In this subsection, we present details on how we process the dataset, turn it into a network graph and finally how we produce, and process features that belong to the graph. Topics to be covered are:

splitting the data,

preprocessing,

feature importance and selection,

computation of similarity between samples, and

generating of the raw graph.

After preprocessing the data, the next step is to split the dataset into training and test samples for validation purposes. We selected cross-validation (CV) as the validation method since it is the de facto standard in ML research. For CV, the full dataset is split into k folds; and the classifier model is trained using data from (k-1) folds then tested on the remaining kth fold. Eventually, after k iterations,, average performance scores (like F1 measure or ROC) of all folds are used to benchmark the classifier model.

A crucial step of CV is selecting the right proportion between the training and test subsamples, i.e., number of folds. Determining the most appropriate number of folds k for a given dataset is still an open research question17, besides de facto standard for selecting k is accumulated around k=2, k=5, or k=10. To address the selection of the right fold size, we have identified two priorities:

Priority 1Class Balance: We need to consider every split of the dataset needs to be class-balanced. Since the number of class types has a restrictive effect on selecting enough similar samples, detecting the effective number of folds depends heavily on this parameter. As a result, whenever we deal with a problem which has low represented class(es), we selected k=2.

Priority 2High Representation: In our model, briefly, we build a network from the training subsamples. Efficient network analysis depends on the size (i.e., number of nodes) of the network. Thus, maximize training subsamples with enough representatives from each class (diversity) is our priority as much as we can when splitting the dataset. This way we can have more nodes. In brief, whenever we do not cross priority 1, we selected k=5.

By balancing these two priorities, we select efficient CV fold size by evaluating the characteristics of each datasets in terms of their sample size and the number of different classes. The selected fold value for each dataset will be specified in the Experiments and results section. To fulfill the class balancing priority, we employed stratified sampling. In this model, each CV fold contains approximately the same percentage of samples of each target class as the complete set.

Preprocessing starts with the handling of missing data. For this part, we preferred to omit all samples which have one or more missing feature(s). By doing this, we have focused merely on developing the model, skipping trivial concerns.

As stated earlier, GSNAc can work on datasets that may have both numerical and categorical values. To ensure proper processing of those data types, as a first step, we separate numerical and categorical features18. First, in order to process them mathematically, categorical (string) features are transformed into unique integers for each unique category by a technique called labelization. It is worth noting that, against the general approach, we do not use the one-hot-encoding technique for transforming categorical features, which is the method of creating dummy binary-valued features. Labelization does not generate extra features, whereas one-hot-encoding extend the number of features.

For the numerical part, as a very important stage of preprocessing, scaling19 of the features follows. Scaling is beneficial since the features may have a very different range and this might affect scale-dependent processes like distance computation. We have two generally accepted scaling techniques which are normalization and standardization. Normalization transforms features linearly into a closed range like [0, 1], which does not affect the variation of values among features. On the other hand, standardization transforms the feature space into a distribution of values that are centered around the mean with a unit standard deviation. This way, the mean of the attribute becomes zero and the resultant distribution has a unit standard deviation. Since GSNAc is heavily dependent on vectorial distances, we do not prefer to lose the structure of the variation within features and this way our choice for scaling the features becomes normalization. Here, it is worth mentioning that all the preprocessing is applied on the training part of the data and transformed on the test data, ensuring no data leakage occurs.

Feature Importance (FI) broadly refers to the scoring of features based on their usefulness in prediction. It is obvious that in any problem some features might be more definitive in terms of their predictive capability of the class. Moreover, a combination of some features may have a higher effect than others in total than the sum of their capacity in this sense. FI models, in general, address this type of concern. Indeed, almost all ML classification algorithms use a FI algorithm under the hood; since this is required for the proper weighting of features before feeding data into the model. It is part of any ML classifier and GSNAc. As a scale-sensitive model, vectorial similarity needs to benefit much from more distinctive features.

For computing feature importance, we preferred to use an off-the-shelf algorithm, which is a supervised k-best feature selection18 method. The K-best feature selection algorithm simply ranks all features by evaluating features ANOVA analysis against class labels. ANOVA F-value analyzes the variance between each feature and its respective class and computes F-value which is the ratio of the variation between sample means, over the variation within the samples. This way, it assigns F values as features importance. Our general strategy is to keep all features for all the datasets, with an exception for genomic datasets, that contain thousands of features, we practiced omitting. For this reason, instead of selecting some features, we prefer to keep all and use the importance learned at this step as the weight vector in similarity calculation.

