Page 450«..1020..449450451452..460470..»

Prediction of the axial compression capacity of stub CFST columns using machine learning techniques | Scientific … – Nature.com

The scatter plots in Fig.7 illustrate the relationship between experimental and predicted outcomes for various ML models applied to training and testing datasets for columns with different cross-section shapes. It can be observed that the data points tightly gather around the diagonal line for most of ML models, signifying a strong alignment between the model predictions and experimental results. This alignment signifies the reliability and prediction accuracy of the developed models. Table 4 introduces evolution metrics to assess the performance of the established ML models, including the mean (), coefficient of variance (CoV), coefficient of determination (R2), root mean squared error (RMSE), the mean absolute percentage error (MAPE), and a20-index, defined as follows:

$$begin{aligned} & mu = frac{1}{n}mathop sum limits_{i = 1}^{n} frac{{y_{i} }}{{hat{y}_{i} }}, ;;R^{2} = 1 - frac{{mathop sum nolimits_{i = 1}^{n} left( {hat{y}_{i} - y_{i} } right)^{2} }}{{mathop sum nolimits_{i = 1}^{n} left( {y_{i} - overline{y}} right)^{2} }} RMSE = sqrt {frac{1}{n}mathop sum limits_{i = 1}^{n} left( {hat{y}_{i} - y_{i} } right)^{2} } , \ & MAPE = frac{1}{n}mathop sum limits_{i = 1}^{n} left| {frac{{hat{y}_{i} }}{{y_{i} }} - 1} right| times 100% \ end{aligned}$$

(12)

where ({y}_{i}) and ({widehat{y}}_{i}) are the actual and predicted output values of the i-th sample, respectively, (overline{y }) is the mean value of experimental observations, and n is the number of specimens in the database. The a20-index16,38 measures the percentage of samples with actual to prediction ratio, ({widehat{y}}_{i}/{y}_{i}), falling within the range of 0.801.20.All data generatedand algorithms introduced in this study are included in thesupplementary file.

Comparison between proposed equations and ML models for training and testing datasets.

As shown in Table 4, all introduced ML models display mean , R2, and a20-index values close to 1.0 and small values for CoV, MAPE%, and RMSE for different cross sections. The prediction results of all introduced models exhibit CoV less than 0.076, MAPE% lower than 6%, and RMSE less than 552kN, indicating minimized scattering in the prediction results compared to the experimental results. Table 4 reveals that the CATB, GPR, and XGB models introduce the best evaluation metrics for the testing subsets, with MAPE% values equal to 1.394%, 1.518%, and 2.135% for CCFST, RCFST, and CFDST column datasets, respectively. In addition, PSVR can accurately predict the capacity of stub CFST columns with MAPE% values equal to 2.497 and 5.151 for CCFST and RCFST columns, respectively. The superior predictive capability of PSVR demonstrates that the SVR model, incorporating the metaheuristic optimization methods39 like the PSO algorithm, can significantly enhance the performance of the SVR model.

Furthermore, the evolution metrics of the testing resemble those of the training set, except for the GPR and CATB models. However, the performance of GPR and CATB models in the testing set is comparable to that of the remaining data-driven models and even better than that of other ML models. In addition, when examining the R2 value and a20-index for the entire dataset, it was found that they are nearly identical to those of the test and training subdatasets. Such robust and stable alignment between the performance of sub-datasets signifies a minimal occurrence of overfitting during the training process of the models.

Although the GPR, CATB, XGB models stands out with significantly superior results compared to other models, extracting an explicit design formula from these models is a challenging task. In contrast, the proposed equations extracted from the SR algorithm offer a distinct advantage by providing simple and practical explicit design formulas, making them more accessible and easier to interpret, even with slightly lower accuracy than the introduced ML models. Although ANN could provide accurate and explicit formulas for strength prediction, utilizing the network in engineering design might not be practical due to the lengthy formulas of the ANN model.

The compressive strength predictions of CFST columns by the proposed equations were compared with the existing code formulas, including EC430 and AISC36029 for different types of columns. As observed in Table 4, for all types of CFST columns, the proposed equations attain a mean, R2, and a20-index nearly equal to 1.0 with CoV less than 0.076 and MAPE% less than 5.9, while EC4 and AISC360 show CoV larger than 0.091, 0.168 with MAPE% larger than 7.1% and 15%, respectively. In addition, the AISC360 predictions, compared to EC4 predictions, appear to overestimate the axial capacity for different cross sections with a higher mean approaching 1.20, lower a20-index, and relatively high error indices. The RMSE and MAPE of AISC36029 predictions are approximately two to six times those of EC430, indicating the better performance of EC4 compared to AISC360. In addition, AISC360 introduces an a20-index with a value approximately 50% lower than that obtained from the EC4 results. This discrepancy could stem from the absence of confinement effect calculations in AISC36029. Although all cited codes standards display a safe design, the error indices introduced by the ML models and proposed equation are significantly small compared to these standards. Specifically, the proposed equations demonstrate superior performance compared to these standards across all evaluation criteria.

