Learning Archives - Business Intelligence Info

Tag Archives: Learning

AI Weekly: Machine learning could lead cybersecurity into uncharted territory

February 15, 2020 Big Data

Once a quarter, VentureBeat publishes a special issue to take an in-depth look at trends of great importance. This week, we launched issue two, examining AI and security. Across a spectrum of stories, the VentureBeat editorial team took a close look at some of the most important ways AI and security are colliding today. It’s a shift with high costs for individuals, businesses, cities, and critical infrastructure targets — data breaches alone are expected to cost more than $ 5 trillion by 2024 — and high stakes.

Throughout the stories, you may find a theme that AI does not appear to be used much in cyberattacks today. However, cybersecurity companies increasingly rely on AI to identify threats and sift through data to defend targets.

Security threats are evolving to include adversarial attacks against AI systems; more expensive ransomware targeting cities, hospitals, and public-facing institutions; misinformation and spear phishing attacks that can be spread by bots in social media; and deepfakes and synthetic media have the potential to become security vulnerabilities.

In the cover story, European correspondent Chris O’Brien dove into how the spread of AI in security can lead to less human agency in the decision-making process, with malware evolving to adapt and adjust to security firm defense tactics in real time. Should costs and consequences of security vulnerabilities increase, ceding autonomy to intelligent machines could begin to seem like the only right choice.

We also heard from security experts like McAfee CTO Steve Grobman, F-Secure’s Mikko Hypponen, and Malwarebytes Lab director Adam Kujawa, who talked about the difference between phishing and spear phishing, addressed an anticipated rise in personalized spear phishing attacks ahead, and spoke generally to the fears — unfounded and not — around AI in cybersecurity.

VentureBeat staff writer Paul Sawers took a look at how AI could be used to reduce the massive job shortage in the cybersecurity sector, while Jeremy Horwitz explored how cameras in cars and home security systems equipped with AI will impact the future of surveillance and privacy.

AI editor Seth Colaner examines how security and AI can seem heartless and inhuman but still relies heavily on people, who are still a critical factor in security, both as defenders and targets. Human susceptibility is still a big part of why organizations become soft targets, and education around how to properly guard against attacks can lead to better protection.

We don’t know yet the extent to which those carrying out attacks will come to rely on AI systems. And we don’t know yet if open source AI opened Pandora’s box, or to what extent AI might increase threat levels. One thing we do know is that cybercriminals don’t appear to need AI to be successful today.

I’ll leave it to you to read the special issue and draw your own conclusions, but one quote worth remembering comes from Shuman Ghosemajumder, formerly known as the “click fraud czar” at Google and now CTO at Shape Security, in Sawers’ article. “[Good actors and bad actors] are both automating as much as they can, building up DevOps infrastructure and utilizing AI techniques to try to outsmart the other,” he said. “It’s an endless cat-and-mouse game, and it’s only going to incorporate more AI approaches on both sides over time.”

For AI coverage, send news tips to Khari Johnson and Kyle Wiggers and AI editor Seth Colaner — and be sure to subscribe to the AI Weekly newsletter and bookmark our AI Channel.

Thanks for reading,

Khari Johnson

Senior AI Staff Writer

Let’s block ads! (Why?)

Big Data – VentureBeat

7 Questions on How to Use Machine Learning for Anomaly Detection

January 29, 2020 TIBCO Spotfire

Reading Time: 3 minutes

Asking questions is one of the best ways to learn. But sometimes you don’t know where to start or what to ask—especially on a topic like anomaly detection that you’re still becoming familiar with. In that case, it’s best to listen to the questions of others and let their line of thinking guide your learning. Below are some questions we received during our “Ask Me Anything: Anomaly Detection” webinar to help you get started.

What’s the difference between outliers and anomalies?

Outliers are observations that are distant from the mean or location of a distribution. However, they don’t necessarily represent abnormal behavior or behavior generated by a different process. On the other hand, anomalies are data patterns that are generated by different processes.

Are there any applications for anomaly detection in pharmaceuticals?

There are many applications of anomaly detection in the pharmaceutical life sciences space. Including, process monitoring and quality control using Statistical Process Control (SPC) or Quality Control (QC) and Multivariate SPC (MSPC) charts in pharmaceutical manufacturing. Timely anomaly detection is critical to avoid abnormal events and adhere to safety standards. Identifying anomalies in over-the-counter transactions can be used to fight prescription abuse in pharma retail data. And real-time detection of anomalies in multi-parameter clinical trial data helps ensure the success of clinical trials.

Are GANs also used for anomaly detection? If so, could you please provide an industry use case.

Generative Adversarial Networks (GANs) are novel unsupervised learning methods that are very efficient in identifying anomalies. Since GANs are iterative by design and the adversarial training is optimized to leverage reconstructed samples to reduce residual loss, they work well in semi-structured and unstructured data. They are specifically very useful for medical image analysis (helping radiologists find hard-to-identify tumors), facial recognition, text to image translation, etc.

Does correlation data affect anomaly detection? In what ways and how could we diminish the effects? Is it preferable to clean and remove correlations before starting anomaly detection?

As mentioned in the webinar, we do not think that correlations would affect anomaly detection, but we have many techniques available to help determine how to treat correlated variables. One suggestion would be to reduce the number of dimensions with a technique like Principle Component Analysis (PCA).

