Page 653«..1020..652653654655..660670..»

The last Machine Learning curriculum you will ever need! – Medium

Your decision to embark on a new AI journey is commendable, and at this stage, I assume youve already solidified your commitment, driven by a compelling reason. The crucial first step is to anchor yourself to this motivation let it serve as the key to unlocking your untapped potential and surmounting any challenges ahead.

Now that youve set your course, you venture online in search of valuable resources. However, what you likely didnt anticipate was navigating through a myriad of courses that often exhibit substantial topic overlap. Its a common challenge faced by many beginners, and truth be told, even seasoned professionals can find themselves entangled in the complexity of choices at times.

Luckily, I am here to help. However, there are a few things that you need to remember.

Embarking on the path of learning is akin to setting out on a journey, not a sprint. Frequently, I receive inquiries such as, How soon until I secure my first job? or What projects should I showcase on my Resume for an internship? I empathize with these queries, having navigated those uncertainties myself. However, to truly comprehend AI and harness its potential for your aspirations, impatience is not your ally. While it might be tempting to expedite the process by combining various tools to achieve quick results, the essence of learning AI lies in embracing the journey. Along this path, challenges will arise that demand a profound understanding of AI principles for long-term success.

Broadening your understanding of machine learning requires a multi-faceted approach. Delve into the intricacies of AI topics by tapping into a variety of resources. Whether its diverse instructors, a range of courses, insightful books, research papers, or thought-provoking blogs immersing yourself in multiple perspectives is the key. This comprehensive exploration not only deepens your insights but also equips you with a well-rounded understanding of artificial intelligence.

Resist the urge to dive into a multitude of courses or topics simultaneously. Opt for a focused approach master one concept thoroughly before venturing into the next. This sequential immersion ensures a solid foundation and a more effective learning experience.

Theory, while essential, is only a fraction of the learning journey. The true grasp of concepts comes from hands-on implementation. Remember, knowledge solidifies when you actively engage with it. Trust the process, invest time in implementing solutions, and solve real-world problems. The insights gained from practical application far exceed those acquired through lengthy lectures, forming the bedrock of profound understanding.

The realization struck me post-industry immersion: thriving in the AI domain demands more than just technical acumen. While AI knowledge is crucial, its only the tip of the iceberg. To navigate the dynamic landscape, supplement your expertise with a toolkit that extends beyond AI algorithms embrace essential software tools like GitHub and Docker. Expand your programming language repertoire, hone your paper-reading skills, delve into cloud computing intricacies, and grasp project management essentials. You will require good writing and documentation skills. The key lies in adaptability; remaining flexible and prepared for the demands of the evolving AI industry.

The rest is here:
The last Machine Learning curriculum you will ever need! - Medium

Read More..

Digital staffing company Aya Healthcare picks up Winnow AI to … – FierceHealthcare

Digital staffing company Aya Healthcare acquired Winnow AI to bolster its physician recruitment capabilities as the industry grapples with a historic provider shortage.

Winnow AI is a data-science-driven recruiting solution that identifies predictive matches and referral connections for each open role at a provider organization. The startup combines artificial intelligence with business intelligence to help organizations tap into a unique source of passive physicians who are likely to relocate to their region.

Winnow AI launched just two years ago to "unlock a better approach to physician recruiting," Ray Guzman, co-founder of Winnow AI and CEO of SwitchPoint Ventures, wrote in a LinkedIn post.

"The speed at which Winnow gained traction demonstrated that it was solving a huge pain point, one that demanded disruption of the status quo. Were delighted that Aya has the same massive vision for Winnow that we do, and we look forward to innovating and growing together," Guzman wrote.

Financial details of the deal were not disclosed.

Aya Healthcare, with more than 7,000 global employees, operates adigital staffing platform that providesevery component of healthcare-focused labor services, including travel nursing and allied health, per diem, permanent staff hiring, interim leadership, locum tenens and nonclinical professionals, according to the company. Ayas software suite includes vendor management, float pool technology, provider services and predictive analytics.

The Winnow AI acquisition marks Aya Healthcare's third M&A deal in five months as the company works to build up its AI capabilities for staffing, hiring and retention.

In July, Aya Healthcare picked upFlexwise Health, a company that offers technology to forecast gaps in patient demand and staffing levels. Its aim is to assist hospitals in optimizing resource allocation and cost.

A few weeks later, the company acquiredPolaris AI, a machine learning platform that predicts future patient volume and staffing levels in clinical settings. Polaris utilizes proprietary machine learning algorithms to intelligently inform staffing needs and provide tools for systems to effectively distribute internal resources and plan alternative schedules.

With its latest deal, Winnow AI will operate within Ayas Provider Solutions division. The company says the startup's capabilities complement its DocCafe brand, aphysician talent acquisition platform with the nations largest pool of active job seekers. Ayas Provider Solutions division will now be able to offer both active and passive job seeker recruitment platforms. The division also enables healthcare organizations to hire locum providers and manage their provider recruitment and engagement through Aya Connect, according to the company.

Were able to help healthcare organizations effectively fill their open provider positions by offering Winnow AI to identify passive job seekers and DocCafe to effectively recruit active physician job seekers, said Alan Braynin, president and CEO of Aya Healthcare, in a statement. This acquisition is an example of our never-ending quest to deliver innovative solutions to our clients that create greater efficiencies, generate cost savings, and improve access to care for the communities they serve.