Figure8 displays the prediction errors of the design standards and the developed ML models for different cross sections. It indicates that most of the introduced ML models are more accurate than the design standards, especially for the GPR, CATB, and PSVR models, implying the superiority of these ML methods in estimating the axial capacity of stub CFST columns. In the case of CCFST columns, the CATB, GPR, and PSVR models display more than 95% of test samples within the 10% error range, while the proposed equation, EC4, and AISC360 show 83%, 75%, and 7% of test samples, respectively, within the same range. For RCFST columns, all ML models exhibit accuracy, with over 75% of test samples falling within a 10% error range, while the corresponding proportions for the proposed equation, EC4, and AISC360 are 85%, 73%, and 42%, respectively. Regarding CFDST columns, all ML models, excluding the RF and LGBM models, correctly predict 90% of the specimens within a 10% range error, while the proposed equation, EC4, and AISC360 attain nearly 83%, 68%, and 17% accuracy for the test samples, respectively, within the same error range. Thus, the introduced ML models can be considered valuable tools alongside the design standards in estimating the axial capacity of stub CFST columns.

Prediction errors of design standards and established ML models.

See the original post here:
Prediction of the axial compression capacity of stub CFST columns using machine learning techniques | Scientific ... - Nature.com

Read More..

Virginia legislators should be learning all they can about AI – News From The States

When you let your phone unlock itself by scanning your face, ask Alexa or Siri for directions, shop online, apply for a job or check your account in a banking app, your experience is underpinned by artificial intelligence (AI), technology that makes machines mimic human attributes like thinking, learning and suggesting. As AI continues infiltrating nearly every aspect of our society, its crucial that Virginia lawmakers and experts continue to study its benefits and risks. That task is particularly urgent as the technology has been shown to perpetuate inequity that already exists in many of our systems.

There are several pieces of legislation related to AI up for debate now in Virginias General Assembly. House Bill 249 calls for the development of a comprehensive framework for how law enforcement agencies use AI and a model policy to guide criminal justice systems use of machine learning technology by 2025.

Such measures sure would have been helpful five years ago, before Norfolk police briefly used a facial recognition app developed by Clearview AI to identify suspects in criminal investigations without public knowledge and without oversight from city or state leaders. Not only was that a huge privacy concern, but it could also have misidentified people especially people of color and connected them to a crime they had nothing to do with.

The U.S. National Institute of Standards and Technology (NIST) tested 189 face recognition algorithms, and found that most facial recognition AI systems have a significantly higher rate of false positive identifications for photos of Black, Asian, and indigenous peoples faces than for photos of white peoples faces, Dr. Jennifer Rhee, director of the AI Futures Lab at Virginia Commonwealth University, told me by email. Some facial recognition systems were found to have false positive rates 10 [times] to 100 [times] higher for Black, Native American, and Asian people than for white people.

Despite those very high error rates, versions of this technology are currently being used by some police departments in America, and for surveillance of public spaces and national borders. That may spell serious consequences for the people who are erroneously identified by these systems, Rhee added.

Worker protection laws arent ready for artificial intelligence, automation and other new technology

Facial recognition technology is riddled with bias, because most of the images used to train algorithms to recognize faces are white, male faces. Most of the engineers developing this technology over the past decade are also white and male; they create tech tools in their own image, and any bias they have is reflected in those creations.

Recent studies demonstrate that machine learning algorithms can discriminate based on classes like race and gender, reads a 2018 MIT study led by Joy Buolamwini and Timnit Gebru that evaluated three commercial AI-powered facial recognition tools that consistently misidentified darker-skinned peoples gender, while almost always getting the gender of lighter-skinned people right. Other tools dont recognize Black faces as human, as demonstrated in 2015 when Google Photos automatically sorted 80 photos of a Black man into a folder it labeled gorillas.

Because AI systems are trained on data produced by society, these systems reflect societys biases and power dynamics, Rhee told me by email.

Artificial intelligence has been under study by state lawmakers since 2020, when the Artificial Intelligence Subcommittee of the Joint Commission on Technology and Science was founded. The function of the group, subcommittee chair Del. Cliff Hayes, D-Chesapeake, told me last summer, is to help legislators get a better understanding of how the technology works and its implications in Virginia.

We need to be equipped to deal with this in Virginia because AI and other technologies are evolving so rapidly, we cant necessarily sit around and wait for federal guidelines, said Hayes, who has had a nearly 30-year career in programming and technology management and is currently the city of Portsmouths chief information officer. Its important to view AI and other technological developments holistically, and to not focus exclusively on either the positive or negative aspects, he added.

As employers expand artificial intelligence in hiring, few states have rules

In 2021, Virginias legislature banned police departments across the commonwealth from using facial recognition technology without first gaining permission from the state. But policymakers quickly concluded that approach meant Virginia would miss out on the legitimate ways AI could aid law enforcement and criminal justice efforts.

An all-out ban of AI isnt the answer, we learned, Hayes said. The answer was to dissect and deal with the technology as it advances, so thats what were doing.

In 2022, legislators rolled back the ban, allowing police departments to use facial recognition only in certain instances. That legislation also required the Virginia State Police to develop a model policy regarding the investigative uses of facial recognition technology and mandated that any facial recognition technology used in Virginia be evaluated by the National Institute of Standards and Technology and have an accuracy score of at least 98 percent true positives across all demographic groups. The second measure was intended to address public concerns that the technology is less accurate in recognizing the faces of people with darker skin tones.