What is a suggested algorithm that would be appropriate for anomaly detection related to identifying unusual activities in network activities or data?

As mentioned in the webinar, there are many methods and algorithms that work well for various applications and use cases of anomaly detection. Some of them are Recurrent Neural Network (RNN), Generative Adversarial Network (GAN), Isolation forests, Deep Autoencoders, etc. If you are specifically interested in Network/Graph analytics, the two main methods used for identifying anomalies in network graphs are the Direct Neighbour Outlier Detection Algorithm (DNODA) and Community Neighbour Algorithm (CNA).

In my current work “novelty” is the main thing we struggle to detect. Quality control charts work well for known patterns, but automated identification of new patterns is harder. I’m hoping to get some ideas for tools to help in that area.

For univariate quality control charts, Western Electric rules are used to detect a handful of common patterns. Classic multivariate methods such as Partial Least Squares (PLS) will capture patterns involving more than one variable that would not be detected by univariate methods. Autoencoder is the most comprehensive tool that will cover the widest range of different patterns. It can capture patterns that are multivariate, cyclical, nonlinear, and have interactions. You train the autoencoder using a set of ‘normal’ data’; any patterns present in new data that were not present in the training set will be flagged. We have a template for Spotfire on the TIBCO Community.

Would reducing the dimensions by doing PCA affect the anomalies in a dataset? Would it lead to the disappearance of the anomalies? If so, how could it be prevented?

Doing PCA will capture some percentage of the variance in the original dataset. So the way we use PCA to do anomaly detection is to compute the ‘distance’ from the original point to the point represented in the lower-dimensional space. The larger the distance (i.e. the more that was ‘lost’ when mapping the observation to the lower-dimensional space), the more we regard it as an anomaly.

Chances are you had a similar question to the ones above, but didn’t know how to ask it. In fact, if any of the questions above spark another one in your mind, contact us and we can answer it for you. And watch the full on-demand “Ask Me Anything: Anomaly Detection and Machine Learning” webinar for more on the basics of anomaly detection, common use cases, and some key techniques to keep in mind as you get started.

Let’s block ads! (Why?)

The TIBCO Blog

Spares Pooling: Machine Learning To The Rescue

December 28, 2019 SAP

Spare parts. Every original equipment manufacturer (OEM) has to maintain an inventory of them. Machines break, pieces of equipment fail, and customers need spares to make repairs without delay.

In today’s on-demand global market, spares replenishment must work like magic: prompt, cost-effective, and seamless. Otherwise, prolonged downtime due to equipment failure can have serious direct and indirect impacts, for example, dysfunctional surgery suites in hospitals, flight cancellations, production delays, unfulfilled orders, or poor quality leading to product recalls.

Ensuring the right parts are on hand when needed

Having a sufficient and efficient supply of spare parts is a vexing inventory-management scenario. Forecasting parts-failures at the points of demand and planning for replenishment are essential, as parts are very expensive and costly to keep in stock. Meanwhile, warehouse management teams must routinely juggle multiple, complex scenarios with contradictory insights, forecast failures, and inventory based on complicated predictive analytics.

Demand and consumption forecasting is done at the stockkeeping unit (SKU) level by taking into account the way they handle failures in KPIs. Doing so can minimize the costs of downtime. The key is to apply a risk-ranking system estimates the consequences of a failure to the business, the probability that a failure will occur, and the risk the business is willing to accept.

In terms of a repair strategy, a lot depends on the OEM’s policy on parts replaceability versis repairability. Availability of onsite skills within the maintenance team and time to repair are crucial factors in these decisions. Sometimes, it is more cost-effective to replace the unit and to repair the old unit at one location. But stocking whole units is far more expensive than individual SKUs.

Calculating the efficient safety stock levels at the SKU level must take into account the predictability of demand, delivery lead times, delivery accuracy, obsolescence, transportation costs, and more. The strategy must include reorder points (ROP) and reorder quantities (ROQ), as well as cost-based optimization of order quantities and order cycles at the supplier or product level. Combining automated routine orders with order suggestions for specific important items as well as automatic exception management must also be considered.

Pooling dramatically changes the stocking strategy

Most OEMs have multiple warehouses with millions of dollars’ worth of inventory. If they were to stock at least one piece of each part in all these warehouses, the cost would be prohibitive. This is where parts pooling comes in.

Pooling builds on data that proves aggregation reduces variability and uncertainty. This implies that demand variability will decrease by aggregating spares across locations.

For example, low demand from one warehouse location offsets the demand at another warehouse location. This decline in variability allows a reduction in safety stock and reduces average inventory. When executed well, it is an excellent approach to reduce inventory at individual warehouses without impacting customers.

Shared and pooled warehousing have the potential to transform spares management of small and midsize businesses in competitive markets. This is quite valuable as business networking strategies and collaboration in logistics continue to gain momentum.

Easier said than done

Location-base parts pooling requires adding many highly complicated variables to the current set of variables. In this case, predictive modeling is no longer sufficient. It is time to move to artificial intelligence (AI) and machine learning models that take into account multiple variables and blend them to enable better decision-making in two critical areas:

Warehouse location and product flow: Decisions on whether to have many warehouses close to customers or to have more centralized locations can balance the advantages of aggregated demand and lead times. Proximity to customers and other factors may make maintaining more warehouses more advantageous.

Transportation: Managing parts pooling across locations means accounting for the geographical area along with logistics like availability and reliability of the ground transportation network, flights, or water corridors.

