In part 2 of our machine learning tutorial, learn how to use the supervised learning approach to machine learning to produce the best predictions.

The following article is comprised of excerpts from the course “Fundamental Machine Learning” that is part of the Machine Learning Specialist certification program from Arcitura Education. It is the second part of the 13-part series, “Using machine learning algorithms, practices and patterns.”

Supervised learning is one of the most widely used machine learning approaches. It can be useful for predicting financial results, detecting fraud, recognizing objects in images and evaluating or assessing risk. The aim of supervised learning is to allow machine learning functions to work in such a way that enables the input data to be used to predict the output class for each new data instance for which the classification is not already known.

With supervised learning, the input data and output data (also called the class) are known in advance. This allows the model to be trained so that it produces the best predictions of classes for the training data by knowing when the model did and did not make a classification error (Figure 1). Subsequent to training the model with the labeled data set, the trained model can then be used to classify future input data with unknown classification.

As part of a supervised learning process, the machine learning system is required to identify circles in images from the data it receives. The result is a form of predictive modeling, whereby the machine recognizes the circles from the other shapes in images (Figure 1).

As the machine learning system continues to make decisions based on the data presented to it, the results of its decisions are reviewed (supervised) by the algorithm. When incorrect decisions are made during training with the labeled data, the algorithm has the opportunity to make adjustments as part of the training process .

Once the model is trained and deployed, the machine learning system can make decisions based on new data it processes.

The following are common supervised learning algorithms, algorithm types and practices:

• Classification
• Decision tree
• Regression
• Predictive modeling
• Ensemble models and methods

Read on to learn more about these five algorithms.

Classification

A problem is referred to as a classification when the output is expected to be a category — such as a circle, a color, a type of car or an outcome. Take, for example, a problem that asks Based on historical weather data, will it rain on February 12? If the model predicts two class values, then the output could be rain and no rain. A classifier would not produce the probability percentage for rain, such as 90% chance of rain. For these types of problems several different classification algorithms and techniques can be used (Figure 4).

When the model is trained, different labels are assigned to different data classes (such as circles and triangles) to enable the model to determine which category the data belongs to.

Where classifications can help

To understand how classification algorithms work in machine learning, consider this hypothetical business case.

A toy company needs to determine how to distribute its inventory of toys across retail locations in different countries. It cannot give each retail store the same inventory because it has learned, over the years, that some types of toys consistently sell better in certain regions than others. Some toys, in particular, hardly sell at all in some regions.

They use a classification algorithm to mine the transaction records in their database as labeled examples in order to classify toy types for each region:

• Class A: types of toys that sell at a poor rate in this region.
• Class B: types of toys that sell at an average rate in this region.
• Class C: types of toys that sell at an exceptional rate in this region.

Once the classifier is trained, all new toys types are classified into one of the three classes for each region. Finally, the toy company proceeds to set inventory quantities for the stores in the different regions using this classification system. Toys that fall under Class A are stocked in modest quantities. Toys that fall under Class B are stocked at standard quantities. Toys that fall under Class C are stocked at high quantities.

The toy company uses a further classification algorithm specifically for toys that fall under Class A to break down those toys into additional classes. They study this information to try to better understand what may have been leading to poor sales in the past. For example, causes of poor sales may have been due to inaccurate marketing, regional competitors offering similar toys at a lower cost or cultural incompatibility with some types of toys.

Decision tree

Decision tree algorithms are a form of classification algorithms that use rules organized into a flowchart. The rules are carried out sequentially so that once one decision is made, the algorithm moves on to the next

When a decision tree approach works

If our hypothetical toy company wanted to establish a new feature for its online shopping site, it might use the decision tree approach. When a customer logs in and begins browsing the store, the site needs to be able to provide recommendations of toy products that match the customer profile information.

