Category Archives: Machine Learning

Not familiar with BizarroMath? Youre in luck; Ive launched a web site for it where you can get an explanation of the numbers and browse the data.

Week 1

Week 2

Week 3

All lines courtesy of DraftKings Sportsbook as of 8:00am, Monday, September 19, 2022.

Iowa football continues to be the #1 supplier of high-quality material for the Sickos Committee.

This week, BizarroMath went 4-8 ATS and 5-7 O/U. Combined with the prior record of 11-8 and 6-13, respectively, the algorithm is 15-16 ATS and 11-20 O/U on the season after three full weeks of play. Not a great outing in the second straight strange week of Division I NCAA Football, but were still learning about these teams.

Vegas Says: MI -46.5, U/O 57.5

BizarroMath Says: MI -64.10 (MI cover), 71.66 (over)

Actual Outcome: MI 59, UCONN 0 (ATS hit, O/U hit)

One Sentence Recap: Michigan aint played nobody.

Vegas Says: OU -11.5, O/U 64.5

BizarroMath Says: OU -7.90 (NE cover), 60.03 (under)

Actual Outcome: OU 49, NE 14 (ATS miss, O/U hit)

One Sentence Recap: We should all be rooting for Mickey Josephs no-nonsense, just play the damn game style, which is a welcome departure from Scott Frosts chesty preening, but Nebraska still seems mired in a deep hole of undisciplined play and softness at the point of attack.

Vegas Says: n/a

BizarroMath Says: n/a

Actual Outcome: SILL 31, NU 24

One Sentence Recap: I watched the Salukis play many at a game at the UNI Dome in Cedar Falls over the years and I was probably less surprised than many that they pulled off this upset.

Vegas Says: Pk, O/U 58.5

BizarroMath Says: PUR -2.22 (Purdue win), O/U 47.9 (under)

Actual Outcome: SYR 32, PUR 29 (ATS miss, O/U miss)

One Sentence Recap: Much like the Penn State game, this game was there for the taking and Purdue simply refused, and I want to reiterate that Ive been skeptical since before the season began that the 2022 Edition of Purdue would be able to maintain the momentum from last year.

Vegas Says: IN -6.5, O/U 59.0

BizarroMath Says: WKY -12.90 (WKY cover), 65.99 (over)

Actual Outcome: IND 33, WKY 30 (ATS hit, O/U hit)

One Sentence Recap: BizMas prediction of a WKY upset damn near came true, but Tom Allens sweeping, must win now changing in the offseason seem to be paying dividends as the Hoosiers are figuring some things out and finding ways to win.

Vegas Says: RUT -17.5, O/U 44

BizarroMath: RUT -23.74 (RUT cover), 44.62 (over)

Actual Outcome: RUT 16, TEM 14 (ATS miss, O/U miss)

One Sentence Recap: Im pretty sure Rutgers is close to its pre-season O/U win total already, as the Scarlet Knights, like the other Eastern Red Team, keep finding ways to win.

Vegas Says: PSU -3, O/U 49

BizarroMath: PSU -2.71 (Auburn cover), 44.70 (under)

Actual Outcome: PSU 41, AUB 12 (ATS miss, O/U miss)

One Sentence Recap: It just means more.

Vegas Says: MN -27.5, O/U 46.5

BizarroMath: MN -23.95 (CO cover), 44.20 (under)

Actual Outcome: MN 49, CO 7 (ATS miss, O/U miss)

One Sentence Recap: Minnesota aint played nobody.

Vegas Says: WI -37.5, O/U 46.5

BizarroMath: WI -38.71 (WI cover), 50.1 (over)

Actual Outcome: WI 66, NMSU 7 (ATS hit, O/U hit)

One Sentence Recap: Theres nothing interesting about this game other than two observations: (1) this is the most points Wisconsin has scored in the Paul Chryst era; (2) Wisconsin has the same problem as Iowa in that Chryst has probably hit his ceiling and isnt going to elevate the program any further, but he wins too much to let him go.

Vegas Says: OSU -31.5, O/U 61

BizarroMath: OSU -28.36 (Toledo cover), 66.12 (over)

Actual Outcome: OSU 77, TOL 21 (ATS miss, O/U hit)

One Sentence Recap: OSUs opponent-adjusted yards surrendered this year is an absurd 2.28, which could be more a function of the small sample size we have for their opponents than anything, but this is why I blend data, folks.

Vegas Says: -MSU 3, O/U 57.5

BizarroMath: MSU -8.44 (MSU cover), 50.28 (under)

Actual Outcome: WA 39, MSU 28 (ATS miss, O/U miss)

One Sentence Recap: Ive told you my numbers dont like the Spartans, and Washington just showed us why.

Vegas Says: IA -23, O/U 40

BizarroMath: IA -2.48 (Nevada cover), 47.22 (over)

Actual Outcome: IA 27, NE 0 (ATS miss, O/U miss)

One Sentence Recap: Weird how when you inject a bunch of scholarship players back into your line-up, and play a defense of dubious quality, you can kind of, sort of, move the ball a little bit, even with an historically incompetent offense.

Vegas Says: MD -3.5, O/U 69.5

BizarroMath: SMU -1.31 (SMU cover), 75.22 (over)

Actual Outcome: MD 34, SMU 27 (ATS hit, O/U miss)

One Sentence Recap: First Four Games Maryland has scored 121 points through 3 games; I put the O/U on how many more games before Iowa breaks that mark at 5.5.

Now that I have the http://www.BizarroMath.com web site up and running, you can take a look at Iowas game-by-game projections and season projections yourself. Im going to not post the images this week and leave it to you to visit the site if you want to see the data. This is not a clickbait money scheme. There are no ads on that site, I wrote the HTML by hand because Im old and thats how I roll, and I make $0 off you visiting that site.