In June, Attorney General Jason Miyares co-authored a letter written by attorneys general of 23 states entreating the National Telecommunications and Information Administration to review AI governance policies. The letter cited data privacy, the need to develop AI safely without hampering innovation and the technologys possible impact on individuals legal and personal information as causes for concern and caution.

Last September, Gov. Glenn Youngkin issued a similarly themed executive order on artificial intelligence that focused on four areas: legal requirements, especially those related to privacy or intellectual property; policy standards for state agencies; appropriate IT safeguards such as cybersecurity and firewalls; and training for students.

That sounded good and all, except there wasnt a focus area dedicated specifically to one of the foremost challenges of AI: its well-known, deeply researched biases that fuel systemic discrimination against some groups of people, usually those who are already marginalized. This directive made me wonder if the governor truly had an interest in safeguarding Virginians of color and impoverished people from this pervasive problem.

Youngkins latest AI-related executive order, released in January, steps up to meet that challenge. Standards for how the technology should be used and taught in the states public schools and guidelines for AIs use in state agency databases issued through the order do strive to prevent harmful AI practices based on inherent biases that lead to discriminatory outcomes and mitigate any risk of bias and discrimination by AI Systems. Thats a sign the governor is listening to and learning from experts and everyday citizens concerned about how AI could unfairly impact some of us; its a good start.

Sen. Suhas Subramanyam, D-Loudoun, a few weeks ago introduced Senate Joint Resolution 14, which orders the Joint Legislative Audit and Review Commission to continue studying the impact of AI in the state with a specific focus on understanding deep fakes, data privacy implications, and misinformation, making sure the technology doesnt indirectly or directly lead to discrimination, finding ways to promote equity in AI algorithms, and looking at how AI can improve government operations and services.

Subramanyam, who was a technology policy adviser in the administration of former President Barack Obama, said his professional as well as personal experiences prompted him to put forward this measure.

We shouldnt necessarily stifle and overregulate [artificial intelligence], but we should look at how we can prevent bad practices and behaviors that stem from it, he told me during a 9 p.m. interview earlier this week. (It was the only time he could talk after a full day at Capitol Square; I whispered my questions to him from my bedroom closet, hoping not to wake my kids. That he would make time to speak on AI after hours spoke volumes about his interest in and passion for the subject.)

I come from an immigrant family, and one of the things Ive found, as someone with a funny name like mine and a background like mine, you tend to find a lot of overgeneralizations or caricatures of your culture already, and those things also show up in the datasets that we use with AI and emergent tech, Subramanyam said. If youve got a dataset thats not representative of all communities and cultures, then you will have an outcome that represents that characterization and misrepresentation.

Finally, among the roughly two dozen measures dealing with AI that are on the table this session is Senate Bill 621, which would establish a commission dedicated solely to advising the governor on AI policy in Virginia; I think Youngkin could surely use the help, and future leaders could too. Through this bill and others, Virginia now has the opportunity to implement artificial intelligence, machine learning and other emerging technologies in a responsible, ethical way that takes into account the specific harms possible for many of its citizens.

Alexa, play Ice Cubes You Can Do It.

Read the original post:
Virginia legislators should be learning all they can about AI - News From The States

Read More..

Operationalizing Machine Learning to Drive Business Value | by dparente | Daniel Parente | Feb, 2024 – Medium

MLOps Pipeline

MLOps, or Machine Learning Operations, brings together processes, best practices, and technologies to manage putting machine learning models into production environments at scale. It fills a major gap enterprises face in getting return from AI and analytics investments.

Research shows only 15% of major companies have widespread machine learning applications running across their business. So the majority of expensive modeling work stays stuck in labs and pilot projects. MLOps fixes this bottleneck by automating the steps needed to deploy, monitor, and update models in reliable pipelines.

Key business benefits MLOps delivers includes:

Without MLOps, models degrade, data science productivity drops, and adoption stalls. Adding MLOps boosts ROI on analytics spending by maintaining model performance post-deployment.

MLOps engineers build the continuous development and deployment capabilities for machine learning models to run successfully as applications. Their expertise combines software engineering, data engineering, and DevOps skills tailored for operationalizing analytics.

Their key responsibilities include:

The rest is here:
Operationalizing Machine Learning to Drive Business Value | by dparente | Daniel Parente | Feb, 2024 - Medium

Read More..

Nous Achieves Microsoft AI and Machine Learning Advanced Specialization – AiThority

Nous Infosystems, a leading global IT solutions provider delivering innovative technology services and solutions, announced that it has achieved theMicrosoft AI and Machine Learning Specialization. This advanced specialization solidifies Nous position as a trusted partner for clients and affirms exceptional capabilities in delivering successful solutions leveraging advanced AI and Machine Learning (AI/ML) capabilities.

With a proven track record of delivering innovative technology services and solutions, Nous continues to showcase its dedication to excellence and the ability to provide outstanding real-world solutions in the rapidly evolving field of AI/ML. The achievement serves as a validation of Nous profound expertise, extensive experience, and demonstrated success in facilitating customer adoption of AI/ML implementation on Azure for AI-powered solutions. Earning the AI and Machine Learning advanced specialization signifies Nous commitment to delivering high-quality solutions and reinforces its position as a trusted partner in the AI and machine learning domain.