In this step, we generate an undirected network graph G, its nodes will be the samples and its edges will be constructed using the distance metrics20 between feature values of the samples. Distances will be converted to similarity scores to generate an adjacency matrix from the raw graph. As a crucial note, we state that since we aim to predict test samples by using G, in each batch, we only process the training samples.

In our study for constructing a graph from a dataset we defined edge weights as the inverse of the Euclidean distance between the sample vectors. Simply, Euclidean distance (also known as L2-norm) gives the unitless straight line (shortest) distance between two vectors in space. In formal terms, for f-dimensional vectors u and v, Euclidean distance is defined as:

$$dleft(u,vright)=sqrt[2]{sum_{f}{left({u}_{i}-{v}_{i}right)}^{2}}$$

A slightly modified use of the Euclidean distance is introducing the weights for dimensions. Recall from the discussion of the feature importance in the former sections, some features may carry more information than others. So, we addressed this factor by computing a weighted form of L2 norm based on distance which is presented as:

$${dist_L2}_{w}left(u,vright)=sqrt[2]{sum_{f}{{w}_{i}({u}_{i}-{v}_{i})}^{2}}$$

where w is the n-dimensional feature importance vector and i iterates over numerical dimensions.

The use of the Euclidean distance is not proper for the categorical variables, i.e. it is ambiguous and not easy to find how much a canarys habitat sky is distant from a sharks habitat sea. Accordingly, whenever the data contains categorical features, we have changed the distance metric accordingly to L0 norm. L0 norm is 0 if categories are the same; it is 1 whenever the categories are different, i.e., between the sky and the sea L0 norm is 1, which is the maximum value. Following the discussion of weights for features, the L0 norm is also computed in a weighted form as ({dist_L0}_{w}left(u,vright)=sum_{f}{w}_{j}(({u}_{j}ne {v}_{j})to 1)), where j iterates over categorical dimensions.

After computing the weighted pairwise distance between all the training samples, we combine numerical and categorical parts as: ({{dist}_{w}left(u,vright)}^{2}={{dist_L2}_{w}left(u,vright)}^{2}+ {{dist_L0}_{w}left(u,vright)}^{2}). With pairwise distances for each pair of samples, we get a n x n square and symmetric distance matrix D, where n is the number of training samples. In matrix D, each element shows the distance between corresponding vectors.

$$D= left[begin{array}{ccc}0& cdots & d(1,n)\ vdots & ddots & vdots \ d(n,1)& cdots & 0end{array}right]$$

We aim to get a weighted network, where edge weights represent the closeness of its connected nodes. We need to first convert distance scores to similarity scores. We simply convert distances to similarities by subtracting the maximum distance on distances series from each element.

$$similarity_s(u,v)=mathrm{max}_mathrm{value}_mathrm{of}(D)-{dist}_{w}left(u,vright)$$

Finally, after removing self-loops (i.e. setting diagonal elements of A to zero), we use adjacency matrix A to generate an undirected network graph G. In this step, we delete the lower triangular part (which is symmetric to the upper triangular part) to avoid redundancy. Note that, in transition from the adjacency matrix to a graph, the existence of a (positive) similarity score between two samples u and v creates an edge between them, and of course, the similarity score will serve as the vectorial weight of this particular edge in graph G.

$$A= left[begin{array}{ccc}-& cdots & s(1,n)\ vdots & ddots & vdots \ -& cdots & -end{array}right]$$

The raw graph generated in this step is a complete graph: that is, all nodes are connected to all other nodes via an edge having some weight. Complete graphs are very complex and sometimes impossible to analyze. For instance, it is impossible to produce some SNA metrics such as betweenness centrality in this kind of a graph.

View post:
Explainable artificial intelligence through graph theory by generalized social network analysis-based classifier | Scientific Reports - Nature.com

This episode originally aired on July 20, 2022.

Advanced computer programs influence, and can even dictate, meaningful parts of our lives. Think streaming services, credit scores, facial recognition software.

And as this technology becomes more sophisticated and more pervasive, its important to understand some basic terminology.

On this Labor Day, were revisiting an episode in which we explore the terms algorithm, machine learning and artificial intelligence. Theres overlap, but theyre not the same things.

