Category Archives: Machine Learning

In classical computers, bits are stored as either a 0 or a 1 in binary notation. Quantum computers use quantum bits or qubits which can be both 0 and 1, this is called superimposition. Last year Google and NASA claimed to have achieved quantum supremacy, raising some controversies though. Quantum supremacy means that a quantum computer can perform a single calculation that no conventional computer, even the biggest supercomputer can perform in a reasonable amount of time. Indeed, according to Google, the Sycamore is a computer with a 54-qubit processor, which is can perform fast computations.

Machines like Sycamore can speed up simulation of quantum mechanical systems, drug design, the creation of new materials through molecular and atomic maps, the Deutsch Oracle problem and machine learning.

When data points are projected in high dimensions during machine learning tasks, it is hard for classical computers to deal with such large computations (no matter the TensorFlow optimizations and so on). Even if the classical computer can handle it, an extensive amount of computational time is necessary.

In other words, the current computers we use can be sometime slow while doing certain machine learning application compared to quantum systems.

Indeed, superposition and entanglement can come in hand to train properly support vector machine or neural networks to behave similarly to a quantum system.

How we do this in practice can be summarized as

In practice, quantum computers can be used and trained like neural networks, or better neural networks comprises some aspects of quantum physics. More specifically, in photonic hardware, a trained circuit of quantum computer can be used to classify the content of images, by encoding the image into the physical state of the device and taking measurements. If it sounds weird, it is because this topic is weird and difficult to digest. Moreover, the story is bigger than just using quantum computers to solve machine learning problems. Quantum circuits are differentiable, and a quantum computer itself can compute the change (rewrite) in control parameters needed to become better at a given task, pushing further the concept of learning.

See original here:
Is Quantum Machine Learning the next thing? | by Alessandro Crimi | ILLUMINATION-Curated | Oct, 2020 - Medium

Suggestions were formulated to reduce bias and error related to the use of machine learning (ML) approaches in cardiology research, and published in the Journal of American College of Cardiologists: Cardiovascular Imaging.

The use of ML approaches for cardiovascular research has recently increased, as the technology offers approaches to automatically discover relevant patterns among datasets. This review authored by members of the American College of Cardiology Healthcare Innovation Council, points to the fact that many studies using ML approaches may have uncertain real-world data sources, inconsistent outcomes, possible measurement inaccuracies, or lack of validation and reproducibility.

The authors provide here a framework to guide cardiovascular research in the form of a checklist.

When considering employing a ML approach for their research work, investigators should initially determine whether it would be applicable for the specific study aim. An important caveat of ML is that it requires large sample sizes. Therefore, if collecting and labeling fewer than hundreds of samples per class is not feasible, overfitting is likely be a relevant concern. When sufficient samples are available, ML approaches are best suited for unstructured data, exploratory study objectives, or for feature selection purposes.

Next, data should be standardized, if necessary. During this process, redundant features are normalized, duplicates are removed, outliers removed or corrected for, and missing data removed or imputed. As a general rule, the ratio of observations to measurements should be 5. In cases in which this ratio is too large, dimension reduction may be considered.

Many ML approaches are available to researchers, and the choice of which model to implement is critical. Some models are preferable for high-dimensional data (regression or instance-based learning) or imaging data (convolutional neural networks). The authors recommend selecting the simplest algorithm that is appropriate for ones dataset.

Several methods are available to assess and evaluate models. Model assessment should always be performed through random division of the data into training, testing, and validation sets. Cross-validation and bootstrapping methods are best suited for big data, and jack-knifing methods for smaller datasets. Model evaluation should include appropriate plots (Bland-Altman). In addition, inter-observational variability should be reported, and misclassification risk be made clear.

To maintain a level of reproducibility across studies, the authors encourage researchers to release the code and data used, when possible. All chosen variables and parameters, as well as specific versions of software and libraries should be clearly indicated.

The authors acknowledge that these methods are complex, and while they have the opportunity to advance the field of cardiology, especially personalized medicine, many concerns remain when translating these findings into clinical practice. This checklist should assist researchers in reducing bias or error when designing and carrying out future studies.

