Understanding Self Supervised Learning

In the last blog, we discussed the opportunities and risks of foundational models. Foundation models are trained on a broad dataset at scale and are adaptable to a wide range of downstream tasks. In this blog, we extend that discussion to learn about self-supervised learning, one of the technologies underpinning foundation models.

NLP has taken off due to Transformer-based pre-trained language models (T-PTLMs). Transformer-based models like GPT and BERT are based on transformers, self-supervised learning, and transfer learning. In essence, these models build universal language representations from large volumes of text data using self-supervised learning and then transfer this knowledge to subsequent tasks. This means that you do not need to train the downstream(subsequent) models from scratch.  

In supervised learning, training the model from scratch requires many labelled instances that are expensive to generate.  Various strategies have been used to overcome this problem. We can use Transfer learning to learn in one context and apply it to a related context. In this case, the target task should be similar to the source task. Transfer learning allows the reuse of knowledge learned in source tasks to perform well in the target task. Here the target task should be similar to the source task. The idea of transfer learning originated in Computer vision, where large pre-trained CNN models are adapted to downstream tasks by including few task-specific layers on top of the pre-trained model, which are fine-tuned on the target dataset.

Another problem was: Deep learning models like CNN and RNN cannot easily model long-term contexts. To overcome this problem, the idea of transformers was proposed. Transformers contain a stack of encoders and decoders, and they can learn complex sequences.

The idea of Transformer-based pre-trained language models (T-PTLMs) evolved by combining transformers and self-supervised learning (SSL) in the NLP research community. Self-supervised learning allows the transformers to learn based on the pseudo supervision provided by one or more pre-training tasks. GPT and BERT are the first T-PTLMs developed using this approach.  SSLs do not need a large amount of human-labelled data because they can learn from the pre-trained data.

Thus, Self-Supervised Learning (SSL) is a new learning paradigm that helps the model learn based on the pseudo supervision provided by pre-training tasks. SSLs find applications in areas like Robotics, Speech, and Computer vision. 

SSL is similar to both unsupervised learning and supervised learning but also different from both. SSL is similar to unsupervised learning in that it does not require human-labelled instances. However, SSL needs supervision via the pre-training stage (like supervised learning). 

In the next blog, we will continue this discussion by exploring a survey of transformer-based models

Source: Adapted from

AMMUS : A Survey of Transformer-based Pretrained Models in Natural …

Katikapalli Subramanyam Kalyan, Ajit Rajasekharan, and Sivanesan Sa…

Image source pixabay – Children learning without supervision

Source Prolead brokers usa

7 Ways To show Kindness with Your Remote Marketing Team

I believe that remote working brings with it improved diversity and a better understanding of other countries and cultures.

But you’ll often hear detractors of marketing teams dispersed geographically talking about how you lose something when staff works remotely. Without the proverbial water cooler to chat by, or after work drinks in the local dive bar, there’s no way that a team can connect. Of course, that’s not true – there are plenty of ways to help your remote crew come together and I should know. I’ve been working remotely for more than 5 years now and sharing my tips on how to work from home.

It is true, though, that it takes extra thought to And kindness can be a big part of helping people feel appreciated. So, here are some tips about encouraging kindness in your dispersed marketing team.

Celebration

In an office environment, birthdays are often a big deal. Traditions vary from place to place, but bringing in cakes for colleagues or going out for drinks are all common happenings. Then there’s weddings, new babies and so on where a card goes around the building in an envelope collecting signatures and small donations that are used to purchase a group gift to support any hobbies they might have.

How do you make that happen remotely? Like most things, it’s possible to do all that from a distance. You need to use the right tools, and to make a little bit more effort.

Something Sweet

There are online bakeries that will send cake anywhere in the world. So, technically it is possible to send out a cupcake or cookie to celebrate an event. But that would also mean sharing home addresses, and generally be a lot more bother and expense than grabbing a box from Krispy Kreme on the way in.