A decision tree algorithm is developed to narrow down which products to display by carrying out a sequence of decisions. It first decides whether the customer previously purchased a toy over $100 within the past 30 days. If the customer did, it then determines whether the customer has previously purchased toys for boys or girls or both. Based on the outcome of that decision, the algorithm produces a classification which is used to choose one or more products to display on the website while the customer is browsing.

For example, the leaf classifying customers as the type to make a recent purchase over $100 for a boy (named Class_Recent_Purchase_Boy by the analyst) would be presented with the top three best-selling toys for boys on the webpage.

The leaf classifying customers as the type of customer to make a recent purchase over $100 for a girl (named Class_Recent_Purchase_Girl by the analyst) would be presented with the top three best-selling toys for girls on the webpage.

Another leaf would further classify customers who buy toys for both genders (i.e.,  toys that are gender neutral) and customers who spent less than $100 within in the last 30 days. For these customers, generic recommendations are presented on the webpage.

Regressions  

Regression is a method of modeling a target value based on independent predictors. A regression algorithm attempts to estimate the relationship across variables (Figure 6). Take, for example, the problem asked above: Based on historical weather data, what will the high temperature likely be on February 12?

Regression analysis is mostly used for tasks such as forecasting and determining cause and effect relationships. They are used when the results are expected to be quantifiable, such as a length, width, age and so on.

What a regression algorithm reveals

What if our hypothetical toy company is planning its marketing campaign for a new line of products. As part of the marketing campaign, it wants to send out promotional emails to its previous customers. Instead of sending out one email to all customers with information about all of its new products, it would like to customize the emails so they promote types of toys that certain customers are more likely to purchase.

They use a regression algorithm to match categories of toys with customer profiles to produce probabilities of certain types of customers purchasing certain types of toys. Historical customer transaction data is used to determine the types of toys a customer previously purchased, how frequently those types of toys were purchased and how much, on average, the customer spent on those types of toys. This information is used to generate the probability of a customer purchasing a new toy of the same type at a given price point.

For example, the algorithm determines that a customer who had previously purchased action figure toys five times over the past six months and who spent an average of $80 on each purchase is 71% likely to purchase a new action toy of the same type that is priced at $40.

The purchase probability of other toys is also assessed and compared, and the promotional email sent to that customer is customized to highlight the new toys with at least a 60% purchase probability at the toy’s current price.

Ensemble methods

Ensemble methods help to improve the outcome of applying machine learning by combining several learning models instead of using a single model. The ensemble learning approach results in better prediction compared to when using a single learning model.

Some industry experts refer to algorithms used in ensemble methods as meta algorithms because they combine several learning techniques into one single model and use a complex method of prediction.

How ensemble methods can help

What if the toy company has a limited number of toy samples for a new line of toys it is releasing, and it would like to send the samples only to high-value customers to reward them for their loyalty? It is challenging to determine which customers should be valued more highly than others.

The company uses an ensemble method that will be applied to a combination of models, each of which represents a factor in determining the customer’s value to the company. The following models are identified:

• total amount spent during all transactions to date;
• length of time the customer has made purchases; and
• whether the types of previous purchases are the same toy type as the sample.

The ensemble method factors in all three models. Ensemble voting is then used to determine whether the customer is considered high-value and will, therefore, receive the toy sample.

An ensemble model is the result of applying different ensemble methods. Ensemble methods can be used for stacking (improving prediction), boosting (bias) and bagging (decreasing the variance) purposes.

The post The supervised approach to machine learning appeared first on Artificial Intelligence.

Leading organizations understand the benefits they can gain by migrating existing apps to a microservices architecture and adopting the approach for new builds. However, there are inherent challenges application designers, architects and developers face around scalability, performance and deployment.

Use these 10 key microservices design principles as guidelines to build applications that meet expectations and avoid short- and long-term lifecycle issues. They range from conceptual, such as how to define the scope of a service and maintain data autonomy, to practical, such as what patterns to use to maximize network traffic performance.