Within months of implementing warehouse pooling optimized through machine learning, most OEMs’ customer KPIs improve while simultaneously minimizing the capital invested in spares.

AI and machine learning models merge cost, operations, and criticality analysis for the relevant scenarios while factoring location criteria at a granular level to automate actions and enable better decisions. Optimization enable OEMs to reduce inventory without impacting customers by using a pooling model.

To reduce risk, OEMs can start by implementing an ML model with a few fixed locations and fewer optimization criteria, like the cost of parts, shipping costs, lead times, and the probability of failures. As the system matures, they can improve this model through a dynamic pooling model capable of handling more factors.

Lenient spares inventory management is a strict no-no as the costs of non-availability can be disastrous. Machine learning-enabled systems are set to transform the way warehouses operate. But before OEMs take the plunge, they must make sure they have the data and talent they need.

See some real-world examples of how machine learning is establishing itself as a mainstream technology tool for companies, enabling them to improve productivity, planning, and ultimately, profits, in the study “Machine Learning In The Real World.”

Let’s block ads! (Why?)

Digitalist Magazine

The Benefits of Doing Machine Learning and Analytics in the Cloud

December 17, 2019 TIBCO Spotfire

Reading Time: 4 mins 30 sec.

Analytics with data science has been one of the last enterprise systems to move to the cloud, but the situation has changed fundamentally in just the last year or two.

Suddenly, there is a proliferation of cloud-based databases and open-source machine learning development frameworks like SageMaker and TensorFlow—all of them now being heavily promoted by the major cloud vendors (Amazon, Microsoft, Google, and more).

The cloud is quickly becoming everyone’s preferred way of doing machine learning and analytics. If you know your way around all the available components, it can be easy to build even the most sophisticated machine learning models for everything from image recognition to fraud detection in the cloud.

What to use—when and how

There’s a ton of rich functionality available in the cloud that you can spin up right now. Over the last few years, there’s been a real shift from heavyweight on-premises installations of data science and predictive analytics to the more lightweight approach that the cloud offers.

The breadth of capabilities that the cloud providers have created combined with the ease of use of a data science platform like TIBCO’s, organizations can spin up environments very quickly without a great deal of IT overhead. This combination of scalability and flexibility is the central value of the cloud when it comes to doing analytics. In fact, with TIBCO^® Data Science, you can create solutions across all of these various cloud environments without needing to learn the nuances of each.

Here’s a helpful chart of the technologies available for artificial intelligence (AI) and machine learning in the cloud:

The Benefits of Doing Machine Learning and Analytics in the Cloud

Proof it’s as easy as it sounds

It really is that easy. The proof is in our customers’ success stories. Below are a couple of case studies that show how easy it can be to build applications based on sophisticated AI and machine learning, using the cloud.

Tipping Point Community fights poverty with data:

As a non-profit organization looking to better understand the drivers behind poverty, Tipping Point started a project to explore correlations between parking citations, late fees, and low-income individuals. Using TIBCO Data Science’s collaborative interface for business users and deploying machine learning models in the cloud to discover insights, Tipping Point found a disproportionate impact on low-income drivers. We’re proud to say these data-driven recommendations that Tipping Point made to the San Francisco Office of Financial Justice led to a change in policy to make the system fairer.

Leidos unlocks big data potential for healthcare analytics:

Leidos partnered with TIBCO Data Science, an enterprise-class cloud platform that leverages Amazon Web Services (AWS), to allow users to create machine learning workflows. By using the cloud, Leidos opened up collaboration across teams and was able to perform quicker analyses. It was able to analyze healthcare data to determine the cause of disease outbreaks like HIV and Zika, consolidate data around emerging healthcare policies, and explore human factors affecting space exploration for NASA.

Across these and many other examples, performing analytics and machine learning in the cloud gives organizations the ability to uncover hidden patterns, anticipate outcomes, and react quickly to real-world events. Data science teams can spin up new systems in the cloud in a matter of hours, performing advanced analytics in a low-code, visual data science environment like TIBCO’s to find the answers to their toughest problems, fast.

Connect teams & scale algorithms with cloud-enabled tech

The TIBCO Data Science platform provides an interactive interface for teams to collaborate on projects in the cloud. Teams scattered around the globe, across different departments, in different roles, can connect through the web-based interface to solve difficult data science problems together. Furthermore, the algorithms TIBCO provides through a simple drag and drop interface are not only easy to use, without requiring a lot of code, but they’re also readily scalable and immensely powerful. But if coding is your thing, you can also use the embedded Jupyter Notebooks that are built into the platform.

Watch this webinar for more on available cloud technologies and the many benefits of doing machine learning and analytics in the cloud.

Get started right now. Go to the AWS marketplace and sign up for a preview of the TIBCO Data Science platform, connect to open sources of data, and start building machine learning models today.

Let’s block ads! (Why?)

The TIBCO Blog

AI Weekly: NeurIPS proves machine learning at scale is hard

December 13, 2019 Big Data

The world’s largest AI research conference is underway in Vancouver, Canada. Researchers are presenting more than 1,400 papers at the Neural Information Processing Systems (NeurIPS) conference, ranging from work that organizers believe has had the greatest impact over the past decade to Yoshua Bengio’s continued march toward consciousness for deep learning.