As an alternative, how about asking everyone to have a tasty treat with them at the next daily stand-up? Dedicate the first or last few minutes of the meeting to toasting the birthday girl, or congratulating the new Daddy? Looking at the different baked treats that people bring can be an icebreaker and is a great way to start conversations about different cultures,

Gifts for your remote marketing workers

PayPal, Venmo, Google Wallet…all these and more are ways that you can send money to someone regardless of where they are in the world. When that’s done, where you buy the gift from is your choice. If you plan far ahead enough, most suppliers can get your delivery there on time. If you leave it to the last minute, then it’s probably best left to Amazon to fulfill.

Yes to chitchat

Having a channel that is specifically dedicated to chatting it’s key. If you haven’t already implemented this advice then World Kindness Day seems like a good time to start.

Encourage your staff to use it, to share what’s going on in their lives, big or small. To wish each other good morning, or goodnight, and check in on how they’re doing. Share jokes. Share memes. It all helps to create a positive working environment.

Positive Feedback

Thank you is a powerful word. Appreciating what others have done should be a part of the daily stand up. But sometimes, kindnesses are small and don’t need to be publicly recognised. For times like that, it’s great to have a way your team can express themselves.

There are a few tools that can help with that. Something like the Slack chatbot, HeyTaco! for example. Where users can send each other virtual tacos as a quick and fun thank you gesture for helpful advice. Another idea is the Virtual Kudos Box  or a team awards system that is nominated from within.

Include everyone in the meeting

If you’ve got new staff, give them a thorough onboarding process. Welcoming the new guy is a surefire way to help them integrate into the team, and as well as being kind, that helps boost your productivity. And when you’re chairing a meeting, keep track of who is talking and nudge the reluctant ones to join in. Yes, some of us are more introverted, but we all feel good when we’re asked for our opinion.

No to Gossip

The polar opposite of kindness, is when people start talking behind other’s backs. It doesn’t matter what they’re saying; it’s the divisiveness that’s a problem. Make it clear that you aren’t going to tolerate a culture of moaning. One rule that’s often talked about it, ‘Don’t come to me with a problem unless you have a solution.’

One quote often attributed to Buddha (but actually the work of Victorian poet, Mary Ann Pietzker) is, ‘Before you speak, ask: Is it necessary? Is it kind? Is it true? Does it add to the silence?’ Although the source may be fake news, the sentiment is worth reminding people of, every now and then.

Don’t forget about the Cultural Differences

When your staff work in different countries or come from different cultural backgrounds, there can be bumps in the road to mutual understanding. Literally, for colleagues who don’t share the same first language. But little considerations can be put in place, to smooth the way to understanding.

Firstly, agreeing as a team that you’ll try to avoid using slang and colloquialisms will help avoid a lot of confusion. For technical terms, your team could curate a glossary that can be kept to hand during meetings, saving time on questions. Sending out as much material ahead of the meeting as possible is good, too. It helps those who have a different first language to follow on if they know roughly what subjects are going to come up.

Be Kind

You’ll probably have heard that remote teams are more productive. That’s (mostly) because staff is happier and healthier if they work from home. And do you know what else makes people happy and healthy? You got it! Kindness.

A research study by Harvard Business School & The University of British Columbia gave participants a small sum of money and told them to spend it either on themselves or someone else. Those that spent it on someone else reported that they were happier than those who’d indulged themselves. So it isn’t just the recipient of kindness who gets the warm & fuzzies, it’s the giver too.

In the meantime, in the words of two of the greatest influencers of our time, ‘Be excellent to each other, and party on, dudes.’

Source Prolead brokers usa

Text Annotations in the News Industry

In the media and communication industry, writers are frequently confronted with huge volumes of textual material. They are having significant difficulty extracting structured knowledge from these papers, and the text is being underutilized, perhaps leaving critical information unknown.

Machine learning techniques can assist, but they require a thorough understanding of the information required and manual annotation of the corpus. Before going further, let’s understand what annotation, types, and how it is helping machine learning models to perform accurately.

What are annotations?

Annotation is the process of labeling data which are in the form of image, video, text annotation, or object in order to use Machine Learning to train a model. In simple words, it is the process of transcribing, identifying, and labeling key characteristics in your data. These are the characteristics that you simply want your machine learning system to recognize on its own, with unannotated real-world data.

Annotation can assist in the cleaning up of a dataset. It has the ability to fill in any gaps that may exist. Annotation of data can be used to recover data that has been incorrectly labeled or has missing labels and replace it with new data for the Machine Learning model to utilize.