Winnows AI predicts which physicians are likely to change jobs and where they are most likely to relocate. These insights equip medical leaders and in-house recruiters to drive novel candidate options and referrals and to create perfectly aligned provider teams, leading to faster, more efficient physician recruitment, the company said.

"Winnow AI offers a more targeted approach to building all-star teams by pinpointing candidates who match the profiles of a companys best doctors, Guzman said in a statement. "Ayas ability to scale Winnows innovative solution will help healthcare companies dramatically improve their ability to attract, hire, and retain the best-fit providers for their organizations.

Companies tackling the physician and nursing workforce shortage are attracting big money from investors. There's been a lot of funding activity around startups offering tools and platforms aimed at making it easier to find, fill and upskill for healthcare jobs. According to Pitchbook, asample list of 17 funded startups tied to the space collectively raised more than $700 million from June 2021 to June 2022.

As of June 2022, those companies, including Trusted Health, IntelyCare andNomad Health, raised over $1.15 billion.

Two years ago, Aya Healthcare acquiredVizient's Contract Labor Management business unit and transitioned it to Vaya Workforce Solutions. That business operates as a vendor-neutral workforce solutions provider covering whole-house contract labor needs.

Excerpt from:
Digital staffing company Aya Healthcare picks up Winnow AI to ... - FierceHealthcare

Read More..

Ethnic disparity in diagnosing asymptomatic bacterial vaginosis … – Nature.com

Dataset

The dataset was originally reported by Ravel et al.16. The study was registered at clinicaltrials.gov under ID NCT00576797. The protocol was approved by the institutional review boards at Emory University School of Medicine, Grady Memorial Hospital, and the University of Maryland School of Medicine. Written informed consent was obtained by the authors of the original study.

Samples were taken from 394 asymptomatic women. 97 of these patients were categorized as positive for BV, based on Nugent score. In the preprocessing of the data, information about community group, ethnicity, and Nugent score was removed from the training and testing datasets. Ethnicity information was stored to be referenced later during the ethnicity-specific testing. 16S rRNA values were listed as a percentage of the total 16S rRNA sample, so those values were normalized by dividing by 100. pH values ranged on a scale from 1 to 14 and were normalized by dividing by 14.

Each experiment was run 10 times, with a different random seed defining the shuffle state, to gauge variance of performance.

Four supervised machine learning models were evaluated. Logistic regression (LR), support vector machine (SVM), random forest (RF), and Multi-layer Perceptron (MLP) models were implemented with the scikit-learn python library. LR fits a boundary curve to separate the data into two classes. SVM finds a hyperplane that maximizes the margin between two classes. These methods were implemented to test whether boundary-based models can perform fairly among different ethnicities. RF is a model that creates an ensemble of decision trees and was implemented to test how a decision-based model would classify each patient. MLP passes information along nodes and adjusts weights and biases for each node to optimize its classification. MLP was implemented to test how a neural network-based approach would perform fairly on the data.

Five-fold stratified cross validation was used to prevent overfitting and to ensure that each ethnicity has at least two positive cases in the test folds. Data were stratified by a combination of ethnicity and diagnosis to ensure that each fold has every representation from each group with comparable distributions.

For each supervised machine learning model, hyper parameter tuning was performed by employing a grid search methodology from the scikit-learn python library. Nested cross validation with 4 folds and 2 repeats was used as the training subset of the cross validation scheme.

For Logistic Regression, the following hyper-parameters were tested: solver (newton-cg, lbfgs, liblinear) and the inverse of regularization strength C (100, 10, 1.0, 0.1, 0.01).

For SVM, the following hyper-parameters were tested: kernel (polynomial, radial basis function, sigmoid) and the inverse regularization parameter C (10, 1.0, 0.1, 0.01).

For Random Forest, the following hyper-parameters were tested: number of estimators (10, 100, 1000) and maximum features (square root and logarithm to base 2 of the number of features).

For Multi-layer perceptron, the following hyper-parameters were tested: hidden layer size (3 hidden layers of 10,30, and 10 neurons and 1 hidden layer of 20 neurons), solver (stochastic gradient descent and Adam optimizer), regularization parameter alpha (0.0001, or .05), and learning rate (constant and adaptive).

The models were evaluated using the following metrics: balanced accuracy, average precision, false positive rate (FPR), and false negative rate (FNR). Balanced accuracy was chosen to better capture the practical performance of the models while using an unbalanced dataset. Average precision is an estimate of the area under the precision recall curve, similar to AUC which is the area under the ROC curve. The precision-recall curve is used instead of a receiver operator curve to better capture the performance of the models on an unbalanced dataset39. Previous studies with this dataset reveal particularly good AUC scores and accuracy, which is to be expected with a highly unbalanced dataset.

The precision-recall curve was generated using the true labels and predicted probabilities from every fold of every run to summarize the overall precision-recall performance for each model. Balanced accuracy and average precision were computed using the corresponding functions found in the sklearn.metrics package. FPR and FNR were calculated computed and coded using Equations below39.

Below are the equations for the metrics used to test the Supervised Machine Learning models:

$${Precision}=frac{{TP}}{{TP}+{FP}}$$

(1)

$${Recall}=frac{{TP}}{{TP}+{FN}}$$

(2)

$${Balanced},{Accuracy}=frac{1}{2}left(frac{{TP}}{{TP}+{FN}}+frac{{TN}}{{TN}+{FP}}right)$$

(3)

$${FPR}=frac{{FP}}{{FP}+{TN}}$$

(4)

$${FNR}=frac{{FN}}{{FN}+{TP}}$$

(5)

where TP is the number of true positives, TN is the number of true negatives, FP is the number of false positives, and FN is the number of false negatives.

$${Average},{Precison}=sum _{n}left({R}_{n}-{R}_{n-1}right){P}_{n}$$

(6)

where R denotes recall, and P denotes precision.