We called up a few experts to help us get a firm grasp on these concepts, starting with a basic definition of algorithm. The following is an edited transcript of the episode.

Melanie Mitchell, Davis professor of complexity at the Santa Fe Institute, offered a simple explanation of a computer algorithm.

An algorithm is a set of steps for solving a problem or accomplishing a goal, she said.

The next step up is machine learning, which uses algorithms.

Rather than a person programming in the rules, the system itself has learned, Mitchell said.

For example, speech recognition software, which uses data to learn which sounds combine to become words and sentences. And this kind of machine learning is a key component of artificial intelligence.

Artificial intelligence is basically capabilities of computers to mimic human cognitive functions, said Anjana Susarla, who teaches responsible AI at Michigan State Universitys Broad College of Business.

She said we should think of AI as an umbrella term.

AI is much more broader, all-encompassing, compared to only machine learning or algorithms, Susarla said.

Thats why you might hear AI as a loose description for a range of things that show some level of intelligence. Like software that examines the photos on your phone to sort out the ones with cats to advanced spelunking robots that explore caves.

Heres another way to think of the differences among these tools: cooking.

Bethany Edmunds, professor and director of computing programs at Northeastern University, compares it to cooking.

She says an algorithm is basically a recipe step-by-step instructions on how to prepare something to solve the problem of being hungry.

If you took the machine learning approach, you would show a computer the ingredients you have and what you want for the end result. Lets say, a cake.

So maybe it would take every combination of every type of food and put them all together to try and replicate the cake that was provided for it, she said.

AI would turn the whole problem of being hungry over to the computer program, determining or even buying ingredients, choosing a recipe or creating a new one. Just like a human would.

So why do these distinctions matter? Well, for one thing, these tools sometimes produce biased outcomes.

Its really important to be able to articulate what those concerns are, Edmunds said. So that you can really dissect where the problem is and how we go about solving it.

Because algorithms, machine learning and AI are pretty much baked into our lives at this point.

Columbia Universitys engineering school has a further explanation of artificial intelligence and machine learning, and it lists other tools besides machine learning that can be part of AI. Like deep learning, neural networks, computer vision and natural language processing.

Over at the Massachusetts Institute of Technology, they point out that machine learning and AI are often used interchangeably because these days, most AI includes some amount of machine learning. A piece from MITs Sloan School of Management also gets into the different subcategories of machine learning. Supervised, unsupervised and reinforcement, like trial and error with kind of digital rewards. For example, teaching an autonomous vehicle to drive by letting the system know when it made the right decision not hitting a pedestrian, for instance.

That piece also points to a 2020 survey from Deloitte, which found that 67% of companies were already using machine learning, and 97% were planning to in the future.

IBM has a helpful graphic to explain the relationship among AI, machine learning, neural networks and deep learning, presenting them as Russian nesting dolls with the broad category of AI as the biggest one.

And finally, with so many businesses using these tools, the Federal Trade Commission has a blog laying out some of the consumer risks associated with AI and the agencys expectations of how companies should deploy it.

Read more:
The ABCs of AI, algorithms and machine learning (re-air) - Marketplace

Image: LNER

Digital transformation is all about delivering change, so how do you do that in an industry that's traditionally associated with largescale infrastructures and embedded operational processes?

Danny Gonzalez, chief digital and innovation officer (CDIO) at London North Eastern Railway (LNER), says the answer is to place technology at the heart of everything your business does.

"We firmly believe that digital is absolutely crucial," he says. "We must deliver the experiences that meet or exceed customers' expectations."

Delivering to that agenda is no easy task. Gonzalez says the rail journey is "absolutely full" of elements that can go wrong for a passenger, from buying a ticket, to getting to the train station, to experiencing delays on-board, and onto struggling to get away from the station when they reach their destination.

SEE: Digital transformation: Trends and insights for success

LNER aims to fix pain points across customer journeys, but it must make those changes in a sector where legacy systems and processes still proliferate. Gonzalez says some of the technology being used is often more than 30 years' old.

"There's still an incredible amount of paper and spreadsheets being used across vast parts of the rail industry," he says.

"Our work is about looking at how things like machine learning, automation and integrated systems can really transform what we do and what customers receive."

Gonzalez says that work involves a focus on the ways technology can be used to improve how the business operates and delivers services to its customers.