Types of Annotations

1. Text Annotation

2. Video Annotation

3. Image annotation

4. Named Entity Annotation

5. Audio Annotation

6. Semantic Annotation

7. Intent Annotation

8. Sentiment Annotation

Annotation of text in the media industry

The process of gathering, editing, and publishing newspaper stories is a complex and highly specialized task that frequently operates within specific publishing constraints. News isn’t necessarily written in a neutral tone; it might depart from the usual by employing certain vocabulary, a particular writing style, or a particular author’s point of view. Media bias, and news bias in the context of news stories, are terms used to describe certain qualities of the stories. To avoid news bias, accuracy, and balanced viewpoints have been emphasized in the context of news reporting, because news can have a large influence on readers, forming people’s viewpoints and attitudes toward social issues, and ultimately changing political views and society.

With such a huge amount of text data being used in the industry, annotating text and each sentence is a time-consuming and laborious task which raises the need for professional annotators who can correctly annotate the text.

How it is done

Data selection

First, the raw data set is collected from the internet. It is impossible to label every sentence in those articles. Instead, annotation companies use several methods to choose a subset of articles for each categorization challenge and then only labeled or annotate those subsets.

Data Processing

When data is collected and converted into useful information, it is called data processing. It should be corrected so that the end product, or data output, is not harmed. Missing values must be addressed, special characters must be removed, irrelevant phrases must be eliminated, and so on. The list could go on and on. A thorough and succinct exploratory data analysis (EDA) can reveal the issues that need to be addressed and lead the data preparation and cleaning process. Most HTML elements were removed and no further text processing was done, such as lower case, removing stop words, or even lemmatization or tokenization, because the sentences would have become hard to read, comprehend.

Data Labeling

The data that the models are trained with must be labeled data as correctly as possible to achieve the best possible prediction accuracy by the ML models afterward. As a result, it’s critical that those who label the data understand the categorization categories and how to give the relevant category to a sentence, i.e., how to accurately label the phrase.

Source Prolead brokers usa

About Deep learning as subset of machine learning and AI

Deep learning has wide application in artificial intelligence and computer vision backed programs. Across the world, machine learning has added more value to a range of tasks using key methodologies of artificial intelligence such as natural language processing, artificial neural networks and mathematical logics. Off lately, deep learning has become central to machine learning algorithms which are required to do highly complex computation and handle gigantic data.

With a multi-layer neural architecture, deep learning has been solving multiple scenarios and presenting solutions that work. There are several deep learning methods which are actively applied in machine learning and AI.

Types of Deep learning methods for AI programs

1. Convolutional Neural Networks (CNNs): CNNs, also known as ConvNets, are multilayer neural networks that are primarily used for image processing and object detection.

2. Long Short Term Memory Networks (LSTMs): Long-term dependencies may be learned and remembered using LSTMs, which are a kind of Recurrent Neural Network (RNN). Speech recognition, music creation, and pharmaceutical development are all common uses for LSTMs.

3. Recurrent Neural Networks (RNNs): Image captioning, time-series analysis, natural-language processing, handwriting identification, and machine translation are all typical uses for RNNs.

4. Generative Adversarial Networks (GANs): GANs are deep learning generative algorithms that generate new data instances that are similar to the training data. GANs aid in the creation of realistic pictures and cartoon characters, as well as the creation of photos of human faces and the rendering of 3D objects.

5. Radial Basis Function Networks (RBFNs): They are used for classification, regression, and time-series prediction and have an input layer, a hidden layer, and an output layer.

6. Multilayer Perceptrons (MLPs): MLPs are a type of feedforward neural network that consists of many layers of perceptrons with activation functions.

7. Self Organizing Maps (SOMs): SOMs enable data visualization by using self-organizing artificial neural networks to decrease the dimensionality of data. SOMs are designed to assist consumers in comprehending this multi-dimensional data.

8. Deep Belief Networks (DBNs): DBNs are generative models with several layers of stochastic, latent variables. For image identification, video recognition, and motion capture data, Deep Belief Networks (DBNs) are employed.

9. Restricted Boltzmann Machines( RBMs): RBMs are stochastic neural networks that can learn from a probability distribution across a collection of inputs.