But even as the conference showed theoretical research and neuroscience-related papers on the rise alongside categories like algorithms and deep learning, the mushrooming of the event itself — and the associated growing pains — was a constant theme, and it speaks to the growth of the AI field in general.

Organizers said that at the start of the conference Sunday, they expected about 400 people to show up for registration. Instead, 4,200 queued up in the conference center registration line. All told, NeurIPS 2019 welcomed 13,000 attendees, up 40% from the prior year. In all, the conference has quadrupled in size in the past five years.

Organizers started to use a lottery system this year, after 9,000 tickets to 2018 NeurIPS sold out in 12 minutes.

Researchers submitted a record 6,600 papers for consideration.

With industry, governments, and academia investing billions, machine learning is red hot right now. The AI Index 2019 report released Wednesday found that private investment, research, and the number of PhD candidates in AI are all growing at healthy rates.

But that growth can come at a cost. As intellectually driven as NeurIPS can be, crowds can get in the way of the intellectual exchanges and scientific critique the event is intended to spawn.

Organizers invited all conference attendees to share feedback in a town hall Thursday, and one of the scientists’ main concerns was how to deal with the event’s rapid growth.

Poster sessions are the most valuable part of the conference, according to many who spoke at the town hall, since they allow you to speak directly with research authors and gain exposure to a broad range of work. At one point, the poster sessions was filled to capacity, making those dialogs difficult.

This year’s organizers are considering a lot of potential ways forward: NeurIPS regional meetups were held in more 35 countries this year, and organizers considered splitting the global conference into regional events.

“The idea is instead of having one big meeting, we could have a bunch of satellite meetings,” Sejnowski said. “I think the idea is we’ll have a bigger audience and greater participation if we had more than one location.”

Should the conference break into regional events, Seejnowski said organizers could randomly dispatch speakers to regional locations and livestream their talks for consumption around the world.

NeurIPS livestreamed major presentations this year and gave researchers ways to critique and question posters online, but a regional approach naturally limits in-person interactions and could also dilute the kind of valuable interactions the global event is intended to produce.

NeurIPS 2020 is back in Vancouver and 2021 will move to Sydney, in a larger space than the Vancouver Convention Center. Organizers may aim for larger venues in the future, like convention centers in places like New Orleans and San Diego, which have capacity of 30,000. Doing so could allow doing away with the lottery system.

Among other changes under consideration for next year: a policy to exclude certain vendors from the industry expo. No rules currently exist to exclude specific vendors. The issue came up following a question from the audience about why the National Security Agency (NSA), a U.S. intelligence organization, had a booth at a conference held in Canada. Next year, researchers may also be asked to share the carbon footprint of training their AI model.

Toward a more open and welcoming NeurIPS

Like last year, attendees from parts of Asia and Africa encountered issues with Canadian immigration officials that kept them from reaching Vancouver. Diversity and inclusion chair Katherine Heller said that as of Monday, 158 initially denied visas were approved, 14 were in progress, and five were denied. Despite the change, some poster sessions displayed signs that said the presenter received a visa too late to attend.

Since the conference will take place in Vancouver again next year, organizers hope to improve communication with the Canadian government in 2020.

But organizers seem to be trying hard to make NeurIPS a more accepting and welcoming place. A child care program was full to capacity for the kids of about 75 researchers. That program is expected to expand next year.

UC Berkeley professor Celeste Kidd gave the opening keynote address, and touched on how tech can drive false facts and misunderstandings of the #metoo movement.

{Dis}ability in AI had its first meeting, and workshops highlighted the work of diverse machine learning practitioners as well as AI that can positively impact the lives of people of African or Latin descent, women, and followers of Islam and Judaism.

At the Queer in AI workshop, people got to choose whether they wanted to be photographed or not, and you could hear presentations about research that seeks to remove gender identity from AI systems.

Some workshops even opened with acknowledgment that the convention took place on the land of the native Coast Salish people.

Machine learning at scale

What you hear about at NeurIPS are the advances in stochastic gradient descent, covariants, and all manner of technical discoveries or steps forward, but there are also workshops on disaster response, privacy, machine learning for the developing world, climate change, and using computer vision to battle cancer.

It’s an insane rush of theoretical, complex approaches to improve the performance of machine intelligence, but also a drive to solve real-world problems and contribute to society.

The other challenge in front of the community, it seems, is growth management. Perhaps regional events help, but interest in a unified international event does not appear to be subsiding.

AI is the top focus of computer science school PhD programs today, and other conferences like ICLR or CVPR are also experiencing growth.

In other words, growth in the AI field will likely continue to drive the growth of NeurIPS. The organizers of the biggest machine learning research conference of the year have to deal with the challenges that growth brings, but that’s a nice problem to have. While they’re making progress toward solving some of those challenges, it seems clear that the importance of physical access to the conference will not subside anytime soon and will likely continue to grow in importance.

For AI coverage, send news tips to Khari Johnson and Kyle Wiggers and AI editor Seth Colaner — and be sure to subscribe to the AI Weekly newsletter and bookmark our AI Channel.

Thanks for reading,

Khari Johnson

Senior AI Staff Writer

Let’s block ads! (Why?)

Big Data – VentureBeat

Rensselaer focuses IBM’s AiMOS supercomputer on machine learning

December 6, 2019 Big Data

Sophisticated machine learning applications require not only enormous amounts of training data, but powerful computer hardware on which to train. An analysis conducted by San Francisco research firm OpenAI found that since 2012, the amount of compute used in the largest training runs has been increasing exponentially with a 3.4-month doubling time, and that it’s grown by more than 300,000 times over that same time period.