This manifests as an in-depth blueprint for digital transformation, which Gonzalez refers to as LNER's North Star: "That gives everyone a focus on the important things to do."

As CDIO, he's created a 38-strong digital directorate of skilled specialists that step out of traditional railways processes and governance and into innovation and the generation of creative solutions to intractable challenges.

"It's quite unusual for a railway company to give more permission for people to try things and fail," he says.

Since 2020, the digital directorate in combination with its ecosystem of enterprise and startup partners has launched more than 60 tools and trialled 15 proof-of-concepts.

One of these concepts is an in-station avatar that has been developed alongside German national railway company Deutsche Bahn AG.

LNER ran a trial in Newcastle that allowed customers to interact in free-flowing conversations with an avatar at a dedicated booth at the station. The avatar plugged into LNER's booking engine, so customers could receive up-to-date information on service availability. Following the successful trial, LNER is now looking to procure a final solution for wider rollout.

The company is also working on what Gonzalez refers to as a "door-to-door" mobility-as-a-service application, which will keep customers up to date on the travel situation and provide hooks into other providers, such as taxi firms or car- and bike-hire specialists.

"It's about making sure the whole journey is seamlessly integrated," he says. "As a customer, you feel in control and you know we're making sure that if anything is going wrong through the process that we're putting it right."

When it comes to behind-the-scenes operational activities, LNER is investing heavily in machine-learning technology. Gonzalez's team has run a couple of impactful concepts that are now moving into production.

SEE:What is digital transformation? Everything you need to know about how technology is reshaping business

One of these is a technology called Quantum, which processes huge amounts of historical data and helps LNER's employees to reroute train services in the event of a disruption and to minimise the impact on customers.

"Quantum uses machine learning to learn the lessons of the past. It looks at the decisions that have been made historically and the impact they have made on the train service," he says.

Gonzalez: "We firmly believe that digital is absolutely crucial."

"It computes hundreds of thousands of potential eventualities of what might happen when certain decisions are made. It's completely transforming the way that our service delivery teams manage trains when there's disruption to services."

To identify and exploit new technologies, Gonzalez's team embracesconsultant McKinsey's three horizon model, delivering transformation across three key areas that allows LNER to assess potential opportunities for growth without neglecting performance in the present.

Horizon one focuses on "big, meaty products" that are essential to everyday operations, such as booking and reservations systems, while horizon two encompasses emerging opportunities that are currently being scoped out by the business.

Gonzalez says a lot of his team's activity is now focused on horizon three, which McKinsey suggests includes creative ideas for long-term profitable growth.

He says that process involves giving teams quite a lot of freedom to get on and try stuff, run proof of concepts, and actually understand where the technology works.

Crucial to this work isan accelerator called FutureLabs, where LNER works with the startup community to see if they can help push digital transformation in new and exciting directions.

"We go out with key problem statements across the business and ask the innovators to come and help us solve our challenges and that's led to some of the most impactful things that we've done as a business," says Gonzalez.

FutureLabs has already produced pioneering results. Both the Quantum machine-learning tool and the "door-to-door" mobility service have been developed alongside startup partners JNCTION and IOMOB respectively.

LNER continues to search for new inspiration and has just run the third cohort of its accelerator. Selected startups receive mentoring and funding opportunities to develop and scale up technology solutions.

Gonzalez says this targeted approach brings structure to LNER's interactions and investments in the startup community and that brings a competitive advantage.

"It's not like where I've seen in other places, where innovation initiatives tend to involve 'spray and pray'," he says. "The startups we work with are clear on the problems they're trying to solve, which leads to a much greater success rate."

SEE: Four ways to get noticed in the changing world of work

Gonzalez's advises other professionals to be crystal clear on the problems they're trying to solve through digital transformation.

"Know what the priorities are and bring the business along with you. Its really important the business understands the opportunities digital can bring in terms of how you work as an organisation," he says.

"We're fortunate that we've got a board that understood that rail wasn't where it needed to be in terms of its digital proposition. But we've put a lot of work into creating an understanding of where issues existed and the solutions that we needed if we're going to compete in the future."