The performance of the models were tested against each other as previously stated. Once the model made a prediction, the stored ethnicity information was used to reference which ethnicity each predicted label and actual label belonged to. These subsets were then used as inputs for the metrics functions.

To see how training on data containing one ethnicity affects the performance and fairness of the model, an SVM model was trained on subsets that each contained only one ethnicity. Information on which ethnicity each datapoint belonged to was not given to the models.

To increase the performance and accuracy of the model, several feature selection methods were used to reduce the 251 features used to train the machine learning models. These sets of features were then used to achieve similar or higher accuracy with the machine learning models used. The feature selection methods used included the ANOVA F-test, two-sided T-Test, Point Biserial correlation, and the Gini impurity. The libraries used for these feature selection tests were the statistics and scikit learn packages in Python. Each feature test was performed with all ethnicities, then only the white subset, only Black, only Asian, and only Hispanic.

The ANOVA F-Test was used to select 50 features with the highest F-value. The function used calculates the ANOVA F-value between the feature and target variable using variance between groups and within the groups. The formula used to calculate this is defined as:

$$F=frac{{SSB}/(k-1)}{{SSW}/(n-k)}$$

(7)

Where k is the number of groups, n is the total sample size, SSB is the variance between groups, and SSW is the sum of variance within each group. The two-tailed T-Test was used to compare the BV negative versus BV positive groups rRNA data against each other. The two-tailed T-Test is used to compare the means of two independent groups against each other. The null hypothesis in a two-tailed T-Test is defined as the means of the two groups being equal while the alternative hypothesis is that they are not equal. The dataset was split up into samples that were BV negative and BV positive which then compared the mean of each feature against each other to find significant differences. A p-value <0.05 allows us to reject the null hypothesis that the mean between the two groups is the same, indicating there is a significant difference between the positive and negative groups for each feature. Thus, we use a p-value of less than 0.05 to select important features. The number of features selected were between 40 and 75 depending on the ethnicity group used. The formula for finding the t-value is defined as:

$$t=frac{left({bar{x}}_{1}-{bar{x}}_{2}right)}{sqrt{frac{({{s}_{1}})^{2}}{{n}_{1}}+frac{({{s}_{2}})^{2}}{{n}_{2}}}}$$

(8)

({bar{{rm{x}}}}_{1,2}) being the mean of the two groups. ({{rm{s}}}_{1,2}) as the standard deviation of the two groups. ({{rm{n}}}_{1,2}) being the number of samples in the two groups. The p-value is then found through the t-value by calculating the cumulative distribution function. This defines probability distribution of the t-distribution by the area under the curve. The degrees of freedom are also needed to calculate the p-value. They are the number of variables used to find the p-value with a higher number being more precise. The formulas are defined as:

$${rm{df}}={n}_{1}+{n}_{2}{{{-}}}2$$

(9)

$${p}=2* left(1-{rm{CDF}}left(left|tright|,{rm{df}}right)right)$$

(10)

where ({df}) denotes the degrees of freedom and ({{rm{n}}}_{1,2}) being the number of samples in the group. The Point Biserial correlation test is used to compare categorical against continuous data. For our dataset was used to compare the categorical BV negative or positive classification against the continuous rRNA bacterial data. Each feature has a p-value and correlation value associated with it which was then restricted by an alpha of 0.2 and further restricted by only correlation values >0.5 showing a strong correlation. The purpose of the alpha value is to indicate the level of confidence of a p-value being significant. An alpha of 0.2 was chosen because the Point Biserial test tends to return higher p-values. This formula is defined as:

$${{r}}_{{pb}}=frac{left({M}_{1}-{M}_{0}right)}{{rm{s}}},sqrt{{pq}}$$

(11)

where M1 is the mean of the continuous variable for the categorical variable with a value of 1; M0 is the mean of the continuous variable for the categorical variable with a value of 0; s denotes the standard deviation of the continuous variable; p is the proportion of samples with a value of 1 to the sample set; and q is the proportion of samples with a value of 0 to the sample set.

Two feature sets were made from the Point Biserial test. One feature set included only the features that were statistically significant using a p-value of <0.2 which returned 60100 significant features depending on the ethnicity set used. The second feature set included features that were restricted by a p-value<0.2 and greater than a correlation value of 0.5. This second feature set contained 815 features depending on the ethnicity set used.

Features were also selected using Gini impurity. Gini impurity defines the impurity of the nodes which will return a binary split at a node. It will calculate the probability of misclassifying a randomly chosen data point. The Gini impurity model fitted a Random Forest model with the dataset and took the Gini scores for each feature based on the largest reduction of Gini impurity when splitting nodes. The higher the reduction of Gini value, the impurity after the split, the more important the feature is used in predicting the target variable. The Gini impurity value varies between 0 and 1. Using Gini, the total number of features were reduced to 310 features when using the ethnicity-specific sets and 20 features when using all ethnicities. The formula is defined as:

$${Gini}=1-sum {{p}_{i}}^{2}$$

(12)

where ({{rm{p}}}_{{rm{i}}}) is the proportion of each class in the node. The five sets of selected features from each of the five ethnicities were used to train a model using four supervised machine learning algorithms (LR, MLP, RF, SVM) with the full dataset using our nested cross-validation schemed as previously described. All features were selected using the training sets only, and they were applied to the test sets after being selected for testing. Five-fold stratified cross validation was used for each model to gather including means and confidence intervals.