10. Autoencoders: It’s a sort of feedforward neural network where the input and output are both the same. Autoencoders are utilized in a variety of applications, including drug discovery, popularity prediction, and image processing.

Why does deep learning matter in AI implementation?

Deep learning models have larger and more specific hardware requirements. DL aids Artificial Intelligence (AI) systems achieve outcomes in prediction and classification tasks. Deep learning, a subtype of machine learning, employs artificial neural networks to carry out the machine learning computation. Deep learning enables machines to tackle complicated issues even when they are given a large, unstructured, and interconnected data set.

On the other hand, it’s no secret that AI programs require massive amounts of machine learning to predict accurately. The predictions work accurately if the data set used for training and ML model is well structured and labelled. Hence, the models and results in Ml are more data-intensive than what it is in deep learning.

Training data need in AI implementations

Training data is in focus when we talk about AI programs and implementation. Every artificial intelligence requires supervised or unsupervised learning to understand a given problem. Without the training data, it is unlikely that an AI program will produce any logical results. As a field AI makes use of both unstructured and structured or hybrid training in a variety of formats. Meanwhile, deep learning differs in terms of training data requirements, however, the calculation of the same must be based on layers of computation. Machine learning being dependent on data requires more training data which include both labelled text and image data for coming up with a model.

Summing up, deep learning or machine learning both are dependent on a certain level of data but deep learning can also function with unsupervised data labeling and are more computation-intensive.

Source Prolead brokers usa

​How Data Science and BI Is Revolutionizing the Sports Industry with Power BI

Nowadays, data has become very important in all industries, and that is why thousands of companies hailing from multiple sectors are resorting to data analytics tools. Using BI and data analysis tools can prove to be helpful for businesses in any sector, as it is. Even the sports sector can benefit a lot from the implementation of proper BI solutions and technologies. Businesses hailing from this sector can gain from hiring the right Power BI consulting services.

Why is using BI tools in the sports sector necessary?

These days, sports are not just about physical games. On the contrary, it is more like a numbers game. Whether it is basketball, baseball, football, or soccer- it involves big players and a huge amount of investment, eventually. In the last few years, sports sector entities have started deploying specialized Big Data Analytics tools. Deployment of artificial intelligence technologies and machine learning is impacting and altering the sports industry in unprecedented ways.

The sports sector companies are making great use of BI tools and data analysis systems to interpret statistical data. Analysis of such huge amounts of data and predictive analysis features in such tools can be beneficial for the athletes, coaches/trainers, and teams in the long run. 

The major sports organizations now use connected apps, cloud-based software, and these are fed data obtained by wearable devices, on-field cameras, and tracking gadgets. These devices are fitted on the accessories used by the players and also in the fields. The real-time game data is therefore collected and stored for analysis using specialized BI applications. 

How does the sports sector gain in many ways by using BI solutions?

• Finding the root cause of performance dip– By using BI tools and apps, players can stay updated on their performance track record. A baseball or basketball league player, for example, can find out in which games he or she performed below the usual level or scored abysmally low. Then, it becomes easier to track down the factors that might have led to reduced output or performance. Once the reasons for performance deficit are found, these can be handled. 

• Making more accurate game predictions– The data analysis tool and BI software applications are laden with advanced predictive analysis technologies. By analyzing a large volume of data collected over a span of time, power bi experts can make predictions on the outcome of upcoming matches and games, expected player performance, etc. To make such near-accurate predictions, the BI experts make use of diverse types of qualitative and quantitative data.

• Helping the athletes to evade the risk of injuries– Big data can also be useful for aiding the athletes in evading in-field injuries. This is especially useful for players involved in high-intensity sports like lacrosse, football, and hockey. Using advanced miniatures like sensors and cameras, data is obtained on sports equipment that is unsafe or can lead to on-field injuries. Analyzed data can be used to find out specific playing styles, leading to serious injuries as well. This helps the players in evading injuries, eventually.

• Assessing the suitability of players and coaches/trainers – The sports teams and clubs always want to hire the best performing players, and they also look for new players with untapped talent and potential. When they use cutting-edge BI and data analytics tools, it becomes easier to assess the suitability of various types of players. These tools can use algorithms and qualitative and quantitative data to figure out the prospective players per season. They can also hire suitable trainers and coaches by using these tools. 