Recommended AI News: Secureworks Threat Score Ushers In A New Age Of Cybersecurity AI

Securing the Microsoft AI and Machine Learning Specialization signifies Nous Infosystems steadfast commitment to excellence in technical proficiency and business acumen, emphasizedAnurag Chauhan, CEO at Nous Infosystems. Our attainment of this certification reinforces our dedication to delivering cutting-edge solutions, bringing significant value to our clients across various industries by leveraging AI and Machine Learning.

Dan Rippey, Program Director for the Microsoft AI Cloud Partner Program, highlights, Nous has demonstrated exceptional technical prowess and proven success in aligning outcomes with the Microsoft Cloud. As a Solutions Partner for Digital & App Innovation (Azure) and Data & AI (Azure), Nous consistently showcased its technical capabilities and experience. Their specialization in AI and Machine Learning underscores their deep expertise in critical technical scenarios, solidifying their position as a trusted partner in the Microsoft ecosystem.

Recommended AI News: Rivet Revenue Diagnostics Diagnoses Financial Wellness for Healthcare

Earning the Microsoft AI and Machine Learning Specialization is a demonstration of our teams dedication to staying at the forefront of artificial intelligence and machine learning, says,Sreenivasan Narayanan, EVP Microsoft Global Alliances at Nous Infosystems. This certification validates our expertise in leveraging Microsofts advanced AI/ML tools, enabling us to provide cutting-edge solutions to our clients. It reflects our commitment to delivering innovative AI/ML projects and solidifies our position as a reliable partner for organizations seeking transformative solutions in this dynamic technological domain.

Recommended AI News: CDNetworks Upgrades WAAP Solution with Its Latest AI-Powered Cloud Security 2.0 Platform

[To share your insights with us as part of editorial or sponsored content, please write to sghosh@martechseries.com]

Read more from the original source:
Nous Achieves Microsoft AI and Machine Learning Advanced Specialization - AiThority

Read More..

Bell to Deploy Deep Learning AI on Systems and Data in 18-Month Partnership With Mila – The Fast Mode

Mila and Bell last Friday announced an 18-month collaborative project to apply deep learning neural network algorithms to Bell's systems and data. Bell has made significant investments to develop extensive data analytics capabilities and AI applications in multiple areas of its operations, and this collaboration is the latest step in advancing its AI expertise.

Mila researchers will work alongside Bell's Machine Learning and AI teams to build on those investments by using cutting-edge deep learning neural network techniques to identify opportunities for improving business performance and customer experience.

These neural network deep learning models, inspired by the human brain, teach computers to recognize complex patterns in pictures, text, sounds and other data to produce accurate insights and predictions.

By advancing its understanding of deep learning AI techniques, Bell will continue to enhance its customer experience and accelerate its transition from a traditional telecommunications company to a technology services leader. As part of the collaboration, Bell and Mila will write a paper highlighting their technical findings in support of global AI advancement.

Stphane Ltourneau, Executive Vice President of Mila

Mila is very pleased to work with Bell and apply its renowned expertise in deep learning to the telecommunications sector. Through this collaboration, we look forward to combining Mila's research capabilities with Bell's extensive industry knowledge in order to highlight and harness AI's potential in this evolving field.

Michel Richer, SVP, Enterprise Solutions, Data Engineering & AI, Bell Canada

Becoming an AI leader is key to our transition from a traditional telco to a tech services leader. Working with a global leader like Mila right here in Montral is a great opportunity for Bell to benefit from leading-edge researchers, advance our AI and Cloud expertise, further improve the customer experience and establish our role as a technology services leader.

See the rest here:
Bell to Deploy Deep Learning AI on Systems and Data in 18-Month Partnership With Mila - The Fast Mode

Read More..

Can you explain the artificial intelligence? | by Vikram gounder | Feb, 2024 – Medium

Photo by Maximalfocus on Unsplash

Artificial intelligence (AI) is a term that is often used in the world of technology, but many people are not fully aware of what it means. In simple terms, AI can be defined as the ability of a computer or machine to think, learn, and make decisions similar to a human. It is a rapidly growing field that has the potential to transform various industries and aspects of our daily lives.

To understand AI better, it is essential to break it down into its two main components: machine learning and deep learning. Machine learning is the process of training a computer to recognize patterns and make decisions based on data, without the need for explicit instructions. This is done by feeding the computer large amounts of data, and it uses algorithms to learn from that data and improve its performance. For example, if you show a computer thousands of pictures of cats, it can learn to recognize a cat in a new image without being explicitly told that it is a cat.

Deep learning is a subset of machine learning that uses artificial neural networks to process and analyze large sets of data. These neural networks are modeled after the human brain and can learn and make decisions on their own. They can also improve their performance over time by learning from their mistakes.

So how does AI work? First, data is collected and fed into an AI system. This can be in the form of text, images, or any other type of data. The AI system then analyzes this data, looking for patterns and connections. It uses these patterns to make decisions or predictions. The more data it is exposed to, the better it becomes at making accurate decisions.

One of the most significant advantages of AI is its ability to handle large amounts of data at a speed and accuracy that is impossible for humans. This makes it useful in various fields, including finance, healthcare, transportation, and even entertainment. For example, AI is used in stock market analysis to make predictions and inform investment decisions. In healthcare, AI can help doctors make accurate diagnoses by analyzing patient data and medical records. In transportation, self-driving cars use AI to navigate and make decisions on the road.