Originally posted here:
Ethnic disparity in diagnosing asymptomatic bacterial vaginosis ... - Nature.com

Read More..

The Role of Machine Learning in Precision Synthesis – The Medicine Maker

With the aim to overcome key barriers to applying machine learning (ML) to real experiments and processes for example, the fact that ML typically struggles with sparse data (data with gaps) our latest project, in partnership with the Centre for Process Innovation (CPI) and with funding from Innovate UK, focuses on the potential for ML to act as a catalyst for manufacturing oligonucleotide therapeutics. We are improving predictive modelling tools, experimental program design, optimal process parameter discovery, and target output identification.

Oligonucleotides are difficult to manufacture particularly at scale.They are large, complex molecules that require a multi-stage synthetic process, interleaved with significant purification and analysis stages. The presence of impurities or small variations in reaction conditions and process steps can make significant differences to the structure, yield, and quality of the end product. Synthesis is expensive, meaning that experimental data is often sparse, and research teams would prefer to extract as much value as they can from the data that exists. Alongside these common industry problems, oligonucleotide manufacturing also has significant sustainability challenges; namely, large amounts of waste produced, poor atom economy, and low use of renewable feedstocks.

For these reasons, oligonucleotide manufacturing is an ideal target for ML, which can help detect subtle, non-linear relationships in multi-parameter data that might otherwise be missed. I expect ML to help research teams better understand the key factors driving oligonucleotide manufacturing processes, leading to improved design and control of these processes.

The importance of oligonucleotide therapies cannot be understated. Despite significant advances in medicine, there is still a large gap between the number of diseases and disorders that are druggable with approved therapies. Oligonucleotide therapies represent a relatively new and innovative approachwith the potential to treat a wide range of diseases, including rare genetic disorders, certain types of cancer, and neurodegenerative conditions.The high specificity of oligonucleotides and their ability to target gene mutations or protein expression means that they are a form of personalized medicine, with fewer off-target effects and, potentially, fewer side effects than small molecules. Given the promise, many companies within the pharma industry are either investing heavily in platform R&D to progress oligonucleotide pipelines or forming partnerships and collaborationsto advance these to commercialization.

Going back to ML adoption in this space, oligonucleotides will likely suffer the same challenges seen in other modalities and sectors. Traditional ML methods and algorithms require large, high-quality datasets for training. And as noted, In oligonucleotide manufacturing it is challenging to obtaining sufficient data especially for highly complex and nonlinear proprietary processes. Over-simplified models may not provide meaningful insights.Building models that can generalize across different data formats and processes for different pharma companies will also be challenging. As will the integration of ML solutions into existing manufacturing systems, where it is important to work seamlessly with automation and control systems. The final barriers to adoption are simply inertia or a lack of knowledge and understanding of ML technologies.

Certainly, a small number of specialist companies have made progress in addressing the manufacturing challenges of oligonucleotides, but their insights and models are often proprietary (and pharma is an industry where knowledge is not widely shared). As with many challenges, collaboration is likely key; pre-competitive projects could combine expertise, with ML models acting as a vehicle for capturing and sharing knowledge among the collaborating organizations. This way, what is learnt can be shared to accelerate progress and drive innovation.

And in my view, its absolutely worth it! A somewhat consistent rule of thumb for ML technology when applied to the DoE is a reduction of around 5080 percent in the number of experiments required to achieve a given objective.Furthermore, it could generate new insights and guide informed decision making. Yes, its speculative but the effective use of ML could drive two- to five-fold reductions in the problematic process development phase of bringing new oligonucleotides to market.

CEO and co-founder of Intellegens, a machine learning (ML) software company, focused on research-intensive industries such as chemicals, life sciences, and materials. Originally a spinout from the University of Cambridge, Intellegens worked alongside automotive giant Rolls Royce in a research project to extract more value from rare experimental and process data to optimize the design of a new superalloy.

Continue reading here:
The Role of Machine Learning in Precision Synthesis - The Medicine Maker

Read More..

Looking beyond the AI hype: Delivering real value for financial … – Fintech Nexus News

If a financial institution looks beyond the hype of AI and tempers its expectations, it can use AI to deliver measurable business results. Thats been the experience of Amounts director of decision science Garrett Laird.

Given the interest in Chat GPT and related tools, the recent buzz around AI is understandable. Like many in fintech, Laird reminds the excited that AI has been around in such forms as machine learning for years. Avant has used machine learning in credit underwriting for at least a decade.

Its not a silver bullet, Laird said. It does some things really, really well. But it wont solve all your problems, especially in our space.

Financial products are highly regulated, right? These new LLMs (large language models) are entirely unexplainable; theyre pretty much true black-box models, so they limit the applications and use cases.

Laird sees clear use cases in outlier detection and unsupervised learning. He credits the current AI fervor with igniting interest in LLMs. As businesses look for ways to deploy LLMs, they are also looking at other AI types.

Regulations prevent AI from being used everywhere in financial services. Laird cited the many protected classifications that dictate how and where advertisements and solicitations can be sent. If your AI model cannot explain why one customer got an offer while another did not, youre asking for trouble.