Reference

Sengupta P P, Shrestha S, Berthon B, et al. Proposed Requirements for Cardiovascular Imaging-Related Machine Learning Evaluation (PRIME): A Checklist. JACC Cardiovasc Imaging. 2020;13(9):2017-2035.

Read more from the original source:
Requirements for the Use of Machine Learning in Cardiology Research - The Cardiology Advisor

Samsung is rolling out a new course offering an introduction to machine learning and artificial intelligence in the UAE.

The course, which is part of its global Future Academy initiative, will target UAE residents between the ages of 18 and 35 with a background in science, technology, engineering and mathematics and who are interested in pursuing a career that would benefit from knowledge of AI, the South Korean firm said.

The five-week programme will be held online and cover subjects such as statistics, algorithms and programming.

The launch of the Future Academy in the UAE reaffirms our commitment to drive personal and professional development and ensure this transcends across all areas in which we operate, said Jerric Wong, head of corporate marketing at Samsung Gulf Electronics.

In July, Samsung announced a similar partnership with Misk Academy to launch AI courses in Saudi Arabia.

The UAE, a hub for start-ups and venture capital in the the Arab world, is projected to benefit the most in the region from AI adoption. The technology is expected to contribute up to 14 per cent to the countrys gross domestic product equivalent to Dh352.5 billion by 2030, according to a report by consultancy PwC.

In Saudi Arabia, AI is forecast to add 12.4 per cent to GDP.

Held under the theme be ready for tomorrow by learning about it today, the course will be delivered through a blended learning and self-paced format. Participants can access presentations and pre-recorded videos detailing their course materials.

Through the Future Academys specialised curriculum, participants will learn about the tools and applications that feature prominently in AI and machine learning-related workplaces, Samsung said.

The programme promises to be beneficial, providing the perfect platform for determined beginners and learners to build their knowledge in machine learning and establishing a strong understanding of the fundamentals of AI, it added.

Applicants can apply here by October 29.

Updated: October 6, 2020 07:57 PM

The rest is here:
Samsung launches online programme to train UAE youth in AI and machine learning - The National

Does the life of an astronomer or planetary scientists seem exciting?

Sitting in an observatory, sipping warm cocoa, with high-tech tools at your disposal as you work diligently, surfing along on the wavefront of human knowledge, surrounded by fine, bright people. Then one dayEureka!all your hard work and the work of your colleagues pays off, and you deliver to humanity a critical piece of knowledge. A chunk of knowledge that settles a scientific debate, or that ties a nice bow on a burgeoning theory, bringing it all together. ConferencestenureNobel Prize?

Well, maybe in your first year of university you might imagine something like that. But science is work. And as we all know, not every minute of ones working life is super-exciting and gratifying.

Sometimes it can be dull and repetitious.

Its probably not anyones dream, when they begin their scientific education, to sit in front of a computer poring over photos of the surface of Mars, counting the craters. But someone has to do it. How else would we all know how many craters there are?

Mars is the subject of intense scientific scrutiny. Telescopes, rovers, and orbiters are all working to unlock the planets secrets. There are a thousand questions concerning Mars, and one part of understanding the complex planet is understanding the frequency of meteorite strikes on its surface.

NASAs Mars Reconnaissance Orbiter (MRO) has been orbiting Mars for 14.5 years now. Along with the rest of its payload, the MRO carries cameras. One of them is called the Context (CTX) Camera. As its name says, it provides context for the other cameras and instruments.

MROs powerhouse camera is called HiRISE (High-Resolution Imaging Science Experiment). While the CTX camera takes wider view images, HiRISE zooms in to take precision images of details on the surface. The pair make a potent team, and HiRISE has treated us to more gorgeous and intriguing pictures of Mars than any other instrument.

But the cameras are kind of dumb in a scientific sense. It takes a human being to go over the images. As a NASA press release tells us, it can take 40 minutes for one researcher to go over a CTX image, hunting for small craters. Over the lifetime of the MRO so far, researchers have found over 1,000 craters this way. Theyre not just looking for craters, theyre interested in any changes on the surface: dust devils, shifting dunes, landslides, and the like.

AI researchers at NASAs Jet Propulsion Laboratory in Southern California have been trying to do something about all the time it takes to find things of interest in all of these images. Theyre developing a machine learning tool to handle some of that workload. On August 26th, 2020, the tool had its first success.