If you prefer to have the data presented in-line here, let me know, I will do that next week. Please answer the poll below to help me figure out how best to do this.

5%

21%

54%

13%

4%

1%

Also Caveat: If you come back to these links in the future, they will be updated with the results of future games, which also is a reason to post the data here for posterity, I suppose. Anyway, I may change the web site in the future to provide week-by-week updates showing the net changes. If youre interested in that, please let me know.

On to the analysis.

We finally have two FBS games worth of data on Iowa and can we start jumping to conclusions. Iowas raw PPG against D1 competition are at 17.0, which is good for #108 in the country. Iowas raw YPG are 243.50, which puts the Hawkeyes at #115. Iowas raw YPP are 4.20, ranking the Black and Old Gold at #110. The team is very slowly crawling out of the Division 1 cellar, but didnt exactly light the world on fire Saturday in a wet, frequently-interrupted outing against a Nevada team widely regarded as being Not Very Good.

We dont have enough data for opponent-adjustments for Iowa at this point (I require at least 3 adjustable games). Iowas blended data is what is used for the projections, and you can review that on the BizarroMath.com web site. Suffice it to say that Iowas outing against Nevada was similar in profile to what the team looked like last year. But, the schedule is a bit tougher this year, and Iowa needed some good fortune last year to make the Big 10 Championship game. I know nobody wants to hear it, but if this offense can climb up out of the triple digit rankings and get even to the 80th-90th range, that just might be enough to stay in the conference race.

But this season may simply boil down to schedule. Wisconsins cross-over games are @Ohio State, @MSU, and Maryland at home. Thats about as hard as it gets without playing either Michigan or Penn State. Minnesotas cross-over games are @Penn State, @Michigan State, and Rutgers. Iowas are @Rugers, Michigan, @Ohio State. From most to least difficult, Id say Iowa has the worst draw, then Wisconsin, then Minnesota. The Gophers also get Purdue and Iowa at home, and have Nebraska, Wisconsin, and Illinois on the road. The Badgers have Illinois and Purdue at home and go on the road to play Nebraska, Iowa, and (dont laugh) Northwestern. The Badgers are 1-6 at Northwestern this Century. The schedule generally favors the Gophers, and with Iowa playing Michigan and Ohio State in October, we shouldnt be surprised if the Hawkeyes are out of the division race before November.

That said, Iowas game-by-game odds are moving in the right direction. Iowa is a significant underdog vs. Michigan and Ohio State as expected, and a slight dog to Wisconsin and (stop traffic) Illinois. Perhaps most alarming is that the Hawkeyes have only a 37.92% chance to beat Minnesota. But! Recall that I am not doing opponent adjustments to the 2022 data yet for Minnesota, so their gaudy numbers are being taken at face value, and theyll drop after the Gophers play Michigan State this weekend.

To give you an idea of how that works, consider Michigan, which has played enough adjustable games that I can run opponent adjustments. Their opposition has been so terrible that BizarroMath discounts the Wolverines raw 55.33 PPG by a whopping 22.54 points. This means that this Wolverine team is expected to put up just 32.80 points against an average D1 defense, to say nothing of what they can do against a Top 5 defense, which Iowa just so happens to have (again, before opponent-adjustments). Michigans adjusted data is thus actually worse than last year, whose offense was, opponent-adjusted, worth 42.17 PPG.

Minnesotas adjustments will come soon enough, and well see them return to deep below the Earth, where filthy rodents belong. But, so, too, will Iowas, and of Iowas three adjustable opponents after this coming weekend - Rutgers, Nevada, and Iowa State - the Cyclones are by far the best team.

Iowa Season Projections

The Nevada win and swing in the statistics towards something more similar to last years putrid but still-better-than-this-crap offensive performance has brightened Iowas season outlook somewhat. Iowas most likely outcome now is 7-5 (27.13% chance), with 6-6 being more likely (25.89%) than 8-4 (17.52%). There is a 92.11% chance that Iowa doesnt reach 9.3, and a 78.42% chance that the Hawkeyes get bowl eligible this year.

The Gilded Rodents flashy numbers have pulled them almost even with Wisconsin, as the Badgers and Gophers are both in the 35-40% range for a division championship. Purdues continued struggles drops the Boilermakers to the four spot, elevating hapless Iowa to the third place in the West, though the Hawkeyes chances of actually winning the damn thing drop to 8.40%, Iowas climb up the division ladder from 5th to 3rd is more a function of the poor play of the teams now ranked lower than anything Iowa is doing on the field.

Im a bit puzzled by the conference race in the East, where Ohio State shot from last weeks 21.53% to this weeks 64.18% chance, but I think its because BizMa now has the Buckeyes with a 77.74% chance of winning The Game, which is the main shift that accounts for this change. Why? Well, this week we have opponent-adjustments for both teams and OSU has played a tougher schedule, so the Buckeyes numbers are not being discounted nearly as much as Michigans.

For example, on offense, OSU is putting up a raw 8.49 YPP, which BizMa is actually adjusting up to 9.58. Michigan, by comparison, is putting up 7.97 YPP, but BizMa is adjusting it down to 6.36 YPP based on the competition. As we move into the conference slate and the quality of each teams opposition evens out, well probably see those numbers flatten out a bit.

I love week 4. Because the number of games I have to track is cut in half.

Vegas Says: n/a

BizarroMath: n/a

One Sentence Prediction: Your Fighting Illini are going to be 3-1 going into conference play, and they have been competitive, if a bit raggety.