Machine learning can be used to become more compliant because you can empirically describe why youre making the decisions youre making, Laird said. When there are humans making decisions everyone has their implicit biases, and those are hard to measure or even know what they are.

With algorithms and machine learning, you can empirically understand if a model is biased and in what ways and then you can control for that. While there are many restrictions on one side, I think many things were doing with machine learning and AI benefit consumers from a discrimination and compliance perspective.

Laird said the training models depend on what their systems are used for. Fraud models must be updated quickly and often with third-party sources, historical information and consumer data.

This is one area where machine learning helps. Machine learning operations can ensure proper validations are completed. They prevent it from picking up discriminatory data or information from protected classes.

Laird said an industry cliche is that 90% of machine learning work is data preparation. That has two parts: having relevant data and ensuring it is accessible in real time so it can make valuable business decisions.

While credit provision might not bring the same urgency as fraud, Laird also advises considering how it can benefit from AI. Credit models must have strong governance and risk management processes in place. They need good data sets. Lenders require a thorough understanding of their customers, which, in the case of mortgages, can take years.

Getting access to the right data is a huge challenge, and then making sure its the right population, Laird said. Thats a trend the industry is moving in: product-specific but also customer-base-specific modelling.

The direction were headed is like the democratization of machine learning for credit underwriting where you have models that are very catered to your very unique situation. That challenges many banks because it takes a lot of human capital. Having it takes a lot of data, and its not something you have overnight.

Also read:

AI lowers the entry barrier for fraudsters by providing sophisticated tools and allowing them to communicate in better-quality English. Combatting them also involves AI as one of many layers.

However, AI is used differently with different fraud types. First-party fraudsters can evade identity checks, which introduce friction for legitimate customers.

Third-party fraud brings challenges to supervised models. Those models are based on learnings from previous cases of such fraud. Their characteristics are identified, and models are developed. AI can help to identify those patterns quickly.

However, the process is never-ending because systems must quickly adjust as fraudsters determine how to beat mitigation challenges. Laird said he focuses on that by deploying velocity checks.

We put a lot of mental effort into identifying ways to pick up on these clusters of bad actors, Laird said. And there are many ways you can do that. A couple of the interesting ones that we employ are velocity checks. A lot of times, a fraud ring will exhibit similar behaviors. They might be applying from a certain geography, have the same bank theyre applying from, or have similar device data. They might use VOIP, any number of like attributes.

Laird said some institutions also use unsupervised learning. They might not have specific targets, but they can detect patterns using clustering algorithms. If a population starts defaulting or claiming fraud, the algorithms can identify similar behaviors that need further scrutiny.

Recent financial sector turbulence lends itself to rising deposit-related fraud. If a banks defences are sub-par, they could find themselves vulnerable to fraud that is already happening.

That is probably a problem thats already starting to rear its head and will only get worse, Laird suggested. I think with all of the movement in deposits that happened this past spring, with SVB and all the other events, there was a mad rush of deposit opening.

And with that, two things always happen. Theres an influx of volume. It makes it easier for fraudsters to slip through the cracks. Also, many banks saw that as an opportunity and probably either rushed solutions out or reduced some of their defences. We think theres probably a lot of dormant, recently opened deposit accounts that are probably in the near future going to be utilized as vehicles for bust-out fraud.

Laird returned to case-specific modelling as a significant emerging trend. FICO and Vantage are good models many use, but theyre generic for everything from mortgages to credit cards and personal loans. Casting a wide net limits accuracy, and given increased competition, more bespoke models are a must.

I can go on Credit Karma and get 20 offers with two clicks of a button, or I can go to 100 different websites and get an offer without impacting my credit, Laird observed. If youre trying to compete with that, if your pricing is just based on a FICO score or Vantage score, youre going to get that 700 FICO customer thats trending towards 650, whereas someone with a more advanced credit model is going to get that 700 thats trending towards 750.

Laird is eagerly watching developments following the Consumer Financial Protection Bureaus recent announcement on open banking. Financial institutions must make their banking data available.

Thats a modelling goldmine, Laird said. Financial institutions had an advantage in lending to their customer bases because only they can access that information. Now that its publicly available, that data can be used by all financial institutions to make underwriting decisions. Laird said its mission-critical for financial institutions to have good solutions.

Financial institutions generally take conservative approaches to AI. Most have used Generative AI for internal efficiencies, not direct customer interactions. That time will come but in limited capacities.

Laird reiterated his excitement about the renewed interest in machine learning. He believes they are well-suited to address the problems.

Im excited that theres that renewed interest in investment and an appetite for starting to leverage AI for fraud, Laird said. Its been there for a while.

I think the increased focus on credit underwriting is another one that I get really excited about because with the new open banking regulations coming out, I think financial institutions that dont embrace it are going to get left behind. Theyre going to be adversely selected; theyre not going to be able to remain competitive. It behooves everyone to start thinking about it and understanding ways to leverage that from not just the traditional fraud focuses but increasingly on the credit side.

Tony is a long-time contributor in the fintech and alt-fi spaces.A two-time LendIt Journalist of the Year nominee and winner in 2018, Tony has written more than 2,000 original articles on the blockchain, peer-to-peer lending, crowdfunding, and emerging technologies over the past seven years.He has hosted panels at LendIt, the CfPA Summit, and DECENT's Unchained, a blockchain exposition in Hong Kong. Email Tony here.