The trend spurred the development of supercomputers like the U.S. Department of Energy’s Sierra and Summit, which leverage dedicated accelerator chips to speed up AI computation. Now, IBM’s Hardware Center, in collaboration with New York State, SUNY Polytechnic Institute, and other members of IBM’s AI Hardware Center, has delivered a new machine for the Department of Computer Science at Rensselaer Polytechnic Institute (RPI) that’s optimized for state-of-the-art machine learning workloads.

It’s dubbed Artificial Intelligence Multiprocessing Optimized System, or AiMOS (in honor of Rensselaer cofounder Amos Eaton), and it will principally tackle projects in biology, chemistry, the humanities, and related domains underway at the new IBM Research AI Hardware Center on the SUNY campus in Albany. It’s built using the same IBM Power Systems technology as the aforementioned Sierra and Summit, including IBM Power9 processors and Nvidia Tesla V100 graphics cards. And at eight petaflops on the High-Performance Linpack benchmark, a yardstick of supercomputing performance, AiMOS is one of the most powerful computers to debut on the Top500 ranking on supercomputers.

In fact, IBM claims AiMOS is currently the most powerful supercomputer housed at a private university, the 24th-most powerful supercomputer in the world, and the third-most energy efficient.

“As the home of one of the top high-performance computing systems in the U.S. and in the world, Rensselaer is excited to accelerate our ongoing research in AI, deep learning, and in fields across a broad intellectual front,” said Rensselaer president Shirley Ann Jackson. “The creation of new paradigms requires forward-thinking collaborators, and we look forward to working with IBM and the state of New York to address global challenges in ways that were previously impossible.”

AiMOS will provide modeling, simulation, and computation to advance the development of computing chips and systems optimized for AI algorithms, according to IBM executive vice president John E. Kelly III, with the goal of helping to make AI systems 1,000 times more efficient within the next decade. “[AI] … will help us solve some of our most pressing problems, from healthcare to security to climate change. In order to realize AI’s full potential, special purpose computing hardware is emerging as the next big opportunity.”

Of course, as speedy as AiMOS may be, it’s behind Intel’s Frontera, which can achieve peak performance of 38.7 quadrillion floating point operations per second. Frontera is the fastest computer designed for academic workloads like modeling and simulation, big data, and machine learning, a distinction it earned for the second time in November by nabbing fifth place in the Top500 list.

And AiMOS will certainly be outgunned by AMD’s Frontier, a machine with more than 1.5 exaflops of theoretical performance that’s anticipated to be delivered to Oak Ridge National Laboratory in 2021. Intel for its part plans to deploy the “exaflop-class” Aurora cluster at the Argonne National Laboratory within the next two years as part of the Energy Department’s Exascale Computing Project, a grant program that seeks to accelerate exascale computing research in the U.S.

Currently, the U.S. hosts five of the 10 fastest computers in the world, with China’s best — the TaihuLight at the National Supercomputing Center in Wuxi (built on Sunway’s SW26010 processor architecture) and the Tianhe-2A in Guangzhou — ranking third and fourth, respectively, at roughly 125 peak petaflops and 100 peak petaflops. Cray’s Piz Daint sits in sixth, ahead of Trinity at Los Alamos National Laboratory, Fujitsu’s AI Bridge Clouding Infrastructure in Japan, and Lenovo’s SuperMUC-NG in Germany.

The race between China and the U.S. is fierce. In Top500 rankings, China two years ago surpassed the United States in total number of ranked supercomputers for the first time, with 202 to 143. That trend accelerated the following year; according to the Top500 fall 2018 report, the number of ranked U.S. supercomputers fell to 108 as China’s total climbed to 229.

China and the U.S. are followed in the largest number of ranked supercomputers by Japan, which has over 30 systems; the U.K., with over 20; France with nearly 20; Germany with over 15; and Ireland with just over 10.

Let’s block ads! (Why?)

Big Data – VentureBeat

A super-fast machine learning model for finding user search intent

November 30, 2019 Big Data

In April 2019, Benjamin Burkholder (who is awesome, by the way) published a Medium article showing off a script he wrote that uses SERP result features to infer a user’s search intent. The script uses the SerpAPI.com API for its data and labels search queries in the following way:

Informational — The person is looking for more information on a topic. This is indicated by whether an answer box or PAA (people also ask) boxes are present.
Navigational — The person is searching for a specific website. This is indicated by whether a knowledge graph is present or if site links are present.
Transactional — The person is aiming to purchase something. This is indicated by whether shopping ads are present.
Commercial Investigation — The person is aiming to make a purchase soon but is still researching. This is indicated by whether paid ads are present, an answer box is present, PAAs are present, or if there are ads present at the bottom of the SERP.

This is one of the coolest ways to estimate search intent, because it uses Google’s understanding of search intent (as expressed by the SERP features shown for that search).

The one problem with Burkholder’s approach is its reliance on the Serp API. If you have a large set of search queries you want to find intent for, you need to pass each query phrase through the API, which then actually does the search and returns the SERP feature results, which Burkholder’s script can then classify. So on a large set of search queries, this is time consuming and prohibitively expensive.