Vegas Says: MI -17, O/U 62.5

BizarroMath: MI -3.81, O/U 58.01 (MD cover, under)

One Sentence Prediction: BizMa sees this game as being much closer than Vegas does, and I think the difference might be a function of where we are in the season, as I just dont see Marylands defense holding Michigan down, and I dont buy that under for even a second, folks.

Vegas Says: PSU -26, O/U 60.5

BizarroMath: PSU -30.47, O/U 58.97 (PSU cover, under)

One Sentence Prediction: I dont know a thing about Central Michigan this year but a final along the lines of 45-13 sounds about right.

Vegas Says: MN -2, O/U 51.0

BizarroMath: MN -8.75, O/U 45.49 (MN cover, under)

One Sentence Prediction: Well soon know if the Gilded Rodents are fools gold, but not this week, as I think Minnesota is going to put up some points here in something like a 42-23 affair.

Vegas Says: CIN -15.5, O/U 54.0

BizarroMath: CIN -25.04, O/U 53.90 (CIN cover, under)

One Sentence Prediction: The Hoosiers either crash hard back down to Terra Firma in an embarrassing road rout, or this winds up being an unexpectedly knotty game.

Vegas Says: IA -7.5, O/U 35.5

BizarroMath: IA -9.85, O/U 32.09 (IA cover, under)

One Sentence Prediction: In Assy Football, the MVP is from one of two separate, yet equally important, groups: the punt team, which establishes poor field position for the opposition; and the punt return team, who try to field the ball outside of the 15 yard line without turning it over; this is their magnum opus.

Vegas Says: OSU -17.5, O/U 56.5

Read more:
Machine Learning Week 4 - Updated Iowa Game by Game Projections, Season Record, and Championship Odds - Black Heart Gold Pants

image:The Archytas Dome region of the lunar surface is the target area for the EXPLORE Lunar Data Challenges 2022. view more

Credit: Credit: NASA/GSFC/Arizona State University/EXPLORE/Jacobs University.https://exploredatachallenges.space/wp-content/uploads/2022/09/Archytas2.png

Lunar enthusiasts of all ages are challenged to help identify features on the Moon that might pose a hazard to rovers or astronauts exploring the surface.

The2022 EXPLORE Lunar Data Challengeis focused on theArchytas Dome region, close to the Apollo 17 landing site where the last humans set foot on the Moon 50 years ago this December.

The Machine Learning Lunar Data Challenge is open to students, researchers and professionals in areas related to planetary sciences, but also to anyone with expertise in data processing. There is also a Public Lunar Data Challenge to plot the safe traverse of a lunar rover across the surface of the Moon, open to anyone who wants to have a go, as well as a Classroom Lunar Data Challenge for schools, with hands-on activities about lunar exploration and machine learning.

Announcing the EXPLORE Machine Learning Lunar Data Challenge during the Europlanet Science Congress (EPSC) 2022 in Granada, Spain, this week Giacomo Nodjoumi said: The Challenge uses data of the Archytas Dome taken by the Narrow Angle Camera (NAC) on the Lunar Reconnaissance Orbiter (LRO) mission. This area of the Moon is packed craters of different ages, boulders, mounds, and a long, sinuous depression, or rille. The wide variety of features in this zone makes it a very interesting area for exploration and the perfect scenario for this Data Challenge.

The Machine Learning Lunar Data Challenge is in three steps: firstly, participants should train and test a model capable of recognising craters and boulders on the lunar surface. Secondly, they should use their model to label craters and boulders in a set of images of the Archytas zone. Finally, they should use the outputs of their models to create a map of an optimal traverse across the lunar surface to visit defined sites of scientific interest and avoid hazards, such as heavily cratered zones.

The public and schools are also invited to use lunar images to identify features and plot a journey for a rover. Prizes for the challenges include vouchers totalling 1500 Euros, as well as pieces of real Moon rock from lunar meteorites.

The EXPLORE project, which is funded through the European Commissions Horizon 2020 Programme, gathers experts from different fields of science and technical expertise to develop new tools that will promote the exploitation of space science data.

Through the EXPLORE Data Challenges, we aim to raise awareness of the scientific tools that we are developing, improve their accuracy by bringing in expertise from other communities, and involve schools and the public in space science research, said Nick Cox, the Coordinator of the EXPLORE project.

The deadline for entries closes on 21 November 2022 and winners will be announced in mid-December on the anniversaries of the Apollo 17 mission milestones.

The 2022 EXPLORE Data Challenges can be found at:https://exploredatachallenges.space

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.

Read more from the original source:
Join the challenge to explore the Moon! - EurekAlert

Artificial intelligence and ML are making considerable inroads in finance. They are the critical aspect of variousfinancial applications, including evaluating risks, managing assets, calculating credit scores, and approving loans.

Businesses use AI and ML:

Taking the above points into account, its no wonder that companies like Forbes and Venture beat are usingAI to predict the cash flow and detect fraud.

In this article, we present the financial domain areas in which AI and ML have a more significant impact. Well also discuss why financial companies should care about and implement these technologies.

Machine learning is a branch of artificial intelligence that allows learning and improvement without any programming. Simply put, data scientists train the MI model with existing data sets and automatically adjust its parameters to improve the outcome.

According to Statista, digital payments are expected to show an annual growth rate of 12.77% and grow to 20% by 2026. This vast number of global revenues, done online requires an intelligent fraud system.

Source: Mordor Intelligence

Traditionally, to check the authenticity of users, fraud-detection systems analyze websites through factors like location, merchant ID, the amount spent, etc. However, while this method is appropriate for a few transactions, it would not cope with the increased transactional amount.

And, analyzing the surge of digital payments, businesses cant rely on traditional fraud-detection methods to process payments. This gives rise to AI-based systems with advanced features.