Read this article:
Looking beyond the AI hype: Delivering real value for financial ... - Fintech Nexus News

Read More..

Applications of Semi-supervised Learning part3(Machine Learning … – Medium

Author : Tao Wang, Yuanbin Chen, Xinlin Zhang, Yuanbo Zhou, Junlin Lan, Bizhe Bai, Tao Tan, Min Du, Qinquan Gao, Tong Tong

Abstract : Supervised learning algorithms based on Convolutional Neural Networks have become the benchmark for medical image segmentation tasks, but their effectiveness heavily relies on a large amount of labeled data. However, annotating medical image datasets is a laborious and time-consuming process. Inspired by semi-supervised algorithms that use both labeled and unlabeled data for training, we propose the PLGDF framework, which builds upon the mean teacher network for segmenting medical images with less annotation. We propose a novel pseudo-label utilization scheme, which combines labeled and unlabeled data to augment the dataset effectively. Additionally, we enforce the consistency between different scales in the decoder module of the segmentation network and propose a loss function suitable for evaluating the consistency. Moreover, we incorporate a sharpening operation on the predicted results, further enhancing the accuracy of the segmentation. Extensive experiments on three publicly available datasets demonstrate that the PLGDF framework can largely improve performance by incorporating the unlabeled data. Meanwhile, our framework yields superior performance compared to six state-of-the-art semi-supervised learning methods. The codes of this study are available at https://github.com/ortonwang/PLGDF.

2.SSASS: Semi-Supervised Approach for Stenosis Segmentation (arXiv)

Author : In Kyu Lee, Junsup Shin, Yong-Hee Lee, Jonghoe Ku, Hyun-Woo Kim

Abstract : Coronary artery stenosis is a critical health risk, and its precise identification in Coronary Angiography (CAG) can significantly aid medical practitioners in accurately evaluating the severity of a patients condition. The complexity of coronary artery structures combined with the inherent noise in X-ray images poses a considerable challenge to this task. To tackle these obstacles, we introduce a semi-supervised approach for cardiovascular stenosis segmentation. Our strategy begins with data augmentation, specifically tailored to replicate the structural characteristics of coronary arteries. We then apply a pseudo-label-based semi-supervised learning technique that leverages the data generated through our augmentation process. Impressively, our approach demonstrated an exceptional performance in the Automatic Region-based Coronary Artery Disease diagnostics using x-ray angiography imagEs (ARCADE) Stenosis Detection Algorithm challenge by utilizing a single model instead of relying on an ensemble of multiple models. This success emphasizes our methods capability and efficiency in providing an automated solution for accurately assessing stenosis severity from medical imaging dat

See the rest here:
Applications of Semi-supervised Learning part3(Machine Learning ... - Medium

Read More..

Applications of Semi-supervised Learning part4(Machine Learning … – Medium

Author : Gaurav Sahu, Olga Vechtomova, Issam H. Laradji

Abstract : This work tackles the task of extractive text summarization in a limited labeled data scenario using a semi-supervised approach. Specifically, we propose a prompt-based pseudolabel selection strategy using GPT-4. We evaluate our method on three text summarization datasets: TweetSumm, WikiHow, and ArXiv/PubMed. Our experiments show that by using an LLM to evaluate and generate pseudolabels, we can improve the ROUGE-1 by 1020% on the different datasets, which is akin to enhancing pretrained models. We also show that such a method needs a smaller pool of unlabeled examples to perform better

2.Semi-supervised machine learning model for Lagrangian flow state estimation (arXiv)

Author : Reno Miura, Koji Fukagata

Abstract : In recent years, many researchers have demonstrated the strength of supervised machine learning for flow state estimation. Most of the studies assume that the sensors are fixed and the high-resolution ground truth can be prepared. However, the sensors are not always fixed and may be floating in practical situations for example, in oceanography and river hydraulics, sensors are generally floating. In addition, floating sensors make it more difficult to collect the high-resolution ground truth. We here propose a machine learning model for state estimation from such floating sensors without requiring high-resolution ground-truth data for training. This model estimates velocity fields only from floating sensor measurements and is trained with a loss function using only sensor locations. We call this loss function as a semi-supervised loss function, since the sensor measurements are utilized as the ground truth but high-resolution data of the entire velocity fields are not required. To demonstrate the performance of the proposed model, we consider Stokes second problem and two-dimensional decaying homogeneous isotropic turbulence. Our results reveal that the proposed semi-supervised model can estimate velocity fields with reasonable accuracy when the appropriate number of sensors are spatially distributed to some extent in the domain. We also discuss the dependence of the estimation accuracy on the number and distribution of sensors.

Read the rest here:
Applications of Semi-supervised Learning part4(Machine Learning ... - Medium

Read More..