On some date between March 2010 and May 2012, a meteor slammed into Mars thin atmosphere. It broke into several pieces before it struck the surface, creating what looks like nothing more than a black speck in CTX camera images of the area. The new AI tool, called an automated fresh impact crater classifier, found it. Once it did, NASA used HiRISE to confirm it.

That was the classifiers first find, and in the future, NASA expects AI tools to do more of this kind of work, freeing human minds up for more demanding thinking. The crater classifier is part of a broader JPL effort named COSMIC (Capturing Onboard Summarization to Monitor Image Change). The goal is to develop these technologies not only for MRO, but for future orbiters. Not only at Mars, but wherever else orbiters find themselves.

Machine learning tools like the crater classifier have to be trained. For its training, it was fed 6,830 CTX camera images. Among those images were ones containing confirmed craters, and others that contained no craters. That taught the tool what to look for and what not to look for.

Once it was trained, JPL took the systems training wheels off and let it loose on over 110,000 images of the Martian surface. JPL has its own supercomputer, a cluster containing dozens of high-performance machines that can work together. The result? The AI running on that powerful machine took only five seconds to complete a task that takes a human about 40 minutes. But it wasnt easy to do.

It wouldnt be possible to process over 112,000 images in a reasonable amount of time without distributing the work across many computers, said JPL computer scientist Gary Doran, in a press release. The strategy is to split the problem into smaller pieces that can be solved in parallel.

But while the system is powerful, and represents a huge savings of human time, it cant operate without human oversight.

AI cant do the kind of skilled analysis a scientist can, said JPL computer scientist Kiri Wagstaff. But tools like this new algorithm can be their assistants. This paves the way for an exciting symbiosis of human and AI investigators working together to accelerate scientific discovery.

Once the crater finder scores a hit in a CTX camera image, its up to HiRISE to confirm it. That happened on August 26th, 2020. After the crater finder flagged a dark smudge in a CTX camera image of a region named Noctis Fossae, the power of the HiRISE took scientists in for a closer look. That confirmed the presence of not one crater, but a cluster of several resulting from the objects that struck Mars between March 2010 and May 2012.

With that initial success behind them, the team developing the AI has submitted more than 20 other CTX images to HiRISE for verification.

This type of software system cant run on an orbiter, yet. Only an Earth-bound supercomputer can perform this complex task. All of the data from CTX and HiRISE is sent back to Earth, where researchers pore over it, looking for images of interest. But the AI researchers developing this system hope that will change in the future.

The hope is that in the future, AI could prioritize orbital imagery that scientists are more likely to be interested in, said Michael Munje, a Georgia Tech graduate student who worked on the classifier as an intern at JPL.

Theres another important aspect to this development. It shows how older, still-operational spacecraft can be sort of re-energized with modern technological power, and how scientists can wring even more results from them.

Ingrid Daubar is one of the scientists working on the system. She thinks that this new tool will help find more craters that are eluding human eyes. And if it can, itll help build our knowledge of the frequency, shape, and size of meteor strikes on Mars.

There are likely many more impacts that we havent found yet, Daubar said. This advance shows you just how much you can do with veteran missions like MRO using modern analysis techniques.

This new machine learning tool is part of a broader-based NASA/JPL initiative called COSMIC (Content-based On-board Summarization to Monitor Infrequent Change.) That initiative has a motto: Observe much, return best.

The idea behind COSMIC is to create a robust, flexible orbital system for conducting planetary surveys and change monitoring in the Martian environment. Due to bandwidth considerations, many images are never downloaded to Earth. Among other goals, the system will autonomously detect changes in non-monitored areas, and provide relevant, informative descriptions of onboard images to advise downlink prioritization. The AI that finds craters is just one component of the system.

Data management is a huge and growing challenge in science. Other missions like NASAs Kepler planet-hunting spacecraft generated an enormous amount of data. In an effort that parallels what COSMIC is trying to do, scientists are using new methods to comb through all of Keplers data, sometimes finding exoplanets that were missed in the original analysis.