An AI and ML-powered payment gateway will look at various factors to evaluate the risk score. These technologies consider a large volume of data (location of the merchant, time zone, IP address, etc.) to detect unexpected anomalies, and verify the authenticity of the customer.

Additionally, the finance industry, through AI, can process transactions in real-time, allowing the payment industry to process large transactions with high accuracy and low error rates.

The financial sector, including the banks, trading, and other fintech firms, are using AI to reduce operational costs, improve productivity, enhance users experience, and improve security.

The benefits of AI and ML revolve around their ability to work with various datasets. So lets have a quick look at some other ways AI and ML are making roads into this industry:

Considering how people invest their money in automation, AI significantly impacts the payment landscape. It improves efficiency and helps businesses to rethink and reconstruct their process. For example, businesses can use AI to decrease the credit card processing (gettrx dot com card processing guide for merchants) time, increase automation and seamlessly improve cash flow.

You can predict credit, lending, security, trading, baking, and process optimization with AI and machine learning.

Human error has always been a huge problem; however, with machine learning models, you can reduce human errors compared to humans doing repetitive tasks.

Incorporating security and ease of use is a challenge that AI can help the payment industry overcome. Merchants and clients want a payment system that is easy to use and authentic.

Until now, the customers have to perform various actions to authenticate themselves to complete a transaction. However, with AI, the payment providers can smooth transactions, and customers have low risk.

AI can efficiently perform high volume; labor-intensive tasks like quickly scrapping data and formatting things. Also, AI-based businesses are focused and efficient; they have minimum operational cost and can be used in the areas like:

Creating more Value:

AI and machine learning models can generate more value for their customers. For instance:

Improved customer experience: Using bots, financial sectors like banks can eliminate the need to stand in long queues. Payment gateways can automatically reach new customers by gathering their historical data and predicting user behavior. Besides, Ai used in credit scoring helps detect fraud activity.

There are various ways in which machine learning and artificial intelligence are being employed in the finance industry. Some of them are:

Process Automation:

Process automation is one of the most common applications as the technology helps automate manual and repetitive work, thereby increasing productivity.

Moreover, AI and ML can easily access data, follow and recognize patterns and interpret the behavior of customers. This could be used for the customer support system.

Minimizing Debit and Credit Card Frauds:

Machine learning algorithms help detect transactional funds by analyzing various data points that mostly get unnoticed by humans. ML also reduces the number of false rejections and improves the real-time approvals by gauging the clients behavior on the Internet.

Apart from spotting fraudulent activity, AI-powered technology is used to identify suspicious account behavior and fraudulent activity in real-time. Today, banks already have a monitoring system trained to catch the historical payment data.

Reducing False Card Declines:

Payment transactions declined at checkout can be frustrating for customers, putting huge repercussions on banks and their reputations. Card transactions are declined when the transaction is flagged as fraud, or the payment amount crosses the limit. AI-based systems are used to identify transaction issues.

The influx of AI in the financial sector has raised new concerns about its transparency and data security. Companies must be aware of these challenges and follow safeguards measures:

One of the main challenges of AI in finance is the amount of data gathered in confidential and sensitive forms. The correct data partner will give various security options and standards and protect data with the certification and regulations.

Creating AI models in finance that provide accurate predictions is only successful if they are explained to and understood by the clients. In addition, since customers information is used to develop such models, they want to ensure that their personal information is collected, stored, and handled securely.

So, it is essential to maintain transparency and trust in the finance industry to make customers feel safe with their transactions.

Apart from simply implementing AI in the online finance industry, the industry leaders must be able to adapt to the new working models with new operations.

Financial institutions often work with substantial unorganized data sets in vertical silos. Also, connecting dozens of data pipeline components and tons of APIS on top of security to leverage a silo is not easy. So, financial institutions need to ensure that their gathered data is appropriately structured.

AI and ML are undoubtedly the future of the financial sector; the vast volume of processes, transactions, data, and interactions involved with the transaction make them ideal for various applications. By incorporating AI, the finance sector will get vast data-processing capabilities at the best prices, while the clients will enjoy the enhanced customer experience and improved security.

Of course, the power of AI can be realized within transaction banking, which sits on the organizations usage. Today, AI is very much in progress, but we can remove its challenges by using the technology. Lastly, AI will be the future of finance you must be ready to embrace its revolution.

Featured Image Credit: Photo by Anna Nekrashevich; Pexels; Thank you!

See more here:
AI and Machine Learning in Finance: How Bots are Helping the Industry - ReadWrite

Participants

This study was conducted as part of the ongoing Study on the Design of a Comprehensive Medical System for Chronic Kidney Disease (CKD) Based on Individual Risk Assessment by Specific Health Examination (J-SHC Study). A specific health checkup is conducted annually for all residents aged 4074years, covered by the National Health Insurance in Japan. In this study, a baseline survey was conducted in 685,889 people (42.7% males, age 4074years) who participated in specific health checkups from 2008 to 2014 in eight regions (Yamagata, Fukushima, Niigata, Ibaraki, Toyonaka, Fukuoka, Miyazaki, and Okinawa prefectures). The details of this study have been described elsewhere11. Of the 685,889 baseline participants, 169,910 were excluded from the study because baseline data on lifestyle information or blood tests were not available. In addition, 399,230 participants with a survival follow-up of fewer than 5years from the baseline survey were excluded. Therefore, 116,749 patients (42.4% men) with a known 5-year survival or mortality status were included in this study.

This study was conducted in accordance with the Declaration of Helsinki guidelines. This study was approved by the Ethics Committee of Yamagata University (Approval No. 2008103). All data were anonymized before analysis; therefore, the ethics committee of Yamagata University waived the need for informed consent from study participants.