1. Ensure high cohesion and low coupling

Cohesion and coupling are two terms often used interchangeably when describing a microservices architecture. The former relates to the degree of intradependence that exists between the modules of an application, and the latter is used for the degree of interdependencies.

You should design microservices so that cohesion is high and coupling is low. This plan creates microservices that are adaptable to changes, scalable and can be extended over time.

The higher the cohesion, the better, because the modules work together. If cohesion is low, the application would send too many communications back and forth between the services, causing degraded performance and scalability.

Two components are loosely coupled when they are not interdependent, i.e., if they can function without the other and if any change in one component wouldn’t break the functionality of the other. Loosely coupled components in an application should be easy to test because the component is isolatable.

2. Define the scope properly

You should define the functionality of a microservice, describing what it is intended to do. The scope of a microservice corresponds to the requirement of an independent business module. It’s important to set a proper scope for each microservice in order to rationalize its size and define its boundaries.

3. Adhere to the Single Responsibility Principle

The Single Responsibility Principle states that a class should never have more than one reason for change. This principle is essential to designing a microservices-based application, because there should not be multiple responsibilities in a single microservice.

4. Design for failure

One of the objectives of microservice architecture is to create fault-tolerant and resilient software systems. Failure or performance issues in one service should not affect other services. A memory leak, database connectivity problems or other issues in one microservice should not bring the entire application down.

Since the services in a microservices-based application are autonomous and independent, they can take advantage of the circuit breaker pattern, which is a means to cut off communication with one or more services that are down or experiencing errors.

5. Build around business capabilities

Each microservice should be designed to solve a business problem. The developer can use the appropriate technology stack for each business problem in each microservice. Unlike a monolithic application, you are not constrained to use a single best-fit homogenous technology stack for the whole architecture. This microservices design principle means developers should choose what’s best and readily available for use in every component of the application.

6. Decentralize data

Unlike in monolithic applications, microservices each maintain their own copy of the data. In other words, each microservice has its own database. You should not set up multiple services to access or share the same database, since it defeats the purpose of autonomous microservice operation.

Data pertaining to a specific microservice is private to that service. Use APIs to let other services access a microservice’s owned data. This design principle enforces centralized access control and enables the developers to implement audit logging and caching seamlessly. Aim for one or two database tables per service.

7. Gear up process automation

A microservices design can deploy in several units, which the application team must manage. Automate the deployment process for microservices-based components via smart iterative release tooling, such as a CI/CD pipeline, potentially coupled with a DevOps culture.

8. Enable interservice communication

When you migrate an existing monolithic application to microservices, you must break apart many interrelated components; these services need a way to communicate. Microservices applications also enable diverse programming languages and approaches, as explained in the fifth microservices design principle, so an application might have services built with different technologies communicating with each other. APIs make it all work.

When you set up microservices APIs, abstract the implementation details of how a service works and only expose API methods to enable external access to the service. In this setup, a microservice can scale independently.

9. Monitor constantly

Microservices in production are distributed and interrelated. It is daunting to manually discover and identify errors. Instead, use an automated monitoring system that can track performance constantly. As part of the microservices design and deployment process, select a tool or set of tools that captures and analyzes data on services’ performance, and generates useful metrics.

10. Manage traffic

Traffic to microservices in an application differs from one to the next. One service might have huge traffic while another is low-demand on the network. In each kind of traffic scenario, performance is an important factor. Take advantage of autoscaling and circuit breaker patterns to maximize performance.

The post Follow these 10 fundamental microservices design principles appeared first on Artificial Intelligence.

Fly into Britain’s Manchester Airport these days and you might spot a new landmark amid the urban sprawl on the ground below. Two huge white cylinders stand sentinel: the only outward sign of a massive biomedical project that promises a revolution in science and health care. And like all revolutions, this one is born in blood.