Applications of Semi-supervised Learning part1(Machine Learning … – Medium

Author : Hao Dong, Gatan Frusque, Yue Zhao, Eleni Chatzi, Olga Fink

Abstract : Anomaly detection (AD) is essential in identifying rare and often critical events in complex systems, finding applications in fields such as network intrusion detection, financial fraud detection, and fault detection in infrastructure and industrial systems. While AD is typically treated as an unsupervised learning task due to the high cost of label annotation, it is more practical to assume access to a small set of labeled anomaly samples from domain experts, as is the case for semi-supervised anomaly detection. Semi-supervised and supervised approaches can leverage such labeled data, resulting in improved performance. In this paper, rather than proposing a new semi-supervised or supervised approach for AD, we introduce a novel algorithm for generating additional pseudo-anomalies on the basis of the limited labeled anomalies and a large volume of unlabeled data. This serves as an augmentation to facilitate the detection of new anomalies. Our proposed algorithm, named Nearest Neighbor Gaussian Mixup (NNG-Mix), efficiently integrates information from both labeled and unlabeled data to generate pseudo-anomalies. We compare the performance of this novel algorithm with commonly applied augmentation techniques, such as Mixup and Cutout. We evaluate NNG-Mix by training various existing semi-supervised and supervised anomaly detection algorithms on the original training data along with the generated pseudo-anomalies. Through extensive experiments on 57 benchmark datasets in ADBench, reflecting different data types, we demonstrate that NNG-Mix outperforms other data augmentation methods. It yields significant performance improvements compared to the baselines trained exclusively on the original training data. Notably, NNG-Mix yields up to 16.4%, 8.8%, and 8.0% improvements on Classical, CV, and NLP datasets in ADBench. Our source code will be available at https://github.com/donghao51/NNG-Mix

2.Segment Together: A Versatile Paradigm for Semi-Supervised Medical Image Segmentation (arXiv)

Author : Qingjie Zeng, Yutong Xie, Zilin Lu, Mengkang Lu, Yicheng Wu, Yong Xia

Abstract : Annotation scarcity has become a major obstacle for training powerful deep-learning models for medical image segmentation, restricting their deployment in clinical scenarios. To address it, semi-supervised learning by exploiting abundant unlabeled data is highly desirable to boost the model training. However, most existing works still focus on limited medical tasks and underestimate the potential of learning across diverse tasks and multiple datasets. Therefore, in this paper, we introduce a textbf{Ver}satile textbf{Semi}-supervised framework (VerSemi) to point out a new perspective that integrates various tasks into a unified model with a broad label space, to exploit more unlabeled data for semi-supervised medical image segmentation. Specifically, we introduce a dynamic task-prompted design to segment various targets from different datasets. Next, this unified model is used to identify the foreground regions from all labeled data, to capture cross-dataset semantics. Particularly, we create a synthetic task with a cutmix strategy to augment foreground targets within the expanded label space. To effectively utilize unlabeled data, we introduce a consistency constraint. This involves aligning aggregated predictions from various tasks with those from the synthetic task, further guiding the model in accurately segmenting foreground regions during training. We evaluated our VerSemi model on four public benchmarking datasets. Extensive experiments demonstrated that VerSemi can consistently outperform the second-best method by a large margin (e.g., an average 2.69% Dice gain on four datasets), setting new SOTA performance for semi-supervised medical image segmentation. The code will be released.

View original post here:
Applications of Semi-supervised Learning part1(Machine Learning ... - Medium

Read More..

Applications of Semi-supervised Learning part2(Machine Learning … – Medium

Author : Yue Fan, Anna Kukleva, Dengxin Dai, Bernt Schiele

Abstract : Semi-supervised learning (SSL) methods effectively leverage unlabeled data to improve model generalization. However, SSL models often underperform in open-set scenarios, where unlabeled data contain outliers from novel categories that do not appear in the labeled set. In this paper, we study the challenging and realistic open-set SSL setting, where the goal is to both correctly classify inliers and to detect outliers. Intuitively, the inlier classifier should be trained on inlier data only. However, we find that inlier classification performance can be largely improved by incorporating high-confidence pseudo-labeled data, regardless of whether they are inliers or outliers. Also, we propose to utilize non-linear transformations to separate the features used for inlier classification and outlier detection in the multi-task learning framework, preventing adverse effects between them. Additionally, we introduce pseudo-negative mining, which further boosts outlier detection performance. The three ingredients lead to what we call Simple but Strong Baseline (SSB) for open-set SSL. In experiments, SSB greatly improves both inlier classification and outlier detection performance, outperforming existing methods by a large margin. Our code will be released at https://github.com/YUE-FAN/SSB.

2.MSE-Nets: Multi-annotated Semi-supervised Ensemble Networks for Improving Segmentation of Medical Image with Ambiguous Boundaries (arXiv)

Author : Shuai Wang, Tengjin Weng, Jingyi Wang, Yang Shen, Zhidong Zhao, Yixiu Liu, Pengfei Jiao, Zhiming Cheng, Yaqi Wang

Abstract : Medical image segmentation annotations exhibit variations among experts due to the ambiguous boundaries of segmented objects and backgrounds in medical images. Although using multiple annotations for each image in the fully-supervised has been extensively studied for training deep models, obtaining a large amount of multi-annotated data is challenging due to the substantial time and manpower costs required for segmentation annotations, resulting in most images lacking any annotations. To address this, we propose Multi-annotated Semi-supervised Ensemble Networks (MSE-Nets) for learning segmentation from limited multi-annotated and abundant unannotated data. Specifically, we introduce the Network Pairwise Consistency Enhancement (NPCE) module and Multi-Network Pseudo Supervised (MNPS) module to enhance MSE-Nets for the segmentation task by considering two major factors: (1) to optimize the utilization of all accessible multi-annotated data, the NPCE separates (dis)agreement annotations of multi-annotated data at the pixel level and handles agreement and disagreement annotations in different ways, (2) to mitigate the introduction of imprecise pseudo-labels, the MNPS extends the training data by leveraging consistent pseudo-labels from unannotated data. Finally, we improve confidence calibration by averaging the predictions of base networks. Experiments on the ISIC dataset show that we reduced the demand for multi-annotated data by 97.75% and narrowed the gap with the best fully-supervised baseline to just a Jaccard index of 4%. Furthermore, compared to other semi-supervised methods that rely only on a single annotation or a combined fusion approach, the comprehensive experimental results on ISIC and RIGA datasets demonstrate the superior performance of our proposed method in medical image segmentation with ambiguous boundaries

Originally posted here:
Applications of Semi-supervised Learning part2(Machine Learning ... - Medium

Read More..