For the validation of a predictive model, the most desirable way is a prospective study on unknown data. In this study, the data on health checkup dates were available. Therefore, we divided the total data into training and test datasets to build and test predictive models based on health checkup dates. The training dataset consisted of 85,361 participants who participated in the study in 2008. The test dataset consisted of 31,388 participants who participated in this study from 2009 to 2014. These datasets were temporally separated, and there were no overlapping participants. This method would evaluate the model in a manner similar to a prospective study and has an advantage that can demonstrate temporal generalizability. Clipping was performed for 0.01% outliers for preprocessing, and normalization was performed.

Information on 38 variables was obtained during the baseline survey of the health checkups. When there were highly correlated variables (correlation coefficient greater than 0.75), only one of these variables was included in the analysis. High correlations were found between body weight, abdominal circumference, body mass index, hemoglobin A1c (HbA1c), fasting blood sugar, and AST and alanine aminotransferase (ALT) levels. We then used body weight, HbA1c level, and AST level as explanatory variables. Finally, we used the following 34 variables to build the prediction models: age, sex, height, weight, systolic blood pressure, diastolic blood pressure, urine glucose, urine protein, urine occult blood, uric acid, triglycerides, high-density lipoprotein cholesterol (HDL-C), LDL-C, AST, -glutamyl transpeptidase (GTP), estimated glomerular filtration rate (eGFR), HbA1c, smoking, alcohol consumption, medication (for hypertension, diabetes, and dyslipidemia), history of stroke, heart disease, and renal failure, weight gain (more than 10kg since age 20), exercise (more than 30min per session, more than 2days per week), walking (more than 1h per day), walking speed, eating speed, supper 2h before bedtime, skipping breakfast, late-night snacks, and sleep status.

The values of each item in the training data set for the alive/dead groups were compared using the chi-square test, Student t-test, and MannWhitney U test, and significant differences (P<0.05) were marked with an asterisk (*) (Supplementary Tables S1 and S2).

We used two machine learning-based methods (gradient boosting decision tree [XGBoost], neural network) and one conventional method (logistic regression) to build the prediction models. All the models were built using Python 3.7. We used the XGBoost library for GBDT, TensorFlow for neural network, and Scikit-learn for logistic regression.

The data obtained in this study contained missing values. XGBoost can be trained to predict even with missing values because of its nature; however, neural network and logistic regression cannot be trained to predict with missing values. Therefore, we complemented the missing values using the k-nearest neighbor method (k=5), and the test data were complemented using an imputer trained using only the training data.

The parameters required for each model were determined for the training data using the RandomizedSearchCV class of the Scikit-learn library and repeating fivefold cross-validation 5000 times.

The performance of each prediction model was evaluated by predicting the test dataset, drawing a ROC curve, and using the AUC. In addition, the accuracy, precision, recall, F1 scores (the harmonic mean of precision and recall), and confusion matrix were calculated for each model. To assess the importance of explanatory variables for the predictive models, we used SHAP and obtained SHAP values that express the influence of each explanatory variable on the output of the model4,12. The workflow diagram of this study is shown in Fig.5.

Workflow diagram of development and performance evaluation of predictive models.

See the rest here:
Prediction of mortality risk of health checkup participants using machine learning-based models: the J-SHC study | Scientific Reports - Nature.com

Photo by University Relations

Khoa Luu and Han-Seok Seo

Could artificial intelligence be used to assist with the early detection of autism spectrum disorder? Thats a question researchers at the University of Arkansas are trying to answer. But theyre taking an unusual tack.

Han-Seok Seo, an associate professor with a joint appointment in food science and the UA System Division of Agriculture, and Khoa Luu, an assistant professor in computer science and computer engineering, will identify sensory cues from various foods in both neurotypical children and those known to be on the spectrum. Machine learning technology will then be used to analyze biometric data and behavioral responses to those smells and tastes as a way of detecting indicators of autism.

There are a number of behaviors associated with ASD, including difficulties with communication, social interaction or repetitive behaviors. People with ASD are also known to exhibit some abnormal eating behaviors, such as avoidance of some if not many foods, specific mealtime requirements and non-social eating. Food avoidance is particularly concerning, because it can lead to poor nutrition, including vitamin and mineral deficiencies. With that in mind, the duo intend to identify sensory cues from food items that trigger atypical perceptions or behaviors during ingestion. For instance, odors like peppermint, lemons and cloves are known to evoke stronger reactions from those with ASD than those without, possibly triggering increased levels of anger, surprise or disgust.

Seo is an expert in the areas of sensory science, behavioral neuroscience, biometric data and eating behavior. He is organizing and leading this project, including screening and identifying specific sensory cues that can differentiate autistic children from non-autistic children with respect to perception and behavior. Luu isan expert in artificial intelligence with specialties in biometric signal processing, machine learning, deep learning and computer vision. He will develop machine learning algorithms for detecting ASD in children based on unique patterns of perception and behavior in response to specific test-samples.

The duo are in the second year of a three-year, $150,000 grant from the Arkansas Biosciences Institute.

Their ultimate goalis to create an algorithm that exhibits equal or better performance in the early detection of autism in children when compared to traditional diagnostic methods, which require trained healthcare and psychological professionals doing evaluations, longer assessment durations, caregiver-submitted questionnaires and additional medical costs. Ideally, they will be able to validate a lower-cost mechanism to assist with the diagnosis of autism. While their system would not likely be the final word in a diagnosis, it could provide parents with an initial screening tool, ideally eliminating children who are not candidates for ASD while ensuring the most likely candidates pursue a more comprehensive screening process.