SerpAPI charges ~$ 0.01 per keyword, so analyzing 5,000 keywords will cost you $ 50. Running these results through Burkholder’s labeler script also takes 3 to 5 hours to get through these 5,000 keywords.

So I got to thinking: What if I adapted Burkholder’s approach so that, rather than use it to classify intent directly, I could use it to train a machine learning model that I would then use to classify intent? In other words, I’d incur one-time costs to produce my Burkholder-labeled training set, and, assuming it was accurate enough, I could then use that training set for all further classification, cost free.

With an accurate training set, anyone could label huge numbers of keywords super quickly, without spending a dime.

Finding a model

Hamlet Batista has written a few stellar posts about how to leverage Natural Language models like BERT for labeling intent.

In his posts, he uses an existing intent labeling model that returns categories from Kaggle’s Question Answering Dataset. While these labels can be useful, they are not really “intent categories” in line with what we typically think of for intent taxonomy categories and instead have labels such as Description, Entity, Human, Numeric, and Location.

He achieved excellent results by training a BERT encoder, getting near 90% accuracy in predicting labels for new/unlabeled search keywords.

The big question for me was, could I leverage the same tech (Uber’s Ludwig BERT encoder) to create an accurate model using the search intent labels I’d get from Burkholder’s code?

It turns out the answer is yes!

How to do it

Here’s how the process works:

1. Gather your list of keywords. If you’re planning on training your own model, I recommend doing so within a specific category/niche. Training on clothing-related keywords and then using that model to label finance related keywords will likely be significantly less accurate than training on clothing related keywords and then using that model to label other unlabeled clothing related keywords. That said, I did try using a model labeled on one category/niche to label another, and the results still seemed quite good to me.

2. Run Burkholder’s script over your list of keywords from Step 1. This will require signing up for SerpAPI.com and buying credits. I recommend getting labels for at least 10,000 search queries with this script to use for training. The more training data, the more accurate your model will likely be.

3. Use the labeled data from the previous step as your training data for the BERT model. Batista’s code to do this is very straightforward, and this article will guide you through the process. I was able to get about ~72% accuracy using about 10,000 labels of training data.

4. Use your model from Step 3 to label unlabeled search data, and then take a look at your results!

The results

I ran through this process using a huge list (13,000 keywords) of clothing/fashion-related search terms from SEMrush as my training data. My resulting model gets just about 80% accuracy.

It seems likely that training the model with more data will continue to improve its accuracy up to a point. If any of you attempt it and improve on 80% accuracy, I would love to hear about it. I think with 20,000+ labeled searches, we could see up to maybe 85-90% accuracy.

This means when you ask this model to predict the intent of unlabeled search queries, 8 times out of 10 it will give you the same label as what would have been returned by Burkholder’s Serp API rules-based classifier. It can also do this for free, in large volumes and incredibly fast.

So something that would have taken a few thousand dollars and days of scraping can now be done for free in just minutes.

In my case I used keywords from a related domain (makeup) instead of clothing keywords, and overall I think it did a pretty good job. Labeling 5,000 search queries took under two minutes with the BERT model. Here’s what my results looked like:

The implications

For SEO tools to be useful, they need to be scalable. Keyword research, content strategy, PPC strategy, and SEO strategy usually rely on being able to do analysis across entire niches/themes/topics/websites.

In many industries, the keyword longtails can extend into the millions. So a faster, more affordable approach to Burkholder’s solution can make a lot of difference.

I forsee AI and machine learning tools being used more and more in our industry, enabling SEOs, paid search specialists, and content marketers to gain superpowers that haven’t been possible before these new AI breakthroughs.

Happy analyzing!

Kristin Tynski is a founder and the SVP of Creative at Fractl, a boutique growth agency based in Delray Beach, FL.

Let’s block ads! (Why?)

Big Data – VentureBeat

AI Weekly: Multimodal learning is in right now — here’s why that’s a good thing

November 23, 2019 Big Data

Data sets are fundamental building blocks of AI systems, and this paradigm isn’t likely to ever change. Without a corpus on which to draw, as human beings employ daily, models can’t learn the relationships that inform their predictions.

But why stop at a single corpus? An intriguing report by ABI Research anticipates that while the total installed base of AI devices will grow from 2.69 billion in 2019 to 4.47 billion in 2024, comparatively few will be interoperable in the short term. Rather than combine the gigabytes to petabytes of data flowing through them into a single AI model or framework, they’ll work independently and heterogeneously to make sense of the data they’re fed.

That’s unfortunate, argues ABI, because of the insights that might be gleaned if they played nicely together. That’s why as an alternative to this unimodality, the research firm proposes multimodal learning, which consolidates data from various sensors and inputs into a single system.

Multimodal learning can carry complementary information or trends, which often only become evident when they’re all included in the learning process. Plus, learning-based methods that leverage signals from different modalities can generate more robust inference than would be possible in a unimodal system.

Consider images and text captions. If different words are paired with similar images, these words are likely used to describe the same things or objects. Conversely, if some words appear next to different images, this implies these images represent the same object. Given this, it should be possible for an AI model to predict image objects from text descriptions, and indeed, a body of academic literature has proven this to be the case.

Despite the many advantages of multimodal approaches to machine learning, ABI’s report notes that most platform companies — including IBM, Microsoft, Amazon, and Google — continue to focus predominantly on unimodal systems. That’s partly because it’s challenging to mitigate the noise and conflicts in modalities, and to reconcile the differences in quantitative influence that modalities have over predictions.