And the upcoming Vera C. Rubin Survey Telescope will be another data-generating monster. In fact, managing all of its data is considered to be the most challenging part of that entire project. Itll generate about 200,000 images per year, or about 1.28 petabytes of raw data. Thats far more data than humans will be able to deal with.

In anticipation of so much data, the people behing the Rubin Telescope developed the the LSSTC Data Science Fellowship Program. Its a two-year program designed for grad school curriculums that will explore topics including statistics, machine learning, information theory, and scalable programming.

Its clear that AI and machine learning will have to play a larger role in space science. In the past, the amount of data returned by space missions was much more manageable. The instruments gathering the data were simpler, the cameras were much lower resolution, and the missions didnt last as long (not counting the Viking missions.)

And though a system designed to find small craters on the surface of Mars might not capture the imagination of most people, its indicative of what the future will hold.

One day, more scientists will be freed from sitting for hours at a time going over images. Theyll be able to delegate some of that work to AI systems like COSMIC and its crater finder.

Well probably all benefit from that.

Like Loading...

Read more from the original source:
Machine Learning Software is Now Doing the Exhausting Task of Counting Craters On Mars - Universe Today

Less-than-one-shot learning enables machine learning algorithms to classify N labels with less than N training examples.

This article is part of ourreviews of AI research papers, a series of posts that explore the latest findings in artificial intelligence.

If I told you to imagine something between a horse and a birdsay, a flying horsewould you need to see a concrete example? Such a creature does not exist, but nothing prevents us from using our imagination to create one: the Pegasus.

The human mind has all kinds of mechanisms to create new concepts by combining abstract and concrete knowledge it has of the real world. We can imagine existing things that we might have never seen (a horse with a long necka giraffe), as well as things that do not exist in real life (a winged serpent that breathes firea dragon). This cognitive flexibility allows us to learn new things with few and sometimes no new examples.

In contrast, machine learning and deep learning, the current leading fields of artificial intelligence, are known to require many examples to learn new tasks, even when they are related to things they already know.

Overcoming this challenge has led to a host of research work and innovation in machine learning. And although we are still far from creating artificial intelligence that can replicate the brains capacity for understanding, the progress in the field is remarkable.

For instance, transfer learning is a technique that enables developers to finetune an artificial neural network for a new task without the need for many training examples. Few-shot and one-shot learning enable a machine learning model trained on one task to perform a related task with a single or very few new examples. For instance, if you have an image classifier trained to detect volleyballs and soccer balls, you can use one-shot learning to add basketball to the list of classes it can detect.

A new technique dubbed less-than-one-shot learning (or LO-shot learning), recently developed by AI scientists at the University of Waterloo, takes one-shot learning to the next level. The idea behind LO-shot learning is that to train a machine learning model to detect M classes, you need less than one sample per class. The technique, introduced in a paper published in the arXiv preprocessor, is still in its early stages but shows promise and can be useful in various scenarios where there is not enough data or too many classes.

The LO-shot learning technique proposed by the researchers applies to the k-nearest neighbors machine learning algorithm. K-NN can be used for both classification (determining the category of an input) or regression (predicting the outcome of an input) tasks. But for the sake of this discussion, well still to classification.

As the name implies, k-NN classifies input data by comparing it to its k nearest neighbors (k is an adjustable parameter). Say you want to create a k-NN machine learning model that classifies hand-written digits. First you provide it with a set of labeled images of digits. Then, when you provide the model with a new, unlabeled image, it will determine its class by looking at its nearest neighbors.

For instance, if you set k to 5, the machine learning model will find the five most similar digit photos for each new input. If, say three of them belong to the class 7, it will classify the image as the digit seven.

k-NN is an instance-based machine learning algorithm. As you provide it with more labeled examples of each class, its accuracy improves but its performance degrades, because each new sample adds new comparisons operations.

In their LO-shot learning paper, the researchers showed that you can achieve accurate results with k-NN while providing fewer examples than there are classes. We propose less than one-shot learning (LO-shot learning), a setting where a model must learn N new classes given only M < N examples, less than one example per class, the AI researchers write. At first glance, this appears to be an impossible task, but we both theoretically and empirically demonstrate feasibility.