Seo said that he became interested in the possibility of using multi-sensory processing to evaluate ASD when two things happened: he began working with a graduate student, Asmita Singh, who had background in working with autistic students, and the birth of his daughter. Like many first-time parents, Seo paid close attention to his newborn baby, anxious that she be healthy. When he noticed she wouldnt make eye contact, he did what most nervous parents do: turned to the internet for an explanation. He learned that avoidance of eye contact was a known characteristic of ASD.

While his child did not end up having ASD, his curiosity was piqued, particularly about the role sensitivities to smell and taste play in ASD. Further conversations with Singh led him to believe fellow anxious parents might benefit from an early detection tool perhaps inexpensively alleviating concerns at the outset. Later conversations with Luu led the pair to believe that if machine learning, developed by his graduate student Xuan-Bac Nguyen, could be used to identify normal reactions to food, it could be taught to recognize atypical responses, as well.

Seo is seeking volunteers 5-14 years old to participate in the study. Both neurotypical children and children already diagnosed with ASD are needed for the study. Participants receive a $150 eGift card for participating and are encouraged to contact Seo athanseok@uark.edu.

About the University of Arkansas:As Arkansas' flagship institution, the UofA provides an internationally competitive education in more than 200 academic programs. Founded in 1871, the UofA contributes more than$2.2 billion to Arkansas economythrough the teaching of new knowledge and skills, entrepreneurship and job development, discovery through research and creative activity while also providing training for professional disciplines. The Carnegie Foundation classifies the UofA among the few U.S. colleges and universities with the highest level of research activity.U.S. News & World Reportranks the UofA among the top public universities in the nation. See how the UofA works to build a better world atArkansas Research News.

See the article here:
Researchers Using Artificial Intelligence to Assist With Early Detection of Autism Spectrum Disorder - University of Arkansas Newswire

Analysis of structure optimization problem of dual-phase materials

For demonstrating the potential of our framework for the structure optimization of multiphase materials in terms of a target property, a simple sample problem is considered. The sample problem is the structure optimization of artificial dual-phase steels composed of the soft phase (ferrite) and hard phase (martensite). Examples of microstructures are shown in Fig.3. The prepared dual-phase microstructures can be divided into four major categories: laminated microstructures, microstructures composed of rectangle- and ellipse-shaped martensite/ferrite grains, and random microstructures. The size of microstructure images is (128times 128~mathrm {pixels}) and the total number of prepared microstructures is 3824. As an example of a target material property, the fracture strain was selected since fracture behavior is strongly related to the geometry of the two phases. The fracture strain is the elongation of materials at break. As shown in Methodology, the fracture strains for each category were calculated on the basis of the GTN fracture model18,19. Figure4 illustrates the relationship between martensite volume fraction and fracture strain for each category. This shows that laminated microstructures have a relatively high fracture strain. Also, microstructures with a lower martensite volume fraction (higher ferrite volume fraction) possess a higher fracture strain.

Examples of artificial dual-phase microstructures used for training. Black and white pixels respectively correspond to the hard phase (martensite) and soft phase (ferrite). The size of microstructure images is (128times 128) pixels. The dataset can be divided into four major categories. (a) Laminated microstructures. This category only has completely laminated microstructures. (b) Microstructures composed of rectangular martensite grains. This category includes partially laminated structures, such as these shown in the lower left panel. (c) Microstructures composed of elliptical martensite grains. (d) The random microstructures.

Relationship between martensite volume fraction and fracture strain, and examples of microstructures. (a) Plot showing correspondence between martensite volume fraction and fracture strain. (b) Examples of microstructures. Their martensite volume fractions and fracture strains are shown in the plot.

Microstructures generated by the machine learning framework trained by several datasets. (a) Examples of microsturctuers generated for several fracture strains by the network trained using All dataset. (b) Each column corresponds to the microstructures obtained by the models trained using all microstructures, only the random microstructures, only the microstructures composed of ellipse-shaped martensite grains, or only the microstructures composed of rectangle-shaped martensite grains. However, the Rectangle dataset is limited to include only the microstructures whose martensite volume fraction is between 20% and 30%. The given fracture strains are 0.1, 0.3, 0.7, and 0.9 for the All, Random, and Ellipse datasets, and 0.05, 0.1, 0.2, and 0.3 for the Rectangle dataset.

To show the applicability of our framework, we prepared several datasets: all microstructures (All), only random microstructures (Random), only microstructures composed of ellipse-shaped martensite grains (Ellipse), and only microstructures composed of rectangle-shaped martensite grains (Rectangle). Then, we trained VQVAE and PixelCNN using these datasets. The Rectangle dataset is limited to include only the microstructures whose martensite volume fraction is between 20% and 30% to consider the case in which martensite grains are located separately from each other.

Figure5a shows examples of microstructures generated for several fracture strains using the network trained by All dataset. Figure5b summarizes the trend of the microstructures obtained by the networks trained using the above datasets with gradually increasing fracture strain. For the All, Random, and Ellipse datasets, we can see the trend that martensite grains become smaller and thinner as the target fracture strain increases. Since the larger area fraction of the soft phase (ferrite) contributes to the realization of higher elongation as we can see in Fig.4, this result is reasonable. In addition, it should be noted that the laminated structure corresponding to the highest fracture strain ((text {FS}=0.9)) was generated only for the All case in which the laminated structures are included in the training dataset. Additionally, in the case of the controlled martensite volume fraction of the input microstructures (Rectangle), the martensite grains tend to arrange more uniformly as the given fracture strain increases.

