Cloud Hosting

Machine Learning is a big field with new research coming out frequently. It is a hot field where academia and industry keep experimenting with new things to improve our daily lives.

In recent years, generative AI has been changing the world due to the application of machine learning. For example, ChatGPT and Stable Diffusion. Even with 2023 dominated by generative AI, we should be aware of many more machine learning breakthroughs.

Here are the top machine learning papers to read in 2023 so you will not miss the upcoming trends.

Singing Voice Beautifying (SVB) is a novel task in generative AI that aims to improve the amateur singing voice into a beautiful one. Its exactly the research aim of Liu et al. (2022) when they proposed a new generative model called Neural Singing Voice Beautifier (NSVB).

The NSVB is a semi-supervised learning model using a latent-mapping algorithm that acts as a pitch corrector and improves vocal tone. The work promises to improve the musical industry and is worth checking out.

Deep neural network models have become bigger than ever, and much research has been conducted to simplify the training process. Recent research by the Google team (Chen et al. (2023)) has proposed a new optimization for the Neural Network called Lion (EvoLved Sign Momentum). The method shows that the algorithm is more memory-efficient and requires a smaller learning rate than Adam. Its great research that shows many promises you should not miss.

Time series analysis is a common use case in many businesses; For example, price forecasting, anomaly detection, etc. However, there are many challenges to analyzing temporal data only based on the current data (1D data). That is why Wu et al. (2023) propose a new method called TimesNet to transform the 1D data into 2D data, which achieves great performance in the experiment. You should read the paper to understand better this new method as it would help much future time series analysis.

Currently, we are in a generative AI era where many large language models were intensively developed by companies. Mostly this kind of research would not release their model or only be commercially available. However, the Meta AI research group (Zhang et al. (2022)) tries to do the opposite by publicly releasing the Open Pre-trained Transformers (OPT) model that could be comparable with the GPT-3. The paper is a great start to understanding the OPT model and the research detail, as the group logs all the detail in the paper.

The generative model is not limited to only generating text or pictures but also tabular data. This generated data is often called synthetic data. Many models were developed to generate synthetic tabular data, but almost no model to generate relational tabular synthetic data. This is exactly the aim of Solatorio and Dupriez (2023) research; creating a model called REaLTabFormer for synthetic relational data. The experiment has shown that the result is accurately close to the existing synthetic model, which could be extended to many applications.

Reinforcement Learning conceptually is an excellent choice for the Natural Language Processing task, but is it true? This is a question that Ramamurthy et al. (2022) try to answer. The researcher introduces various library and algorithm that shows where Reinforcement Learning techniques have an edge compared to the supervised method in the NLP tasks. Its a recommended paper to read if you want an alternative for your skillset.

Text-to-image generation was big in 2022, and 2023 would be projected on text-to-video (T2V) capability. Research by Wu et al. (2022) shows how T2V can be extended on many approaches. The research proposes a new Tune-a-Video method that supports T2V tasks such as subject and object change, style transfer, attribute editing, etc. Its a great paper to read if you are interested in text-to-video research.

Efficient collaboration is the key to success on any team, especially with the increasing complexity within machine learning fields. To nurture efficiency, Peng et al. (2023) present a PyGlove library to share ML ideas easily. The PyGlove concept is to capture the process of ML research through a list of patching rules. The list can then be reused in any experiments scene, which improves the team's efficiency. Its research that tries to solve a machine learning problem that many have not done yet, so its worth reading.

ChatGPT has changed the world so much. Its safe to say that the trend would go upward from here as the public is already in favor of using ChatGPT. However, how is the ChatGPT current result compared to the Human Experts? Its exactly a question that Guo et al. (2023) try to answer. The team tried to collect data from experts and ChatGPT prompt results, which they compared. The result shows that implicit differences between ChatGPT and experts were there. The research is something that I feel would be kept asked in the future as the generative AI model would keep growing over time, so its worth reading.

2023 is a great year for machine learning research shown by the current trend, especially generative AI such as ChatGPT and Stable Diffusion. There is much promising research that I feel we should not miss because its shown promising results that might change the current standard. In this article, I have shown you 9 top ML papers to read, ranging from the generative model, time series model to workflow efficiency. I hope it helps.Cornellius Yudha Wijaya is a data science assistant manager and data writer. While working full-time at Allianz Indonesia, he loves to share Python and Data tips via social media and writing media.