The cylinders pump liquid nitrogen into a facility called the U.K. Biobank. Inside the walls of this anonymous-looking industrial unit, scientists hold the bodily fluids of half a million Britons in state-of-the-art, robot-managed freezers. Research does not come more open-access than this. Blood biochemistry, genetic analysis, images of brains, hearts, and other organs — all the internal secrets of volunteers — are combined with intimate personal confessions about lifestyle, such as how many sexual partners someone’s had, how much alcohol they drink, and if they routinely drive faster than the motorway speed limit.

The results of that largesse are flowing. In a given month, dozens of scientific studies can appear based on U.K. Biobank data. They range from the curious — how many cups of coffee can safely be consumed in a single day — to the fundamental, such as the discovery that specific gene variants are associated with disease or healthy life expectancy. And in an area of research where size is crucial, such studies count their volunteers not by the hundred or the thousand, but by the hundred thousand. More than a century after Ernest Rutherford’s Manchester lab showed the world how to unlock the secrets inside the atom, the city is showcasing how Big Data can answer fundamental questions about human health.

“The U.K. Biobank is the gold standard right now,” says Josh Denny, a researcher in biomedical informatics at Vanderbilt University Medical Center in Nashville, Tennessee. “Worldwide it’s the benchmark of an open-access large database with rich information and genetics.” Denny published an article on this subject — using clinical data to get the most out of genomic research — for the Annual Review of Biomedical Data Science in 2018. “What we do when we bring health care and genetics data together is to get at the outcomes that are important to us,” he says.

Even as results emerge touching on everything from aging to susceptibility to asthma, the biobank effort isn’t without its detractors or bumps in the road. Some worry that the broad nature of the research done with the samples makes it impossible for volunteers to give proper consent. And in October, a high-profile paper was withdrawn because of technical problems in the way biobank data were analyzed.

But to scientists like Denny, the promise is clear. “This is a resource for the world,” he says.

The principle behind the U.K. Biobank is ambitious: to link health outcomes to the genetic data that pour from DNA sequencing machines across the world. Medicine traditionally is guided by a patient’s physical symptoms and measurable changes to physiology — what biologists call the phenotype. Integrating genetic data — a patient’s genotype — into these deliberations could help tailor treatments to boost their effectiveness, or even identify people at higher risk of developing a given disease, who could be offered help earlier. But to make that work, scientists need to connect the dots: match genotype to phenotype, find patterns and connections in the way people’s DNA varies and the way their health does, too.

Those connections are becoming clearer. In February this year, for example, scientists found genetic markers in the biobank data that linked high cholesterol to the development of motor neuron disease. Cholesterol-lowering drugs like statins, the results suggest, might prevent this deadly and incurable condition. Last month, a different team combed through the genetics of 334,000 of the half-million people signed up to the biobank project to identify genes associated with problematic metabolism of uric acid, which causes health problems including the painful condition gout. From head to toe, month by month scientists are using the biobank information to reveal everything from the benefits of being left-handed to the damage that diabetes can do to the heart.

The U.K. project isn’t the first to recruit volunteers to identify links between genes and disease. National efforts are also underway in Estonia, Sweden, Iceland, China, and Mexico. And back in the 1990s, the Icelandic company deCODE set out to build a database of the genes they found in the country’s population. Analysis of the Icelandic data, now owned by the U.S. biopharmaceutical giant Amgen, continues — for example, the company is now working on medicines to mimic the heart-protecting effects of a gene variant carried by one in 120 Icelanders.

Even with its high volunteer numbers, the U.K. Biobank isn’t the largest project of its type either. The British effort can call on data from some 500,000 recruits — but, as it often does, the U.S. military has gone further. In April, the U.S. Million Veterans Program signed up its 750,000th participant since it began in 2011, and still wants more to reach its eponymous goal. The MVP screens the health and genomes of veterans to probe the genetics of post-traumatic stress disorder, diabetes, heart disease, suicide prevention, and other topics of particular relevance to that community.

So, what’s so great about Great Britain’s project? Access. Other biobanks set up around the world are useful projects that can help answer some specific questions, Denny says. But it’s often difficult for outside scientists to get access to the data. Some national projects guard their secrets from foreign eyes as a way to give their own researchers a head start. Others fret about privacy and losing the trust of participants if they were to start sharing their information more widely.