The classic k-NN algorithm provides hard labels, which means for every input, it provides exactly one class to which it belongs. Soft labels, on the other hand, provide the probability that an input belongs to each of the output classes (e.g., theres a 20% chance its a 2, 70% chance its a 5, and a 10% chance its a 3).

In their work, the AI researchers at the University of Waterloo explored whether they could use soft labels to generalize the capabilities of the k-NN algorithm. The proposition of LO-shot learning is that soft label prototypes should allow the machine learning model to classify N classes with less than N labeled instances.

The technique builds on previous work the researchers had done on soft labels and data distillation. Dataset distillation is a process for producing small synthetic datasets that train models to the same accuracy as training them on the full training set, Ilia Sucholutsky, co-author of the paper, told TechTalks. Before soft labels, dataset distillation was able to represent datasets like MNIST using as few as one example per class. I realized that adding soft labels meant I could actually represent MNIST using less than one example per class.

MNIST is a database of images of handwritten digits often used in training and testing machine learning models. Sucholutsky and his colleague Matthias Schonlau managed to achieve above-90 percent accuracy on MNIST with just five synthetic examples on the convolutional neural network LeNet.

That result really surprised me, and its what got me thinking more broadly about this LO-shot learning setting, Sucholutsky said.

Basically, LO-shot uses soft labels to create new classes by partitioning the space between existing classes.

In the example above, there are two instances to tune the machine learning model (shown with black dots). A classic k-NN algorithm would split the space between the two dots between the two classes. But the soft-label prototype k-NN (SLaPkNN) algorithm, as the OL-shot learning model is called, creates a new space between the two classes (the green area), which represents a new label (think horse with wings). Here we have achieved N classes with N-1 samples.

In the paper, the researchers show that LO-shot learning can be scaled up to detect 3N-2 classes using N labels and even beyond.

In their experiments, Sucholutsky and Schonlau found that with the right configurations for the soft labels, LO-shot machine learning can provide reliable results even when you have noisy data.

I think LO-shot learning can be made to work from other sources of information as wellsimilar to how many zero-shot learning methods dobut soft labels are the most straightforward approach, Sucholutsky said, adding that there are already several methods that can find the right soft labels for LO-shot machine learning.

While the paper displays the power of LO-shot learning with the k-NN classifier, Sucholutsky says the technique applies to other machine learning algorithms as well. The analysis in the paper focuses specifically on k-NN just because its easier to analyze, but it should work for any classification model that can make use of soft labels, Sucholutsky said. The researchers will soon release a more comprehensive paper that shows the application of LO-shot learning to deep learning models.

For instance-based algorithms like k-NN, the efficiency improvement of LO-shot learning is quite large, especially for datasets with a large number of classes, Susholutsky said. More broadly, LO-shot learning is useful in any kind of setting where a classification algorithm is applied to a dataset with a large number of classes, especially if there are few, or no, examples available for some classes. Basically, most settings where zero-shot learning or few-shot learning are useful, LO-shot learning can also be useful.

For instance, a computer vision system that must identify thousands of objects from images and video frames can benefit from this machine learning technique, especially if there are no examples available for some of the objects. Another application would be to tasks that naturally have soft-label information, like natural language processing systems that perform sentiment analysis (e.g., a sentence can be both sad and angry simultaneously).

In their paper, the researchers describe less than one-shot learning as a viable new direction in machine learning research.

We believe that creating a soft-label prototype generation algorithm that specifically optimizes prototypes for LO-shot learning is an important next step in exploring this area, they write.

Soft labels have been explored in several settings before. Whats new here is the extreme setting in which we explore them, Susholutsky said.I think it just wasnt a directly obvious idea that there is another regime hiding between one-shot and zero-shot learning.

See original here:
Machine learning with less than one example - TechTalks

Machine learning is a complex field of science that has to do with scientific research and a deep understanding of computer science. Your vendor must have proven experience in this field.

In this post, we have collected 15 top machine learning companies worldwide. Each of them has at least 5 years of experience, has worked on dozens of ML projects, and enjoys high rankings on popular online aggregators. We have carefully studied their portfolios and what ex-clients say about working with them. Contracting a vendor from this list, you can be sure that you receive the highest quality.

