future, although this future still needs a long way to go.

In conclusion, this study demonstrated that ML model was more accurate than traditional serological mixed biomarkers in assessing all four liver fibrosis stages in patients with CHB. In addition, the results of this study promote the goal of assessing liver fibrosis in CHB patients and improving the existing prognostic models, thereby facilitating a future prospective study design and evaluation and clinical disease surveillance and treatment. We also hope to further refine and expand this work to clarify the application of this model to a wider range of liver fibrotic diseases.

Follow this link:
A machine learning-based model analysis for serum markers of liver fibrosis in chronic hepatitis B patients | Scientific ... - Nature.com

Read More..