The U.K. Biobank is unique because open and free data access for everyone was the plan from day one, says Rory Collins, an epidemiologist at the University of Oxford and chief executive of the U.K. Biobank project. “We wanted to build something, a resource, in the same way as they built CERN,” the European particle physics lab near Geneva, he says. “This wasn’t a grant application which has to have a specific hypothesis.” It’s a point that other people attached to the Biobank project make repeatedly: This is a basic science project. If they built it, they thought that scientists would come and want to use it.

They have come, and continue to do so. At last count, 13,000 scientists in 77 countries, from Australia and Malaysia to Russia and Jordan, have been given access to data on topics from cognition and sleep to mental health.

Prompted by a call from British scientists to invest in the promise of DNA, the biobank started life as a funding pledge from Tony Blair’s new Labour government in 1998. Backed by the Medical Research Council, a state funder, and the Wellcome Trust, a biomedical charity, the project was based on the principles of the famous Framingham Heart Study, an influential population cohort study that followed 5,200 residents of Framingham, Massachusetts, as a way to find factors that influence cardiac illness.

The U.K. started to recruit volunteers to its study in 2006 and reached its half-million goal four years later. It focused on individuals ages 40 to 69 because organizers figured it would be most useful to study older people, who tend to more quickly show the signs of ill health that researchers are interested in. (Indeed, the carefully preserved samples at the Manchester HQ now represent the earthly remains of at least 20,000 volunteers who have since passed away.)

Participants weren’t paid and had to spend hours at one of several regional centers, where they surrendered blood and urine, had their health examined and filled in surveys on their habits and lifestyle. As a result, the biobank population is not as diverse as geneticists might like, Collins admits, especially if the results are supposed to be useful around the world. Some 94 percent of people the biobank signed up are white, and certain socioeconomic groups, including young, low-income white men, are underrepresented.

Initially, the blood samples were analyzed for simple variations in genetic sequence, such as single nucleotide polymorphisms. These single base-pair changes in DNA occur at specific places in the genome and can explain traits such as eye color and inherited diseases such as cystic fibrosis and sickle-cell anemia. They also act as markers to indicate risk of complex diseases, including diabetes and Alzheimer’s.

The first of these genotyping data were released for 150,000 biobank participants in May 2015. Results from the other 350,000 were added two years later. That fulfilled the original plan, but as genetic sequencing has become faster and cheaper, other researchers wanted to go further. In 2017, the drug firms GSK and Regeneron offered to sequence the “exome” of 50,000 U.K. Biobank participants. This gives a readout of sections of DNA that actually code for proteins, and is seen as a more powerful way to locate information that could be used to develop medicines.

The companies agreed to pay the bill, but wanted something in return: exclusive access to the data. They were given 6 to 12 months, then in 2019 the information was released to the wider scientific community. A larger group of pharma companies is working on exome sequences for the remaining 450,000 volunteers under the same arrangement.

The post How big data can answer fundamental questions about human health appeared first on Artificial Intelligence.

Most discussions about artificial intelligence are characterised by hyperbole and hysteria. Though some of the world’s most prominent and successful thinkers regularly forecast that AI will either solve all our problems or destroy us or our society, and the press frequently reports on how AI will threaten jobs and raise inequality, there’s actually very little evidence to support these ideas. What’s more, this could actually end up turning people against AI research, bringing significant progress in the technology to a halt.

The hyperbole around AI largely stems from its promotion by tech-evangelists and self-interested investors. Google CEO Sundar Pichaideclared AI to be “probably the most important thing humanity has ever worked on”. Given the importance of AI to Google’s business model, he would say that.

Some even argue that AI is a solution to humanity’s fundamental problems, including death, and that we will eventually merge with machines to become an unstoppable force. The inventor and writer Ray Kurzweil has famously argued this “Singularity” will occur by as soon as 2045.