However, AI is not without its limitations. One of the biggest concerns surrounding AI is the potential loss of jobs. As machines become more intelligent and capable of performing tasks that were once done by humans, there is a fear that many jobs will become obsolete. This is a valid concern, and it is essential for society to find ways to adapt and train individuals for new job opportunities.

Another limitation of AI is the potential for biased decision making. Since AI systems learn from the data they are fed, if the data is biased, it can lead to biased decisions. For example, if an AI system is trained on data that is predominantly male, it may not be able to accurately recognize and cater to the needs of female users.

In conclusion, AI is a complex and rapidly evolving field that has the potential to bring about significant changes in our society. It is a powerful tool that can help us make more informed decisions, increase efficiency, and improve our lives. However, it is essential to use it responsibly and address its limitations to ensure that it benefits everyone. As technology continues to advance, it is crucial for us to understand and embrace AI to stay updated and make the most of its potential.

The rest is here:
Can you explain the artificial intelligence? | by Vikram gounder | Feb, 2024 - Medium

Read More..

Secure your machine learning models with these MLSecOps tips – TechTarget

In recent years, organizations have moved quickly to integrate new AI technology into their business processes, from basic machine learning models to generative AI tools like ChatGPT. Despite offering numerous business advantages, however, the integration of AI expands organizations' attack surfaces.

Threat actors are constantly looking for ways to infiltrate target IT environments, and AI-powered tools can become another entry point to exploit. AI security strategies are essential to safeguard company data from unauthorized access.

MLSecOps is a framework that brings together operational machine learning (ML) with security concerns, aiming to mitigate the risks that AI/ML models can bring to an organization. MLSecOps focuses on securing the data used to develop and train ML models, mitigating adversarial attacks against those models, and ensuring that developed models comply with regulatory compliance frameworks.

ML models can help organizations increase efficiency by automating repetitive tasks, improving customer service, reducing operational costs and maintaining competitive advantages.

But ML adoption also introduces risks at different points, including during the development and deployment phases, especially when using open source large language models (LLMs). The following are among the most significant risks:

The term MLOps refers to the process of operationalizing ML models in production. It involves several phases:

MLSecOps, therefore, is a natural extension of MLOps. Similar to how DevOps evolved into DevSecOps by integrating security practices into the software development lifecycle, MLSecOps ensures that ML models are developed, tested, deployed and monitored using security best practices.

MLSecOps integrates security practices throughout the ML model development process. This integration ensures the security of ML models in two areas:

MLSecOps specifically focuses on the security issues related to ML systems. The following are the five main security pillars that MLSecOps addresses.

Like other software tools, ML systems frequently use components and services from various third-party providers, creating a complex supply chain. A security vulnerability in any component across the ML system supply chain could allow threat actors to infiltrate it and conduct various malicious actions.

Typical supply chain elements for an ML system include the following:

The U.S. was the pioneer in addressing the security aspects related to the software supply chain. In 2021, the Biden administration issued Executive Order 14028, which requires all organizations in both public and private sectors to address security vulnerabilities in their supply chain.

Model provenance is concerned with tracking an ML system's history through development, deployment, training, testing, and monitoring and usage. Model provenance helps security auditors identify who made specific changes to the model, what those changes were and when they occurred.

Some elements included in the model provenance of an ML system include the following:

Model provenance is essential to comply with the various data protection compliance regulations, such as the GDPR in the European Union, HIPAA in the United States and industry-specific regulations such as the Payment Card Industry Data Security Standard.

Governance, risk and compliance (GRC) frameworks are used within organizations to meet government and industry-enforced regulations. For ML systems, GRC spans several elements of MLSecOps, with the primary aim of ensuring that organizations are using AI tools responsibly and ethically. As more organizations build AI-powered tools that rely on ML models to perform business functions, there is a growing need for robust GRC frameworks in the use and development of ML systems.

For instance, when developing an ML system, organizations should maintain a list of all components used in development, including data sets, algorithms and frameworks. This list is now known as the machine learning bill of materials (MLBoM). Similar to a software bill of materials for software development projects, MLBoMs document all components and services used to create AI tools and their underlying ML models.

Trusted AI deals with the ethical aspects of using AI tools for different use cases as more organizations rely on AI tools to perform job functions, including critical ones. There is an emerging need to ensure that AI tools and their underlying ML models are giving ethical responses and are not biased in any way towards characteristics like race, gender, age, religion, ethics or nationality.

One method to check the fairness of AI tools is to request that they explain their answers. For instance, if a user asks a generative AI tool to recommend the best country to visit in summer, the model should provide a justification for its answer. This explanation helps humans understand what factors influenced the AI tool's decision.

Adversarial machine learning is concerned with studying how threat actors can exploit ML systems in various ways to conduct malicious actions. There are four primary types of adversarial ML:

ML development teams can use the MLSecOps methodology efficiently to mitigate cyberattacks when developing ML models. The following are some MLSecOps best practices:

Nihad A. Hassan is an independent cybersecurity consultant, an expert in digital forensics and cyber open source intelligence, and a blogger and book author. Hassan has been actively researching various areas of information security for more than 15 years and has developed numerous cybersecurity education courses and technical guides.