Generated microstructures and trend of martensite volume fraction. (a) Microstructures generated at fixed tensile strength and several fracture strains. The tensile strength is set as 700 MPa. The given FSs are 0.1, 0.3, 0.4, 0.5, 0.7, and 0.9. (b) Trend of martensite volume fraction relative to the change in fracture strain. For each fracture strain, the martensite volume fractions of 3000 microstructures generated corresponding to the fracture strain and fixed tensile strength ((700 mathrm {MPa})) were calculated. The black lines and green triangles in the boxes denote median and mean values, respectively.

From these results, we can conclude that there are at least two different strategies for the realization of a higher fracture strain: one is to decrease the size of martensite grains and also to arrange them uniformly, and the other to alternatively make a completely laminated composite structure27. The fact that the laminated structures never appear without providing the laminated structures in the training dataset implies that there exists an impenetrable wall for a simple optimization process, such as a gradient descent algorithm used to train neural networks, to figure out the robustness of laminated structures from the other structures.

Next, the tensile strength is given in addition to the fracture strain as another label for PixelCNN for considering the balance between strength and fracture strain (ductility). In this case, all microstructure data are used for training. The microstructures are generated at the fixed tensile strength of (700 mathrm {MPa}). The generated microstructures are shown in Fig.6a. The laminated structures seem to be dominantly selected as the target fracture strain increases. The trend that martensite grains become smaller and thinner is not seen when the tensile strength is fixed.

In addition, the martensite volume fractions were calculated for 3000 microstructures generated corresponding to several fracture strains. The tensile strength was fixed at (700 mathrm {MPa}) again. The box plot of the trend of the martensite volume fraction relative to the change in fracture strain is shown in Fig.6b. The martensite volume fraction decreases as the given fracture strain increases. At the same time, the martensite volume fraction approaches a constant value. For the realization of a higher ductility without decreasing the tensile strength, the shape of martensite grains approaches that of the laminated structures as the martensite volume fraction decreases. This result implies that laminated structures can achieve a higher tensile strength with a smaller martensite volume fraction. As a result, the laminated structures can be considered as the optimized structures with respect to the shape of martensite grains for the realization of a higher ductility without decreasing their strength. The laminated structures were actually reported to exhibit improved combinations of strength and ductility27.

Correspondence between the target fracture strains given as inputs and the actual fracture strains. For each target fracture strain, 30 microstructures were generated. Then, fracture strains are calculated using the physical model18,19. (a) Plot of relationship. (b) Box plot of relationship. The black lines and green triangles in the boxes denote median and mean values, respectively. (c) Microstructures whose fracture strains are smaller than 20% of the target fracture strains. The values above the panels denote the given target fracture strains (left) and actual fracture strains (right).

To validate the effectiveness of the present framework, fracture strains are calculated using the physical model18,19 for each microstructure obtained using the framework. In this case, the network trained by giving only fracture strain as the target property is used. Figure7a,b show the correspondence between the target fracture strains for generated microstructures and the actual calculated fracture strains. Also, the coefficient of determination was 0.672. It is clear that our framework captures well the general trend of microstructures relative to the fracture strain. However, it should be noted also that there exist several microstructures whose actual fracture strains are far less than the target strains. Figure7c shows the typical microstructures whose fracture strains are smaller than 20% of the target fracture strains. Additionally, the coefficient of determination for the data without data points corresponding to the microstructures shown in Fig.7c was 0.76. All of them are partially incomplete laminated structures. This can be understood as follows. Although laminated structures has a potential to realize higher fracture strains as shown in Fig.4, this is true only when the microstructures are completely laminated. Even when one martensite layer has a tiny hole, the gap between martensite grains becomes the hot spot that induces much earlier rupture. Thus, the box plot shown in Fig.7b is understood to show decreasing values as a result of an attempt to completely laminate the structures to realize the given target fracture strain. This indicates that the framework recognizes the structures shown in Fig.7c to be structurally close to completely laminated structures even though they have far less fracture strains than the completely laminated structures.

As a consequence, these results illustrate that our framework provides a powerful tool for the optimization of material microstructures in terms of target properties, or at least for capturing the trend of microstructures in terms of the change in target property in various cases.

The above results of the generation of material structures corresponding to the target fracture strain indicate that our framework captures the implicit correlation between the material microstructures and the fracture strain. However, generally, it is difficult to interpret implicit knowledge captured by machine learning methods. For that reason, we cannot hastily conclude that machine learning can understand this problem and acquire meaningful knowledge for material design similarly to humans or that it just obtains physically meaningless problem-specific knowledge. Usually, human researchers attain the background physics by noting a part or behavior that will affect a target property during numerous trial-and-error experiments. Generally, this process is time-consuming. Accordingly, approaching implicit knowledge obtained by machine learning methods could be beneficial for achieving a more efficient way to extract general knowledge for material design. Thus, we discuss how to approach the physical background behind the implicit knowledge captured by our framework. In particular, we investigate whether the machine learning framework can identify a part of material microstructures that strongly affects a target property in a similar way human experts can predict on the basis of their experiences.

To identify a critical part of microstructures, we consider calculating a derivative of material microstructures with respect to the fracture strain. This is based on the assumption that human experts unconsciously consider the sensitivity of material microstructures to a slight change in target property. Accordingly, the following variable (Delta) is defined as the derivative:

$$begin{aligned} Delta :=frac{partial D(mathbb {E}_{P(theta |epsilon _f, M_r)}[ theta ])}{partial epsilon _f}, end{aligned}$$

(3)

where (mathbb {E}_{P(theta |epsilon _f, M_r)}[ theta ]) is the expectation of a spatial arrangement of fundamental structures (theta) according to (P(theta |epsilon _f, M_r)), which is the probability distribution captured by PixelCNN. Here, (M_r) and (epsilon _f) are the reference microstructure under consideration and the calculated fracture strain for the microstructure, respectively. In other words, (mathbb {E}_{P(theta |epsilon _f, M_r)}[ theta ]) is the deterministic function of (epsilon _f) and (M_r). In addition, D is the CNN-based deterministic decoder function; hence, (Delta) has the same pixel size of the input microstructure images.