The hysteria around AI comes from similar sources. The likes of physicist Stephen Hawking and billionaire tech entrepreneur Elon Musk warned that AI poses an existential threat to humanity. If AI doesn’t destroy us, the doomsayers argue, then it may at least cause mass unemployment through job automation.

The reality of AI is currently very different, particularly when you look at the threat of automation. Back in 2013, researchers estimated that, in the following ten to 20 years, 47% of jobs in the US could be automated. Six years later, instead of a trend towards mass joblessness, we’re in fact seeing US unemployment at a historic low.

Even more job losses have been threatened for the EU. But past evidenceindicates otherwise, given that between 1999 and 2010, automation created 1.5 million more jobs than it destroyed in Europe.

AI is not even making advanced economies more productive. For example, in the ten years following the financial crisis, labour productivity in the UK grew at its slowest average rate since 1761. Evidence shows that even global superstar firms, including firms who are among the top investors in AI and whose business models depend on it – such as Google, Facebook and Amazon – have not become more productive. This contradicts claims that AI will inevitably enhance productivity.

So why are the society-transforming effects of AI not materialising? There are at least four reasons. First, AI diffuses through the economy much slower than most people think. This is because most current AI is based on learning from large amounts of data and it is especially difficult for most firms to generate enough data to make the algorithms efficient or simply to afford to hire data analysts. A manifestation of the slow diffusion of AI is the growing use of “pseudo-AI” where a firm appears to use an online AI bot to interact with customers but which is in fact a human operating behind the scenes.

The second reason is AI innovation is getting harder. Machine learning techniques that have driven recent advances may have already producedtheir most easily reached achievements and now seem to be experiencing diminishing returns. The exponentially increasing power of computer hardware, as described by Moore’s Law, may also be coming to an end.

Related to this is the fact that most AI applications just aren’t that innovative, with AI mostly used to fine-tune and disrupt existing products rather than introduce radically new products. For example, Carlsberg is investing in AI to help it improve the quality of its beer. But it is still beer. Heka is a US company producing a bed with in-built AI to help people sleep better. But it is still a bed.

Third, the slow growth of consumer demand in most Western countries makes it unprofitable for most businesses to invest in AI. Yet this kind of limit to demand is almost never considered when the impacts of AI are discussed, partly because academic models of how automation will affect the economy are focused on the labour market and/or the supply side of the economy.

Fourth, AI is essentially not really being developed for general application. AI innovation is overwhelmingly in visual systems, ultimately aimed for use in driverless cars. Yet such cars are most notable for their absence from our roads, and technical limits mean they are likely to remain so for a long time.

New thinking needed

Of course, AI’s small impact in the recent past doesn’t rule out larger impacts in the future. Unexpected progress in AI could still lead to a so-called robocalypse. But it will have to come from a different kind of AI. What we currently call artificial intelligence – big data and machine learning – is not really intelligent. It is essentially correlation analysis, looking for patterns in data. Machine learning generates predictions, not explanations. In contrast, human brains are storytelling devices generating explanations.

As a result of the hype and hysteria, many governments are scrambling to produce national artificial intelligence strategies. International organisations are rushing to be seen to take action, holding conferencesand publishing flagship reports on the future of work. For example, the United Nations University Centre for Policy Research claims that AI is “transforming the geopolitical order” and, even more incredibly, that “a shift in the balance of power between intelligent machines and humans is already visible”.

This unhinged debate about the current and near-future state of AI threatens both an AI arms race and stifling regulations. This could lead to inappropriate controls and moreover loss of public trust in AI research. It could even hasten another AI-winter – as occurred in the 1980s – in which interest and funding disappear for years or even decades after a period of disappointment. All at a time when the world needs more, not less, technological innovation.

The post No, artificial intelligence won’t start a robocalypse or solve humanity’s fundamental problems appeared first on Artificial Intelligence.