Fortunately, there’s hope yet for wide multimodal adoption. ABI Research anticipates the total number of devices shipped will grow from 3.94 million in 2017 to 514.12 million in 2023, spurred by adoption in the robotics, consumer, health care, and media and entertainment segments. Companies like Waymo are leveraging multimodal approaches to build hyper-aware self-driving vehicles, while teams like that led by Intel Labs principal engineer Omesh Tickoo are investigating techniques for sensor data collation in real-world environments.

“In a noisy scenario, you may not be able to get a lot of information out of your audio sensors, but if the lighting is good, maybe a camera can give you a little better information,” Tickoo explained to VentureBeat in a phone interview. “What we did is, using techniques to figure out context such as the time of day, we built a system that tells you when a sensor’s data is not of the highest quality. Given that confidence value, it weighs different sensors against each at different intervals and chooses the right mix to give us the answer we’re looking for.”

Multimodal learning won’t supplant unimodal learning, necessarily — unimodal learning is highly effective in applications like image recognition and natural language processing. But as electronics become cheaper and compute more scalable, it’ll likely only rise in prominence.

“Classification, decision-making, and HMI systems are going to play a significant role in driving adoption of multimodal learning, providing a catalyst to refine and standardize some of the technical approaches,” said ABI Research chief research officer Stuart Carlaw in a statement. “There is impressive momentum driving multimodal applications into devices.”

For AI coverage, send news tips to Khari Johnson and Kyle Wiggers — and be sure to subscribe to the AI Weekly newsletter and bookmark our AI Channel.

Thanks for reading,

Kyle Wiggers

AI Staff Writer

P.S. Please enjoy this video about Bill Gates discussing AI at Bloomberg’s New Economy Forum in Beijing, among other topics like climate change and nuclear power.

Let’s block ads! (Why?)

Big Data – VentureBeat

How to operationalize data science and machine learning without IT hurdles

November 13, 2019 BI News and Info

One of the biggest contributors to the Model Impact Epidemic

You may have heard about the Model Impact Epidemic – essentially a series of obstacles that are preventing organizations from properly operationalizing machine learning models. According to a recent Gartner report, 2 of the 3 top challenges that prevent data science projects from making it into production are IT-related.

Sometimes, IT departments get overloaded with requests for maintaining and operating different applications and platforms. ML Operations is a brand-new field, which requires some additional skills and a full understanding of the life cycle of a data science project.

If you develop DS projects, but they never get into production or the pace at which your processes are deployed is not quick enough, RapidMiner Managed Server is for you.

What is the RapidMiner Server?

RapidMiner Server is the orchestration and automation component of the RapidMiner platform that helps overcome many of the organizational challenges that prevent models from being fully deployed so they can deliver real business impact. Data Science is no longer an exploration task that ends with a report written by a data scientist. Today, impact is delivered only once models are deployed and operationalized. And that’s when RapidMiner Server shines: allowing enterprise customers to quickly deploy models in order to create value.

To help you seamlessly integrate your data science workflow into your production environments, as a part of RapidMiner 9.4, we introduced RapidMiner Managed Server, a new service that makes it easy for both small groups and larger companies to benefit from RapidMiner Server, without the need to add new hardware or wait for IT resources.

With RapidMiner Managed Server, we take care of all IT-related tasks (installation, maintenance, security, etc.), removing many of the roadblocks that users find on their path to applying models in production.

Managed Server: We do it for you

RapidMiner Managed Server is a services offering that harnesses our experienced team of ML Operations experts to install, configure, and maintain a RapidMiner environment for you. The environment can include one or multiple Servers (Development, Test and Production). It can grow horizontally whenever needed by having high availability servers or by adding Job Agents for additional execution power. And it can also include any number of Real-time Scoring Agents for Web Services deployments.

We host the environment in the RapidMiner AI Cloud and it’s tailored to suit your needs. In fact, the first thing we do is to analyze your needs in order to get the right size. We make sure your data scientists will find room for collaboration, remote execution of their projects, and even more importantly, for the deployment that finally provides value for the company. Next, we set up the infrastructure with a focus on network configuration and security. Finally, the RapidMiner platform is deployed and configured.

And then you can start working with it. No need to maintain it, it just works. While you do the data science, we do things like:

Backup everything and restore whenever needed
Manage any incidents that may come up
Elastically grow or shrink the environment according to your needs
Upgrade to new versions as they become available
Monitor the system
Secure the system to the highest standards

All this happens transparently while you use the platform and focus on solving your business problems.

RapidMiner Server as a Service

Internally, your users will be accessing RapidMiner Server in the cloud, “as a service”. However, it can be integrated with your network, so you can access your own data sources. And it can be configured in a completely customized way just for you. It’s a dedicated environment you won’t be sharing with anyone else.

Architecture of the environments

The RapidMiner environments are in the cloud, with the flexibility to make changes whenever they are needed. The whole system is “dockerized” and uses modern, state-of-the-art technology that tightens security, ensures availability, and minimizes risks. Our team of security experts make sure any vulnerability is removed and your data and processes are safe.

The system is also fully monitored, and we periodically share this information with you, so you’ll know if capacity is running low, what time is best to schedule a process, or which processes are using a lot of memory.