Go here to read the rest:
How avatars and machine learning are helping this company to fast track digital transformation - ZDNet

Since the pandemic there has been a race to go digital, organisations and individuals alike. Remote working, remote learning, contactless payments, and online shopping has become the mainstay of our lives. While technology really helped us get through the pandemic, it came with a learning curve, the lack of awareness about dos and donts made many easy targets for hackers and hacking organisations. The rush to go digital exposed a lot of companies to cyber-attacks through ransomware, data breaches, and leaks. The Pegasus and Predator scandals also showed us that hacking now is a business and Politicians are also using the services of these companies for their own benefit. As technology moves ahead, the risk of cyber-attacks increases as well, to explain a bit more about the current cyber security scenario Siddharth Shankar from TimesNow speaks to Asaf Hecht, from CyberArk. Asaf manages one of the research groups in CyberArk Labs. He focuses on researching and discovering the latest attack techniques and applying lessons learned to improve cyber defenses. Prior to CyberArk, Asaf served for eight years in the Israeli Army, as a skilled helicopter pilot and as Team Leader for the advanced cyber-hunting team, an elite force that protects military top-secret networks and reveals APTs.

Excerpts

Asaf: You are absolutely correct about the situation during COVID, We saw it in the entire world that people and organization go on digital and work from home like me today, and everything is going for the internet and there is also a paradigm shift. The legacy of the traditional perimeter is finished. When you have a company and you are in a physical building, you can secure this building with what is going out and where is the entrance and exit.

Nowadays, everything is everywhere, and this is the need of the hour. The trends that you mentioned are correct. Also, contactless payment is also gaining popularity in countries like Israel. From bigger organizations to private ones to even individuals. From the youngest who thrive to use new technology to older people that don't have other options because the bank or store in the neighborhood was closed because it converted online. However, there are a few things that we need to remember regarding the outlook toward risk, it's different from the organisation side and different from the individual side of things.

For individuals, I think awareness is important to everyone and every age. Having said that, I think there are some principles that are very important.

For example, check your bank account and your credit card payment at least once per month. This is because there is a high chance of these kinds of things occurring without your knowledge. If you get scammed and phished and someone stole your credit card and pays for his desire. The most important thing for you is to detect it and then you can automatically alert the credit card company and also you will get a full refund from credit card company. The first principle, again simple awareness. The second principle is to check what goes out from your accounts and submit it for refund if something is wrong.

The third principle, the normal public is not the target, this is a big difference between attacks on organisations and attacks on individuals. While I believe threat actors can attack everyone, everything, and every device, it is only a matter of how much time is consumed, the budget and what are they gaining out of it. I can imagine myself if I have a phone and I don't think someone will want access to my phone, and it will offer him $10,000. I am sure it's not. There is nothing important there.

Siddharth - Phishing and getting money out of people or crypto wallets are quite common, but lately, we have been seeing a lot of state-sponsored threat actors and we have also been seeing private companies which are actually working only on finding exploits, finding zero-day hacks and then selling it to governments and making a big chunk of money. So now this is probably becoming a business as well and a lot of them actually stem from Israel. What are your thoughts on this? Is this quickly becoming a business prospect for hackers?

Asaf - I think it's an interesting trend and I agree with it. In the recent decade, there are few private companies that are mentioning that they do is to develop technology solutions for gaining access and intelligence. I think that in Israeli Cyber intelligence, getting cyber intelligence or getting cyber access to a device or target, has been around for over two decades, and then what happened is the people who were in the army service, some of them completed their service and they still wanted to do what they did. Also, we need to remember that most of the usage of these technologies is for humanitarian reasons for anti-terror fighting and for making sure there is no major terror attack. The fact is that for 20 years, we didn't have a devastating terror attack again (in Israel). And I think a major thing that helped this fact apart from other countermeasures is their demonstration is this kind of spying technologies that sometimes also comes from the private sector.

The challenge and the problem emanate from how you make sure that these kinds of technologies are being handled and sold to the right target. This is a problem, but these are the two sides of things. The world I think needed this kind of spying company to make it a safer place, but the problem from the other side is how to monitor and who these companies sell to and even if they sell it to a government, maybe the government will say, yeah. We are going to use it on a valid target, but I think these private companies don't really can audit, the usage of this third-party government.

Siddharth - Back in 2011, Bill Gates had said that the next big challenge for the world will not be a nuclear war, it would be a virus. We had COVID and the whole world just stopped. Do you think in the future, a full blown war like the Russia-Ukraine conflict will not happen and it will be more cyber warfare?