‘Your United States was normal’: has translation tech really made … – The Conversation

Every day, millions of people start the day by posting a greeting on social media. None of them expect to be arrested for their friendly morning ritual.

But thats exactly what happened to a Palestinian construction worker in 2017, when the caption (good morning) on his Facebook selfie was auto-translated as attack them.

A human Arabic speaker would have immediately recognized as an informal way to say good morning. Not so AI. Machines are notoriously bad at dealing with variation, a key characteristic of all human languages.

With recent advances in automated translation, the belief is taking hold that humans, particularly English speakers, no longer need to learn other languages. Why bother with the effort when Google Translate and a host of other apps can do it for us?

In fact, some Anglophone universities are making precisely this argument to dismantle their language programs.

Unfortunately, language technologies are nowhere near being able to replace human language skills and will not be able to do so in the foreseeable future because machine language learning and human language learning differ in fundamental ways.

For machine translation, algorithms are trained on large amounts of texts to find the probabilities of different patterns of words. These texts can be both monolingual and bilingual.

Bilingual training data comes in the form of human-translated parallel texts. These are almost always based on the standard version of the training language, excluding dialects and slang phrases, as in the example above.

Diversity is a characteristic of all human languages, but diversity is a problem for machines. For instance, deadly means causing death in most varieties of English, and that is what appears in the training data.

The Australian meaning of excellent (from Aboriginal English) puts a spanner in the works. If you input Deadly Awards into any translation app, what youll get in your target language is the equivalent of death-causing awards.

The internal linguistic diversity of English, as of any other language, is accompanied by great diversity across languages. Each language does things differently.

Tense, number or gender, for example, need to be grammatically encoded in some languages but not in others. Translating the simple English statement I am a student into German requires the inclusion of a grammatical gender marking and so will either end up as I am a male student or I am a female student.

Read more: Friday essay: is this the end of translation?

Furthermore, some languages are spoken by many people, have powerful nation states behind them, and are well resourced. Others are not.

Well resourced in the context of machine learning means that large digital corpora of training data are available.

The lists of language options offered by automated translation tools like the list of 133 languages in which Google Translate is currently available erase all these differences and suggest that each option is the same.

Nothing could be further from the truth. English is in a class of its own, with over 90% of the training data behind large language models being in English.

The remainder comes from a few dozen languages, in which data of varying sizes are available. The majority of the worlds 6,000+ languages are simply missing in action. Apps for some of these are now being created from models pre-trained on English, which further serves to cement the dominance of English.

One consequence of inequalities in the training data is that translations into English usually sound quite good because the app can draw both on bilingual and monolingual training data. This doesnt mean they are accurate: one recent study found about half of all questions in Vietnamese were incorrectly auto-translated as statements.

Machine-translated text into languages other than English is even more problematic and routinely riddled with mistakes. For instance, COVID-19 testing information auto-translated into German included invented words, grammatical errors, and inconsistencies.

Machine translation is not as good as most people think, but it is useful to get the gist of web sites or be able to ask for directions in a tourist destination with the help of an app.

However, that is not where it ends. Translation apps are increasingly used in high-stakes contexts, such as hospitals, where staff may attempt to bypass human interpreters for quick communication with patients who have limited proficiency in English.

Read more: The problem with machine translation: beware the wisdom of the crowd

This causes big problems when, for instance, a patients discharge instructions state the equivalent of Your United States was normal an error resulting from the abbreviation US being used for ultrasound in medical contexts.

Therefore, there is consensus that translation apps are suitable only in risk-free or low-risk situations. Unfortunately, sometimes even a caption on a selfie can turn into a high-risk situation.

Only humans can identify what constitutes a low- or high-risk situation and whether the use of machine translation may be appropriate. To make informed decisions, humans need to understand both how languages work and how machine learning works.

It could be argued that all the errors described here can be ironed out with more training data. There are two problems with this line of reasoning. First, AI already has more training data than any human will ever be able to ingest, yet makes mistakes no human with much lower levels of investment in their language learning would make.

Second, and more perniciously, training machines to do our language learning for us is incredibly costly. There are the well-known environmental costs of AI, of course. But there is also the cost of dismantling language teaching programs.

If we let go of language programs because we can outsource simple multilingual tasks to machines, we will never train humans to achieve advanced language proficiency. Even from the perspective of pure strategic national interest, the skills to communicate across language barriers in more risky contexts of economics, diplomacy or healthcare are essential.

Languages are diverse, fuzzy, variable, relational and deeply social. Algorithms are the opposite. By buying into the hype that machines can do our language work for us we dehumanise what it means to use languages to communicate, to make meaning, to create relationships and to build communities.

The author would like to thank Ava Vahedi, a Master of mathematics student at UNSW, for her help in writing this article.

Here is the original post:
'Your United States was normal': has translation tech really made ... - The Conversation

Read More..