RapidMiner Server and Real-time Scoring

RapidMiner Server and Real-time Scoring are included in with RapidMiner Managed Server. They make a good couple, as RapidMiner Server provides user control, collaboration, and scheduling, while Real-time Scoring provides easy deployment of projects as Web Services and has extremely low latencies during execution. Both can be part of your environment.

Conclusions

With RapidMiner Managed Server, we are offering the expertise and knowledge of our group of expert ML Operations administrators. They’ll take care of installation, configuration, maintenance, availability, re-sizing, security, upgrades—all of the nitty gritty—and you just use the platform and get the value: collaboration during development, remote execution, scheduling, deployment, etc.

This is one of the many ways RapidMiner is working to help our customers move their path to AI beyond the ‘science project phase’ and into the world of operational models. In the world of Enterprise AI, models don’t have an impact until they’re fully deployed. Click here for more information on other new offerings and enhancements we’ve provided to help with this initiative.

Let’s block ads! (Why?)

RapidMiner

Amazon researchers reduce data required for AI transfer learning

October 28, 2019 Big Data

Cross-lingual learning is an AI technique involving training a natural language processing model in one language and retraining it in another. It’s been demonstrated that retrained models can outperform those trained from scratch in the second language, which is likely why researchers at Amazon’s Alexa division are investing outsize time investigating them.

In a paper scheduled to be presented at this year’s Conference on Empirical Methods in Natural Language Processing, two scientists at the Alexa AI natural understanding group — Quynh Do and Judith Gaspers — and colleagues propose a data selection technique that halves the amount of required training data. They claim that it surprisingly improves rather than compromises the model’s overall performance in the target language.

“Sometimes the data in the source language is so abundant that using all of it to train a transfer model would be impractically time consuming,” wrote Do and Gaspers in a blog post. “Moreover, linguistic differences between source and target languages mean that pruning the training data in the source language, so that its statistical patterns better match those of the target language, can actually improve the performance of the transferred model.”

In the course of experiments, Do, Gaspers, and team employed two methods to cut the source-language data set in half: the aforementioned data selection technique and random sampling. They pretrained separate models on the two halved data sets and on the full data set, after which they fine-tuned the models on a small data set in the target language

Do and Gaspers note that all of the models were trained simultaneously to recognize intents (requested actions) and fill slots (variables on which the intent acts), and that they took as inputs multilingual embeddings (a word or sequences of words from different languages mapped to a single point in a multidimensional space) to bolster model accuracy. The team combined the multilingual embedding of each input word with a character-level embedding that encoded information about words’ prefixes, suffixes, and roots, and they tapped language models trained on large text corpora to select the source-language data that’d be fed to the transfer model.

Within the system the researchers engineered, a bilingual dictionary translated each utterance in the source data set into a string of words in the target language. Four language models were applied to the resulting strings while a trigram model handled character embeddings. And for each utterance in the the sum of probabilities computed by the four language models, only those which yielded the highest normalized score were selected. .

To evaluate their approach, the team first transferred a model from English to German with different amounts of training data in the target language (10,000 and 20,000 utterances, respectively, versus millions of utterances in the full source-language data set). Then, they trained the transfer model on three different languages — English, German, and Spanish — before transferring it to French (with 10,000 and 20,000 utterances in the target language). They claim that the transfer models outperformed a baseline model trained only on data in the target language, and that relative to the model trained on the target language alone, the model trained using the novel data selection technique showed improvements of 3% to 5% on the slot-filling task and about 1% to 2% on intent classification.

Let’s block ads! (Why?)

Big Data – VentureBeat

Tag Archives: Learning

AI Weekly: Machine learning could lead cybersecurity into uncharted territory

7 Questions on How to Use Machine Learning for Anomaly Detection

What’s the difference between outliers and anomalies?

Are there any applications for anomaly detection in pharmaceuticals?

Are GANs also used for anomaly detection? If so, could you please provide an industry use case.

Does correlation data affect anomaly detection? In what ways and how could we diminish the effects? Is it preferable to clean and remove correlations before starting anomaly detection?

What is a suggested algorithm that would be appropriate for anomaly detection related to identifying unusual activities in network activities or data?

In my current work “novelty” is the main thing we struggle to detect. Quality control charts work well for known patterns, but automated identification of new patterns is harder. I’m hoping to get some ideas for tools to help in that area.

Would reducing the dimensions by doing PCA affect the anomalies in a dataset? Would it lead to the disappearance of the anomalies? If so, how could it be prevented?

Spares Pooling: Machine Learning To The Rescue

Ensuring the right parts are on hand when needed

Pooling dramatically changes the stocking strategy

Easier said than done

AI Weekly: NeurIPS proves machine learning at scale is hard

Toward a more open and welcoming NeurIPS

Machine learning at scale

Rensselaer focuses IBM’s AiMOS supercomputer on machine learning

A super-fast machine learning model for finding user search intent

Finding a model

How to do it

The results

The implications

AI Weekly: Multimodal learning is in right now — here’s why that’s a good thing

How to operationalize data science and machine learning without IT hurdles

One of the biggest contributors to the Model Impact Epidemic

What is the RapidMiner Server?

Managed Server: We do it for you

RapidMiner Server as a Service

Architecture of the environments

RapidMiner Server and Real-time Scoring

Conclusions

Amazon researchers reduce data required for AI transfer learning

Recent Posts

Categories

Archives