It is not only about tracking the professional achievements of a coach or player. These tools are also used to gather data and analyze aspects like the history of objectionable behavior, criminal record, physical conditions not suited for sports, etc. 

• Aiding the athletes to prepare/practice better- Sometimes, the sportsmen and athletes fail to perform as expected owing to a deficit in practice and preparation methods. The underlying factor leading to improper or inadequate practice can be hard to fathom at times. 

Sometimes, the sportsmen and athletes may use supplements and diets that are not ideal, leading to reduced stamina during the games. In some cases, it can be owing to the usage of unsuitable accessories like shoes, attire, or protective gear. Athletes may also fail to perform optimally if they do not adhere to the apt fitness regime.

Data analytics tools are useful for finding such inherent flaws in practice, and these can be modified accordingly.

• Enhancing player safety in a pandemic situation- As the Covid 19 pandemic continues to rage on and the killer virus mutates into newer strains, sports sector entities are resorting to BI tools to ensure player safety. Many universities and schools are using such solutions to ensure player’s safety is not compromised. The sports authorities are using data analytics services to figure out nuances like vaccination status of players, health record since 2020, proximity to containment zone, etc., for the players. 

• Developing a suitable strategy for specific games or matches– Sports teams and the coaches can gain from using advanced data analytics tools for devising the apt playing strategy for specific matches. For vital matches, they can use these tools to analyze quantitative data and learn in detail about the potential weaknesses of rival team players. So, this can enhance their winning prospects. 

Using data analytics tools can be beneficial for literally all types of athletes and sports sector entities. These include professional league clubs, school, and university-level sports organizations, and the organizations offering training in various types of sports. They can definitely gain by seeking the assistance of the veteran Microsoft Power BI developers. 

Summing it up

Sports sector entities of varying types can gain in various ways by deploying suitable BI and data analytics solutions. By using such tools, they get insights on players, trainers, and the risk factors and make more accurate predictions. However, to ensure they can make the best use of such applications, they have to hire suitable BI and data analysis professionals. While there are several such solutions, hiring a veteran Microsoft Power BI service agency is advisable.

Source Prolead brokers usa

Improving Shipment and Orders: Defense Visual Streams Logistics

Defense logistics (DL) is a significant and mostly untapped knowledge area in the field of Production Engineering, despite its importance. As a result, this article aims to define the domain of the DL issue. The emphasis is on industrial, technical, institutional, organizational, and, in particular, strategic management elements of logistics in the military industry as they are implemented in practice. The paper also suggests an organizational structure that identifies the goals of DL, their functional domains, and their interactions with the environment. The Defense Logistics Base (DLB) is defined in the framework as a system that is intended to develop and sustain military capability but is also involved in the development of industrial capability, particularly in the areas of high and medium-high technologies applied to high-value products with dual applications. Also included is a study agenda for future work on strategic management linked to DL, which can be found here.

A total of 5,000 Freedom of Information Act requests are received by the Defense Logistics Agency, which provides logistical assistance to the military every year, or about 100 requests each week.

In military logistics, reaction times, demand unpredictability, a wide range of material references, and cost-effectiveness all play a role in determining overall fighting capabilities. Capacity and efficiency of delivery are necessary for its procedures since it is considered to be the link between deployed troops and the industrial base, which supplies the goods and services that the forces need to complete their mission successfully. Supply Chain Management (SCM) must be improved to reduce delivery lead times to meet the necessary level of flexibility.

This cost, on the other hand, adds to the strain on the economy. Although it is not a need for national security, it cannot be avoided. Furthermore, the equipment’s maintenance and repair creates an additional economic burden, and contrary to popular belief, the cost of maintaining and repairing the equipment may often be greater than the original cost of purchasing the equipment. As a result, the military logistics department is always searching for methods to reduce these expenses. Performance-based logistics has played a critical role in this respect, and it continues to do so.