Best companies for machine learning

1. Serokell

Serokell is a software development company that focuses on R&D in programming and machine learning. Serokell is the founder of Serokell Labs an interactive laboratory that studies new theories of pure and applied mathematics, academic and practical applications of ML.

Serokell is an experienced, fast-growing company that unites qualified software engineers and scientists from all over the world. Combining scientific research and data-based approach with business thinking, they manage to deliver exceptional products to the market. Serokell has experience working with custom software development in blockchain, fintech, edtech, and other fields.

2. Dogtown Media

Dogtown Media is a software vendor that applies artificial intelligence and machine learning in the field of mobile app development. AI helps them to please their customers with outstanding user experience and help businesses to scale and develop. Using machine learning for mobile apps, they make them smarter, more efficient, and accurate.

Among the clients of Dogtown Media are Google, Youtube, and other IT companies and startups that use machine learning daily.

3. Iflexion

This custom software development company covers every aspect of software engineering including machine learning.

Inflexion has more than 20 years of tech experience. They are proficient at building ML-powered web applications for e-commerce as well as applying artificial intelligence technologies for e-learning, augmented reality, computer vision, and big data analytics. In their portfolio, you can find a dating app with a recommender system, a travel portal, and countless business intelligence projects that prove their expertise in the field.

4. ScienceSoft

ScienceSoft is an experienced provider of top-notch IT services that works across different niches. They have a portfolio full of business-minded projects in data analytics, internet of things, image analysis, and e-commerce.

Working with ScienceSoft, you trust your project in the hands of R&D masters who can take over the software development process. The team makes fast data-driven decisions and delivers high-quality products in reduced time.

5. Increon

If you are looking for an innovative software development company that helps businesses to amplify their net impact to customers and employees, pay attention to Increon.

This machine-learning software vendor works with market leaders in different niches and engineers AI strategies for their business prosperity. Icreon has firsthand, real-world experience building out applications, platforms, and ecosystems that are driven by machine learning and artificial intelligence.

6. Hidden Brains

Hidden Brains is a software development firm that specializes in AI, ML, and IoT. During 17 years of their existence, they have used their profound knowledge of the latest technologies to deliver projects for healthcare, retail, education, fintech, logistics, and more.

Hidden Brains offers a broad set of machine learning and artificial intelligence consulting services, putting the power of machine learning in the hands of every startupper and business owner.

7. Imaginovation

Imaginovation was founded in 2011 and focuses on web design and development. It actively explores all the possibilities of artificial intelligence in their work.

The agencys goal is to boost the business growth of its clients by providing software solutions for recommendation engines, automated speech and text translation, and effectiveness assessment. Most high-profile clients are Nestle and MetLife.

8. Cyber Infrastructure

Cyber Infrastructure is among the leading machine learning companies with more than 100 projects in their portfolio. With their AI solutions, they have impacted a whole variety of industries: from hospitality and retail to fintech and Hitech.

The team specializes in using advanced technologies to develop AI-powered applications for businesses worldwide. Their effort to create outstanding projects has been recognized by Clutch, Good Firms, and AppFutura.

9. InData Labs

InData Labs is a company that delivers a full package of AI-related services including data strategy and AI consulting and AI software development. They have plenty of experience working with the technologies of machine learning, NLP, computer vision, and predictive modeling.

InData Labs analyses its clients capabilities and needs, designs a future product concept, inserts the ML system into any production type, and improves the previously built models.

10. Spire Digital

Spire Digital is one of the most eminent AI development companies in the USA. They have worked on more than 600 cases and have deep expertise in applying AI in the fields of finance, education, logistics, healthcare, and media. Among other tasks, Spire Digital helps with building and integrating AI into security systems and smart home systems.

Over more than 20 years, the company has managed to gain main awards including #1 Software Developer In The World from Clutch.co and Fastest Growing Companies In America from Inc. 5000.

Conclusion

Working with a top developer, you choose high-quality software development and extensive expertise in machine learning. They apply the most cutting-edge technologies in order to help your business expand and grow.

Media ContactCompany Name: SerokellContact Person: Media RelationsEmail: Send EmailPhone: (+372) 699-1531Country: EstoniaWebsite: https://serokell.io/

Original post:
Top Machine Learning Companies in the World - Virtual-Strategy Magazine