Here is the original post:
Secure your machine learning models with these MLSecOps tips - TechTarget

Read More..

In the AI science boom, beware: your results are only as good as your data – Nature.com

Hunter Moseley says that good reproducibility practices are essential to fully harness the potential of big data.Credit: Hunter N.B. Moseley

We are in the middle of a data-driven science boom. Huge, complex data sets, often with large numbers of individually measured and annotated features, are fodder for voracious artificial intelligence (AI) and machine-learning systems, with details of new applications being published almost daily.

But publication in itself is not synonymous with factuality. Just because a paper, method or data set is published does not mean that it is correct and free from mistakes. Without checking for accuracy and validity before using these resources, scientists will surely encounter errors. In fact, they already have.

In the past few months, members of our bioinformatics and systems-biology laboratory have reviewed state-of-the-art machine-learning methods for predicting the metabolic pathways that metabolites belong to, on the basis of the molecules chemical structures1. We wanted to find, implement and potentially improve the best methods for identifying how metabolic pathways are perturbed under different conditions: for instance, in diseased versus normal tissues.

We found several papers, published between 2011 and 2022, that demonstrated the application of different machine-learning methods to a gold-standard metabolite data set derived from the Kyoto Encyclopedia of Genes and Genomes (KEGG), which is maintained at Kyoto University in Japan. We expected the algorithms to improve over time, and saw just that: newer methods performed better than older ones did. But were those improvements real?

Scientific reproducibility enables careful vetting of data and results by peer reviewers as well as by other research groups, especially when the data set is used in new applications. Fortunately, in keeping with best practices for computational reproducibility, two of the papers2,3 in our analysis included everything that is needed to put their observations to the test: the data set they used, the computer code they wrote to implement their methods and the results generated from that code. Three of the papers24 used the same data set, which allowed us to make direct comparisons. When we did so, we found something unexpected.

It is common practice in machine learning to split a data set in two and to use one subset to train a model and another to evaluate its performance. If there is no overlap between the training and testing subsets, performance in the testing phase will reflect how well the model learns and performs. But in the papers we analysed, we identified a catastrophic data leakage problem: the two subsets were cross-contaminated, muddying the ideal separation. More than 1,700 of 6,648 entries from the KEGG COMPOUND database about one-quarter of the total data set were represented more than once, corrupting the cross-validation steps.

NatureTech

When we removed the duplicates in the data set and applied the published methods again, the observed performance was less impressive than it had first seemed. There was a substantial drop in the F1 score a machine-learning evaluation metric that is similar to accuracy but is calculated in terms of precision and recall from 0.94 to 0.82. A score of 0.94 is reasonably high and indicates that the algorithm is usable in many scientific applications. A score of 0.82, however, suggests that it can be useful, but only for certain applications and only if handled appropriately.

It is, of course, unfortunate that these studies were published with flawed results stemming from the corrupted data set; our work calls their findings into question. But because the authors of two of the studies followed best practices in computational scientific reproducibility and made their data, code and results fully available, the scientific method worked as intended, and the flawed results were detected and (to the best of our knowledge) are being corrected.

The third team, as far as we can tell, included neither their data set nor their code, making it impossible for us to properly evaluate their results. If all of the groups had neglected to make their data and code available, this data-leakage problem would have been almost impossible to catch. That would be a problem not just for the studies that were already published, but also for every other scientist who might want to use that data set for their own work.

More insidiously, the erroneously high performance reported in these papers could dissuade others from attempting to improve on the published methods, because they would incorrectly find their own algorithms lacking by comparison. Equally troubling, it could also complicate journal publication, because demonstrating improvement is often a requirement for successful review potentially holding back research for years.

So, what should we do with these erroneous studies? Some would argue that they should be retracted. We would caution against such a knee-jerk reaction at least as a blanket policy. Because two of the three papers in our analysis included the data, code and full results, we could evaluate their findings and flag the problematic data set. On one hand, that behaviour should be encouraged for instance, by allowing the authors to publish corrections. On the other, retracting studies with both highly flawed results and little or no support for reproducible research would send the message that scientific reproducibility is not optional. Furthermore, demonstrating support for full scientific reproducibility provides a clear litmus test for journals to use when deciding between correction and retraction.

Now, scientific data are growing more complex every day. Data sets used in complex analyses, especially those involving AI, are part of the scientific record. They should be made available along with the code with which to analyse them either as supplemental material or through open data repositories, such as Figshare (Figshare has partnered with Springer Nature, which publishes Nature, to facilitate data sharing in published manuscripts) and Zenodo, that can ensure data persistence and provenance. But those steps will help only if researchers also learn to treat published data with some scepticism, if only to avoid repeating others mistakes.

Link:
In the AI science boom, beware: your results are only as good as your data - Nature.com

Read More..

Machine learning tool unveils promising drugs to minimise harmful scarring post heart attack – Open Access Government

This cutting-edge computer model promises a breakthrough in cardiac care and also holds the potential to revolutionise drug discovery for various complex diseases.

The research led by computational biologist Dr. Anders R. Nelson and Dr. Jeffrey J. Saucerman from UVAs Department of Biomedical Engineering combines decades of human knowledge with a new approach called logic-based mechanistic machine learning.