Read this article:
Top Machine Learning Papers to Read in 2023 - KDnuggets

Machine Learning is one of the approach of Artificial Intelligence in which Machines become capable of drawing intelligent decisions like humans by learning from its past experiences. In classical methods of Artificial Intelligence, step by step instructions are provided to the machines to solve a problem. Machine Learning combines classical methods of Artificial Intelligence with the knowledge of past to gain human like intelligence.

The book Introduction to Machine Learning with Python has made explanation on Machine Learning with Python from basics to the advanced level so as to assist beginners in building strong foundation and develop practical understanding.

The beginners with zero or little knowledge about Machine Learning can gain insight into this subject from this book. This book explains Machine Learning concepts using real life examples implemented in Python.

The book Introduction to Machine Learning with Python present detailed practice exercises for offering a comprehensive introduction to machine learning techniques along with basics of Python. The book leverages algorithms of machine learning in a unique way of describing real life applications. Though not mandatory, some experience with subject knowledge will fasten the learning process.

About the authors:

Dr. Deepti Chopra has done PhD in the area of Natural Language Processing from Banasthali Vidyapith. Currently, she is working as Associate Professor at JIMS Rohini, Sector 5. Dr. Chopra is an author of five books and two MOOCs. Two of her books have been translated into Chinese and one has been translated into Korean. She has 2 Australian Patents and 1 Indian Patent to her credit. Dr. Chopra has several publications in various International Conferences and journals of repute. Her areas of interest include Artificial Intelligence, Natural Language Processing and Computational Linguistics. Her primary research works involve machine translation, information retrieval, and cognitive computing.

Mr. Roopal Khurana is working as Assistant General Manager at Railtel Corporation of India Ltd., IT Park, Shastri Park, Delhi. Currently, he is working in the field of Data Networking, MPLS Technology. He has done BTech in Computer Science and Engineering from GLA University, Mathura, India. He is a technology enthusiast. Previously, he has worked with companies, such as Orange and Bharti Airtel.

Keywords:

Artificial Intelligence, Computer Science and IT, Machine Learning, Deep Learning, Python Programming, Back propagation, Supervised Learning, Scikit Learn, Unsupervised Learning, Numpy, Decision Trees, Matplotlib, Support Vector Machine, pandas, Neural Network, Logistic Regression, Linear regression, Clustering, Jupyter notebook, Classification.

For more information please visit: http://bit.ly/3JDcY3R

Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.

See the original post:
Introduction to machine learning with python - EurekAlert

by Japan Advanced Institute of Science and Technology

Graphene's 2D nature, single molecule sensitivity, low noise, and high carrier concentration have generated a lot of interest in its application in gas sensors. However, due to its inherent non-selectivity, and huge p-doping in atmospheric air, its applications in gas sensing are often limited to controlled environments such as nitrogen, dry air, or synthetic humid air.

While humidity conditions in synthetic air could be used to achieve controlled hole doping of the graphene channel, this does not adequately mirror the situation in atmospheric air. Moreover, atmospheric air contains several gases with concentrations similar to or larger than the analytic gas. Such shortcomings of graphene-based sensors hinder selective gas detection and molecular species identification in atmospheric air, which is required for applications in environmental monitoring, and non-invasive medical diagnosis of ailments.

The research team led by Dr. Manoharan Muruganathan (formerly Senior Lecturer), and Professor Hiroshi Mizuta at the Japan Advanced Institute of Science and Technology (JAIST) employed the machine learning (ML) models trained on various gas adsorption-induced doping and scattering signals to realize both highly sensitive and selective gas sensing with a single device.

The ML models' performances are often dependent on the input features. 'The conventional graphene-based ML models are limited in their input features', says Dr. Osazuwa Gabriel Agbonlahor (formerly post-doctoral research fellow). The existing ML models only monitor the gas adsorption-induced changes in the graphene transfer characteristics or resistance/conductivity without modulating these characteristics by applying an external electric field.

Hence, they miss distinctive van der Waals (vdW) interaction between gas molecules and graphene, which is unique to individual gas molecules. Hence, unlike the conventional electronic nose (e-nose) models, we can map the external electric field modulated graphene-gas interaction, which enables more selective feature extraction for complex gas environments such as atmospheric air.

Our ML models for the identification of atmospheric gases were developed using the graphene sensor functionalized with a porous activated carbon thin film. Eight vdW complex features were used to monitor the effects of the external electric field on the graphene-gas molecule vdW interaction, and consequently mapped the evolution of the vdW bonding before, during, and after the external electric field application.