If the machine learning framework correctly captures the physical correlation between the geometry of the material microstructures and the fracture strain, (Delta) is expected to correspond to the areas in (M_r) that highly affects the determination of the fracture strain even without giving the physical mechanism itself. For numerical calculation, (Delta) is approximated as

$$begin{aligned} Delta thickapprox {D(mathbb {E}_{P(theta |epsilon _f+Delta epsilon _f, M_r)}[ theta ])-D(mathbb {E}_{P(theta |epsilon _f, M_r)}[ theta ])}/Delta epsilon _f, end{aligned}$$

(4)

where (Delta epsilon _f) is the gap of the fracture strain, which is set as 0.01 in this paper. Because it is difficult to compare quantitatively the distribution of this variable with the critical microstructure distributions obtained from the physical model, in this paper, we only discuss the location of crucial parts. Thus, the denominator (Delta epsilon _f) is ignored for the calculation of (Delta) in the rest of this paper.

Comparison of derivatives of microstructures with respect to the fracture strain obtained using the machine learning framework with the distributions of void volume fractions calculated on the baisis of physical model. (a)(d) Comparisons for several microstructures. The left, middle, and right column correspond to the reference microstructures, the void distributions obtained using the physical model, and the derivative obtained by the machine learning framework, respectively.

Figure8 shows the comparison of the parts of microstructures critically affecting the determination of the fracture strain obtained by the physical model and our machine learning framework. In the case of the results from machine learning, the absolute values of (Delta) defined in Eq.(3) for each pixel are shown as colormaps. On the other hand, because the fracture behavior is formulated as damage and void-growth processes in the physical model, the void distribution in a critical state directly shows the critical points for the determination of fracture strain. Thus, in the case of the physical model, the calculated void distribution in a critical state is shown in Fig.8. The details of the physical model and the experiment for the determination of some parameters are given in Methodology. For ease of comparison, the ranges of visualized values are changed for each image, while the relative relationships among values of each colormap are kept. Thus, we compare the results qualitatively in terms of the distribution of areas having relatively high values in the next paragraph.

Figure8a,b illustrate the crucial parts of the microstructures composed of relatively long and narrow rectangle-shaped martensite grains. We can see an acceptable agreement between the results of the physical and machine learning methods in terms of the overall distribution of crucial areas which are shown in red in the colormaps of Fig.8. In addition, Fig.8c,d show the parts that critically influence the fracture behavior in the microstructures composed of similarly shaped martensite grains. As an important difference between them, in Fig.8c, the rectangle-shaped martensite grains are irregularly arranged and some martensite grains are close to each other, which might critically affect the fracture behavior, whereas in Fig.8d, circular martensite grains are almost regularly arranged. About Fig.8c, the machine learning framework seems to capture the crucial parts that are predicted by the physical model. As mentioned above, the distributions seem to be dominantly affected by the martensite grains being close to each other. In other words, the short-range interactions among a small number of martensite grains are dominant for the determination of the fracture strain in this case. Also, in Fig.8d, both the physical model and the machine learning framework can predict that the crucial parts are uniformly distributed in square areas.

On the other hand, the physical model also predicts the influence of long-range interactions among martensite grains on fracture behavior, which can be seen in Fig.8c,d as a bandlike distribution. However, the bandlike distribution resulting from the long-range interactions does not seem to be captured by the machine learning framework owing to the characteristic of PixelCNN. Because a global stochastic relationship among the fundamental elements is factorized as a product of stochastic local interactions in PixelCNN as defined in Eq.(1), the extent of interaction exponentially decreases as distance increases. Therefore, the long-range interactions are difficult to be captured by PixelCNN. The discussion of the limitation of PixelCNN in capturing long-range interactions and a remedy for this limitation can be found in28. Figure9 illustrates a sample case showing that the relatively long-range interactions are important for the dertermination of fracture strain. In this case, the determination of the part that critically affects the fracture behavior seems to be difficult using the framework based on PixelCNN.

Sample case showing that a relatively long-range interactions among martensite grains are important for the determination of fracture strain.

For incompletely laminated structures such as that shown in Fig.8a, the martensite layers are expanded to achieve a higher fracture strain even though increasing the martensite volume fraction basically contributes to the decrease in the fracture strain, as shown in Fig.4. Similarly, we can see in Fig.8c that the martensite grains tended to expand to fill the hot spots between them. Additionally, as mentioned above, even though completely laminated structures are structurally similar to incompletely laminated structures, the fracture strains of completely laminated structures are much higher than those of incompletely laminated structures. Thus, eliminating tiny holes that could be causes of hot spots and reaching ({ completely}) laminated structures markedly improve their fracture strains. Altogether, these results imply that the framework recognizes the potential of laminated structures to achieve a higher fracture strain in a similar way that human researchers reach an intuition on completely laminated structures as a result of the consideration of reducing the occurrence of hot spots.

From the above results, we can conclude that our framework can identify the areas that critically affect a target property without human prior knowledge when the local topology of microstructures is dominant for the target property. This implies that machine learning designed consistent with metallurgists process of thinking can approach the background or the meaning of the implicitly extracted knowledge in a similar way that humans acquire an empirical knowledge.

Continue reading here:
Identification of microstructures critically affecting material properties using machine learning framework based on metallurgists' thinking process |...