The interdisciplinary team aimed to understand better how drugs impact fibroblasts, cells crucial for heart repair but also known for causing harmful scarring known as fibrosis.

Many common diseases such as heart disease, metabolic disease, and cancer are complex and hard to treat, explains Dr. Nelson. Machine learning helps us reduce this complexity, identify the most important factors contributing to the disease, and better understand how drugs can modify diseased cells.

Unlike previous attempts that focused on specific aspects of fibroblast behaviour, the UVA researchers utilised their innovative machine-learning model to predict the effects of 13 promising drugs on human fibroblasts.

The model identified a potential candidate to prevent scarring and explained how it works. This dual capability is crucial for designing effective clinical trials and understanding potential side effects.

One standout discovery from the study is the potential of the drug pirfenidone, which is already FDA-approved for idiopathic pulmonary fibrosis. The model revealed a new explanation of how pirfenidone suppresses contractile fibres inside fibroblasts, contributing to the hearts stiffening.

The model predicts the effects of an experimental Src inhibitor, WH4023, on another type of contractile fibre, a finding experimentally validated with human cardiac fibroblasts.

While future research is required to validate the efficacy of these drugs in animal models and human patients, the UVA team is optimistic about the transformative potential of mechanistic machine learning.

Dr. Saucerman emphasises, Were looking forward to testing whether pirfenidone and WH4023 also suppress fibroblast contraction in scars in preclinical animal models. We hope this provides an example of how machine learning and human learning can work together to not only discover but also understand how new drugs work.

Read more:
Machine learning tool unveils promising drugs to minimise harmful scarring post heart attack - Open Access Government

Read More..

Machine Learning APIs on the Horizon – Digital Engineering 24/7

Last November, the Autodeskcrowd returned to familiar grounds: the Venetian in Las Vegas for the annual Autodesk University. In the convention center a few corridors away from the slot machines, roulette wheels and Blackjack tables, Autodesk CEO Andrew Anagnost decided it was time to show his hands.

For better or worse, AI[artificial intelligence]has arrived, with all the looming implications for all of us, he said.Weve been working to get you excited about AI for years. But now were moving from talking about it to actually changing your businesses.

Autodesks big bet comes in the form of Autodesk AI, based in part on the technology from BlankAI it acquired. The implementation will lead to 3D models that can be rapidly created, explored and edited in real time using semantic controls and natural language without advanced technical skills, the company announced.

More details come from Jeff Kinder, executive vice president of product development and manufacturing. Debuting in our automotive design studio next year, BlankAI will allow you to pull from your historical library of design work, then use it to rapidly generate new concepts that build on your existing design style, he said. This is where general AI ends, and personalized machine learning begins.

The ability to use natural language to generate 3D assets, as described by Autodesk in its announcement, would be a big achievement itself. But the next step is more tantalizing. The initial AI will no doubt be trained based on publicly available design data and Autodesk data. But the company also revealed it planned to give users a way to further refine the algorithms using their own proprietary data.

At some point, as some of these capabilities get to a level of automation, we may actually license a model to a particular customer, and they can train and improve that model on top of their capabilities. Thats the business model that Microsoft is using right now for some of their tools. I think its a very robust model, said Anagnost during an industry pressQ&A.

Although initially debuting in Autodesks automotive portfolio, Autodesk ultimately aims to include these capabilities as part of its Fusion industry cloud, said Stephen Hooper, Autodesk vice president of design and manufacturing. If the algorithm is trained on your historical data, it understands your design cues, styling cues and brand identity, making it much more helpful in generating your preferred designs.While its clearly on Autodesks roadmap, the exact mechanism remains unclear. Were still evaluating how and when we might provide a private model, Hooper said.

Last year, at PCB West in Santa Clara, CA, Kyle Miller,research and product manager at Zuken, unveiled a new offering from Zuken for its CR-8000printed circuit boarddesign software: the Autonomous Intelligent Place and Route (AIPR). Miller pointed out AI-optimized layout tends to be cleaner, simpler, with fewer clashes, because the software could process complex design hierarchy and signal clusters much better than humans could. What takes Autorouter [another product] a set-up time of 30 minutes and auto-routing time of 15 minutes, might just take AIPR 30 seconds, he said.

AIPR is just a launchpad for the PCB software makers long-term goal. The next step, according to Miller, is to apply machine learning to all the PCB designs available in Zukens library. The outcome is what the company calls the Basic Brain, which enhances the user experience by routing the design utilizing the Smart Autorouter based on learned approaches and strategies.

After that, Zuken plans to offer a tool that its customers can use to apply machine learning to their own library. The company calls it the Dynamic Brain, which learns from your PCB designers, utilizing past design examples and integrating them into AI algorithms. Ultimately, the goal is the Autonomous Brain, an AI-driven powerhouse in continuous learning mode, pushing the boundaries of creativity.

Zukens roadmap is a multiyear roadmap; therefore, Dynamic Brain is not expected to show up in the portfolio soon. Our first goal is to make the base productthe Basic Brainas capable as possible before delivering the Dynamic Brain, Miller said.