Furthermore, although the gas sensing experiments were performed under different experimental conditions e.g., gas chamber pressures, gas concentrations, ambient temperature, atmospheric relative humidity, tuning time, and tuning voltage, the developed models were shown to be robust enough to accommodate these variations in experimental conditions by not exposing the models to these parameters.

Moreover, to test the models versatility, they were trained on atmospheric environments as well as relatively inert environments that are often used in gas sensing e.g., nitrogen and dry air. Hence, a high-performance atmospheric gas "electronic nose" was achieved, distinguishing between the four different environments (ammonia in atmospheric air, acetone in atmospheric air, acetone in nitrogen, and ammonia in dry air) with 100% accuracy.

The research is published in the journal Sensors and Actuators B: Chemical.

More information: Osazuwa G. Agbonlahor et al, Machine learning identification of atmospheric gases by mapping the graphene-molecule van der waals complex bonding evolution, Sensors and Actuators B: Chemical (2023). DOI: 10.1016/j.snb.2023.133383

Provided by Japan Advanced Institute of Science and Technology

Read more:
New machine-learning approach identifies one molecule in a billion selectively, with graphene sensors - Phys.org

It is true that identity fraud has become a buzz word among people and experts, but not in a good way. First coined in 1964, identity theft has since turned prolific, whether it is fraudsters seeking to impersonate others in order to either open a credit card account or gangs laundering money without linking transactions to their real-life identities.

Last year, The National Fraud Intelligence Bureau reported that fraud offences increased by 17% in March 2022, reaching 936,276 compared to the year ending March 2021.

A GlobalData survey conducted in September 2022 stated identity theft came among the top three concerns shared among citizens from the UK, US, France, Germany and Poland. Identity theft, as a practice.

Nir Stern is vice president of product management at AU10TIX, a tech company providing intelligence information and the infrastructure needed to combat fraud.

In an interview with Electronic Payments International, Stern talks about how machine learning, data consortiums as well as external sources can effectively be used to combat identity fraud.

Generally speaking, in the world of financial crime and even beyond, you need to differentiate between account takeover and identity fraud.

In the case of an account takeover, as a consumer, you have an existing relationship with a financial institution, and someone is taking over that account by stealing your credentials (username, password or even one-time password).

This is usually done via social engineering or email scams. In the end, a fraudster tries to make the victim do things for them or provide them with information that they need without the victim knowing about the fraud.

With identity fraud, the purpose usually is to pretend to be another person to start a new interaction with the institution either to commit financial crimes or to accomplish other activities. One example could be money laundering to finance terrorist activity.

Because of regulations, you would need an identity with all relevant information. The methods employed would be different.

First, you need to have access to your victims personal information unless you want to create a synthetic ID.

Secondly, you must create a fake ID that is good quality enough to bypass any security measurements financial institutions have.

There are roughly three types of methods.

There is the very elementary and nave one, like trying to take a photo of the victims ID or download an ID image.

Then there are the more sophisticated fraud techniques. You could go on certain websites, pay a certain amount of money and download any type of digital document. Those should be of sufficient quality so that when you hand them to an expert, they will not notice anything suspicious.

Then you have the top quality fraud, where identity fraudsters steal or buy personal information from databases uploaded on the dark web. Then, they create high-quality, sometimes even physical IDs that are almost impossible to detect with all the information or standard methods of going through fraud. Everything will look perfectly normal because the fraudsters have accessed information through a particular type of data breach.

Those are the trickiest to detect because everything looks completely normal. And when we examine web traffic, we see the most sophisticated fraudsters use this.

So, because fraudsters are using the methods mentioned above, none of the standard ID verification measures employed by banks will be helpful, no matter how sophisticated they are.

The only way to detect it with systems like ours is first to use machine learning that has access to millions of legitimate transactions and can spot different indicators showing what counts as fraudulent activity.

Also, we have a unique system called Instinct, a concerted database where we track transactions worldwide. Through our big customers, we keep an eye on a huge amount of legitimate and fraudulent transactions. We securely store that information, avoiding any storage of sensitive data.

These steps enable us to see repetitions. Because when

these fraudsters invest a lot of time, effort, and money in coming up with a specific tactic, they will not do it just once. They will perform mass attacks.