Asaf - I think there are more challenges, but I think cybersecurity issues are gaining more value because more assets are converting to be online and again, our daily things are online, and sometimes even from the army's perspective, its easier to do something behind the keyboard and not blast anyone and risk your people going to war, I think it's a future threat. While saying it, I think it's also there is a balance thats kind of the nuclear balance that both of the sides have nuclear power, so no one uses it. I think it might go to this one. Maybe the country could devastate the other country in the cyber war, but I think they understand that there might be attacks on the same power as well. It might be a threat that is above our head, but not really will be done with 100% power. I do think and we already saw that in a low power cyber attacks already happening right? Also, even in Ukraine and Russia conflict, Russia on the starting day of the conflict, wiped out hundreds of machines and denial of services attacks on websites in Ukraine and so on.

Siddharth - Asaf now lately, we have seen Predator, we have seen ERMAC, Follina, and a lot of other malware or ransomware coming out. Why is this happening? We have the internet the knowledge is there, and the news spreads. Yet all these things are happening so much more today than say 5 to 10 years ago.

Asaf - Yes, with the popularity of the internet and technologies and phone devices and everything is computerized, and I think that awareness and even the availability of knowledge is very easy to gain for everyone.

As an example, the lapsus$ or phishing attack, will still probably work. It is really a problem, but again we should sleep well, there are cybersecurity vendors out there doing our best. For example, at CyberArk, we try to help organizations across the world and so it will be harder for attackers to achieve their goal also if they attack a company, the damage will be reduced a lot. They will not be at a total loss, and we also see this from the other side of your question, we see more attacking groups yes, and ransomware campaigns because they have money and there are more options for people and also to build an organization and a business. As an example, there is Conti, an attacking group that does ransomware mainly. It's built like a regular company, there is human resources, HR, there is R&D, and there is a kind of marketing to make the tool available to a paying audience. Yeah, this is kind of the new world.

Siddharth - You mentioned the supply chain attacks. Now if the supply chain is crippled for a big company like say Samsung or Apple. It is going to cause a lot of damage and damage reduction will be the biggest thought once the attack has happened. What are the things they should keep in mind before an attack happens and after an attack happens to minimize the damage to them and their consumers?

Asaf - Before the attack happens, we should make sure that our network is there in the most secure place and in the most secure state. There are many protocols and steps and standards. One of the main things is to secure privileged access security and secure identity security. Nowadays, it's not only devices and laptops and phones, but also more of the identity that uses this computer in this form because it could be many identities on the same device, and multiple identities or specific identities could be accessing across multiple devices. We need to secure the focus on identity. How it has been authenticated, what it does and there are again many solutions that can help with this. I would focus on securing the identities and of course making sure to check all the standards.

If we do the preparation right in stage 1, the damage will be limited because one identity will be compromised and one network will be compromised, but the sensitive database is on a different network and there is a segmentation in the network, and so on. If we did the preparation right then the damage should be limited, but still, we should also prepare for this compromise because it might happen at any time and we should also practice it. I think most organisations will suffer from this kind or another compromise, but good preparation will limit the damage when it occurs.

Siddharth - What would your forecast be in terms of trends of security that we will be seeing in the future, like supply chain is one, next what could be it?

Asaf - Interesting question. I think cloud will be major as nowadays cloud is a popular for the technology benefits and so on and I think now that cloud services are being used much more, the attacks on this kind of scenario will be more popular. Some unique specific services like database on cloud and SQL on cloud and virtual machine on cloud and things like this.

Another thing I might say is the attacks in the future will be about machine learning and automation around attacking and automation around the discovery of vulnerabilities, open source is also a popular vector because nowadays because technology is so complex, we have several components on every product. Open source is also another vector.

Siddharth - During WWDC, Apple announced something about a passwordless feature. Do you think this is an interesting concept that will increase security?

Asaf - Yes. There are several disadvantages of having a password. Of course, its hard to remember, people tend to use the same password for 2 different services and so on. The trend of a passwordless future, I think it's good. Mainly it involves some other device or multifactorial with your phone. The passwordless thing is a good solution, From our vulnerability research, we saw after the authentication has been successfully done, it's still a token or digital token or certificate that is being stored in the computer and a device in the cloud. After the authentication phase, the token is not really authenticated more. Nowadays there is a new trend of continuous authentication. You want to continuously authenticate the identity and what it does.

Continued here:
Cyber-attacks in future will be about machine learning and automation around attacking and discovery of vulner - Times Now