The plan to let customers use the AI tool to ingest proprietary data also invites certain questions about security. It has been very important to Zuken from the beginning of this process that we have no internal access to any customer data All training of the Dynamic Brain will be done on the customer site. We have no plan to use cloud-based services for this (unless specifically agreed with a customer). No data is shared with Zuken servers, said Miller.

The training will be done via the Zuken CR-8000 platform, Miller explained. This communicates on the local network only (or wider network if the customer has a secure multisite network) with the AIPR server, which handles the training, he added.

On the first day he took the job as Ansyss CTO, Prith Banerjee decided he was going to focus on AI-powered simulation.In 2018, we started investing in it, specifically to explore opportunities in two areas: Can AI or ML make simulation faster? Can it make simulation easier to use? he asked.

The investment appears to be bearing fruit. Last October, Ansys launched Ansys SimAI, described as a cloud-enabled, physics-neutral platform that will empower users across industries to greatly accelerate innovation and reduce time to market.

Once trained with customer data, Ansys SimAI predicts simulation results, which align with AnsysFluentcalculations. However, Ansys SimAI takes a mere 5 minutes.Image courtesy of Ansys.

The software is a departure from typical simulation products. Its better to think of it as a way to use AI to train the software to develop good finite element analysis (FEA) instincts.

SimAI is a giant leap forward compared to our previous technology, in that the users do not need to parametrize the geometry, said Banerjee.

You feed the software a set of simulation results, then let the AI-like software learn the correlations between the topology and the outcomes.The users can take their simulation results, upload them to SimAI and train the model. After uploading the data, the users select which variables they are interested in and how long they are willing to wait for the training to complete. Once the training is done, they can upload new geometries and make predictions, explained Banerjee.

This is similar to how, over time, a veteranFEAuser learns to anticipate certain stress distributions, deformation and airflow patterns based on the designs topology. Except, with high-performance computing (HPC), the software can learn in a few hours what would have taken a human months or years to learn. But with HPC comes the need to rely on the cloudAnsys cloud, based on Amazon Web Services (AWS) infrastructure.

AWS infrastructure provides state-of-the-art security and is used by many security-sensitive customers spanning defense, healthcare and financial organizations, Banerjee said. In 2018, Ansys launched Ansys Discovery, a fast-paced simulation tool targeting the designers. Since then, cloud has become an integral part of the companys strategy and offerings.

The personalization of machine learning is a trend that weve been seeing in the last five years, said Johanna Pingel, AI product marketing manager at MathWorks. Essentially, you start with out-of-the-box algorithms, but then you want to incorporate your own engineering data, she added.

MathWorks offers MATLAB and Simulink. MATLABapps let you see how different algorithms work with your data. Iterate until youve got the results you want, then automatically generate a MATLAB program to reproduce or automate your work, according to the company.

Once you have an executable program, you may deploy it in Simulink to build system models to conduct what-if analyses.

Suppose youre an autonomous vehicle developer. Its relatively easy to develop or find an out-of-the-box lane-detection algorithm, Pingel pointed out.

But thats just the starting point. You may want to refine it to work for nighttime, or for the UK, where the drivers drive on the left. Without such refinement options, the algorithms scope will likely be too broad to be effective for your enterprises specific needs, she said.

MATLAB is ideal for such training, according to Pingel. She explained, You can import your data through a[graphical user interface (GUI)], and train a model through a GUI. You use a low-code, app-based workflow for training models.

MathWorks has also taken note of its own customers growing interest in ChatGPT-like interactions. In response, in November 2023, the company launched the MATLAB AI Chat Playground, trained on ChatGPT. It appears as a chat panel in MATLAB, allowing users to query the software using natural language. However, the tool is experimental and still evolving, Pingel cautioned.

Although natural language-based input might make engineering tools more accessible,Pingelpointed out the domain knowledge and expertise of the human still remains essential in crafting the input and assessing the output.

Engineers must use their inherent knowledge of the problem when theyre talking to the software about the kind of structural capabilities they want. They have to bring that to the table when theyre using generative AI, she said.

Former SolidWorks CEO and Onshape cofounder John McEleney warned, Im not dismissing the technology, but theres a lot of AI washing happening. Everyone wants to jump on the AI bandwagon with AI this, AI that.

For AI training to be reliable, the sample data pool has to be large enough to represent a rich variety of scenarios.

The question is, do you have enough models to train your AI engine? he asked. If youre a large automotive or aerospace company, sure. But for most midsize manufacturers, maybe not. If your training is based on 50 to 100 models, are you reaching a critical mass? he asked.

McEleney revealed Onshape is currently exploring some internal models to gain insights. It would be logical and reasonable to assume that design assistant-type suggestions will be how we would introduce these features, he said.

Considering how speaking to AI chatbots such as Siri on smartphones has become the norm, McEleney said, You can imagine being able to tell your software, Go do this, and the system being able to find samples from your previous work to execute it for you.

He also foresees users being highly protective of their proprietary data even if they want to benefit from AI training.

So I can see that, at least in the beginning, people will want to do that type of training internally, he added.

Most people would like access to others data, because a larger sample pool makes the AI algorithm more reliable. But the same people are also highly protective of their proprietary data, because it contains IP that gives them a competitive advantage. Thats the dilemma of the AI era.

View original post here:
Machine Learning APIs on the Horizon - Digital Engineering 24/7

Read More..