And then we see cases where everything will look perfectly normal if you look at a single transaction. Still, our systems can detect the same face used on multiple IDs or the same ID with different faces.

The combination of sophisticated machine learning to determine indicators of fraudulent activity and using a data consortium is basically the only way to detect those attacks. Unfortunately, many organisations do not use such capabilities to prevent identity fraud, thereby risking their money.

So, the main differentiator is, first, the fact that many of our competitors and ID-proofing solutions need to provide manual reviews behind the scenes. They will capture the information or images sent over to them. But most of the analysis will be done by human beings. And, as I have mentioned, with all these sophisticated, super complex forgeries, a human being cannot identify ID fraud because the activity looks normal even when you go into detail with your examination.

Thats why first, you need to have a fully automated system based on machine learning/AI to detect those tiny indicators of forgeries that are hard to see otherwise.

Moreover, you also need the capabilities like we do with Instinct to look through a data consortium, detect synthetic fraud or mass attacks, and see whether those attacks are repetitive.

You dont bring a knife to a gunfight. That is probably the key element in fighting identity fraud. If you look at the reports that are always shared, the value of identity theft-related fraud is hundreds of millions. You see specific gangs of identity thieves making millions out of it. As a result, they invest a lot of money into acquiring new technology to commit fraud. So you need to have the equivalent of that technology to fight.

Therefore, you cannot have a solution based on human beings/agents that are supposed to detect identity fraud. It is almost impossible for them.

Next, on top of having a fully automated system, you also need to have a multi-layered approach.

That means not only searching to pinpoint fraudulent activity but also having access to data consortiums, as well as looking into external data sources to determine whether the fraudulent activity is one-off or not.

Thirdly, fraudsters adapt all the time, so you need a fully adaptable system that keeps on learning, a system with access to information not limited to your customers. If you only monitor your customer environment, when the next attack comes, they might not be prepared for it unless they have enough experience.

The other side of this discussion we should have brought up is the legitimate user experience.

If you think about these banks, they want to be secure and protected from fraud. Around 99.9% of their traffic consists of legitimate persons. In that sense, you want to ensure they have the best user experience, especially when you are a digital bank. That is your only channel; you dont have a backup physical branch.

You need to ensure you have a smoother digital user experience and make it secure and comfortable for your customers.

The key for these companies is to have the back-end systems able to detect these forgeries and the front-end solutions that help you, as a legitimate user, take the best profile photos and quickly become a customer.

When using these front-end solutions, make sure that the quality of customer images enables you, as a company, to detect ID theft and prevent fraudsters from using things like external cameras to create deep-fake photos, as well as upload images, or screenshot pictures of people or their IDs.

So all of that, combining a sophisticated front-end solution and a sophisticated back-end solution, is the direction many digital banks and financial institutions take and what traditional banks should do more as well.

When doing ID or digital verification, if you are not a small local bank but a global one willing to extend to other countries, you need companies like AU10TIX to do it for you.

We support over 200 countries and over 4,000 types of IDs. It is virtually impossible for a bank to develop a system that can manage various national ID formats while also successfully detecting forgeries.

Thats not their business. Their business is to give the best user experience, so I believe the best approach to combat fraud is for them to get expert help.

The riskiest thing about blockchain or decentralised financial transactions is that you dont have any indication of where the money goes, no way to turn it back, and no information about who is interacting with the system. So in the case of any account takeovers or social engineering acts that happen, there is nothing you can do.

There are different transaction monitoring techniques for decentralised IDs. A lot of times, these techniques make sure the money doesnt go to unknown wallets. But if its something like a fraudster opening the digital wallet and stealing the money there is very little you can do.

One of the trends we see, and were strategically investing a lot in it, is decentralised ID. It is when you issue a digital ID, which is encrypted and tokenised on your mobile device or digital wallet. That information is not shared with anyone; the signature is kept in the blockchain.

Because it is based on these standards, if you need to use the identification system, you can just share simple information or claims like am I allowed to do it? Am I over 18? Am I a citizen of a particular country?

We see a lot of hype around this technology. Because its all around self-survey identity and significantly related to data privacy, which is key to decentralised finance, the combination of the two may actually be a solution that benefits everyone. You can keep your privacy without the need to be part of the financial institution ecosystem while still making it secure in a way that guarantees only you have access to it.

Continued here:
Machine learning and data consortium can fight identity theft: AU10TIX - Electronic Payments International