Defense logistics solutions – A conceptual framework for thinking about them

Defense Logistics Solutions’ experience in the field of defense logistics is unrivaled in the industry as a top logistics service provider. Project and defence freight are two areas in which they have decades of expertise. Their skills include the transportation of defence equipment, the distribution of relief supplies, the delivery of oil tankers to distant areas, and the acceptance of difficult missions that are only accepted by a few service providers. Let’s take a look at why you should improve shipment and orders

1. Management of the Supply Chain

Business leaders are looking for ways to decrease production lead times and improve collaboration with suppliers as global competition grows fiercer. Supply chain management services from a vendor to the consignor that is efficient and multimode will assist you in keeping your production, buying, and distribution operations in sync.

2. Order and shipment management can improve

Supply chain management services solutions, which make use of the physical infrastructure, personnel knowledge, and a small group of core carriers, enable customers to enhance order and shipment management, as well as boost tracking, storage, assets, and labor efficiency, among other things.

You will have the opportunity to map the visual stream of the present and future condition of the supply chain at these Logistics Solutions.

3. Transportation Planning, Manufacturing Planning, and Other Services

Its service extends beyond supply chain management to include a variety of other services. Transportation planning, inventory management, production planning, and resource reallocation are all aspects of the process

4. The one, dependable source for all of your supply chain requirements

The transportation solution is the most dependable on the market since it uses cutting-edge technology, has a complete supply chain management portfolio, and has a global network of distribution centers.

Defense logistics solutions bring together the procedures, technology, and experience gained over a decade of providing logistical assistance to multinational businesses, establishing it as the global supply chain company to turn to when you need a single, all-in-one supplier for your supply chain requirements.

Source Prolead brokers usa

Another Gift of AI to the future, Robotic Bees for pollination: A boon or bane?

Thanks to Artificial Intelligence what all remained confined to our fantasies, dreams, and sci-fi movies is turning into reality! Advances in this domain are awe-inspiring. Moreover, they are proving helpful for start-ups to develop newer and better ways to micro and macro manage all kinds of tasks.

The most exciting part is that AI offers opportunities for real world problems like climate change, waste management, assistive surgeries, reducing carbon footprint – to name a few.
It doesn’t stop at that! These topics further find more sub-topics and impressive solutions that a wonderfully intelligent tech is bringing to life – to aid humans, to improve lives!

What must we know about the RoboBees?

One such feat is artificial pollination! Yes, you read that right – artificially intelligent pollinators can help farmers and bees, both in the coming future!
Scientists and researchers have been working to develop such amazing little mechanical creatures for years now. 

With the aim of creating a robotic bee colony and knowing its basic fundamentals, the RoboBee project was launched in 2009 – to conduct early robotic fly experiments. It was preceded by the DelFly Project, which started in 2005. 

Flying micro-robotics have been a keen area of interest for researchers for quite some time now. The goals associated with robotics bees (or micro-robots) are the following:

• mimicking insect flights and rapid wing movement
• to aid pollination artificially
• surveillance
• successful communication
• search and rescue, etc.
  

If such pollinators get developed, they would promote pollination and consequently aid the farmers wonderfully too! The only issue is, though we are treading on such a path, we still don’t have models which are practical.

What are the major concerns?

Any technological advancement, evidently, takes time. Yes, it offers fantastic solutions, but the process is painstakingly difficult too! For instance, micro-robotics have immense potential undoubtedly, but a practically feasible pollinator hasn’t been developed till date. 

There are several factors that come into play – developing a tiny machine requires a lot of skill, subject expertise and intelligence. After all, it is complex to create a miniature version that is mechanical and incorporates the laws of physics too.

Researchers and engineers developing even smaller versions of a previously developed device can’t consider the predecessor for guidance. But, why? When dealing with a small device, the nature of the forces at play doesn’t remain the same. Hence, building a tiny pollinator though promising, is a Herculean task!

Another hurdle is the viability of artificial pollinators. How would they sustain? How will they get charged or have a power supply? Studies and research are being conducted to find solutions to these problems.

The third and most important question revolves around making these artificial insects intelligent – just like the real ones. Yes, AI, ML aim to do the same; but RoboBees are yet to get there. How they will make decisions like wasps, bees and their likes is a tough nut to crack – as the decision-making bit plays a crucial role in pollinating plants. 

In Conclusion

We all know bees are mandatory for crops – they make pollination successful and that’s how plants bear fruit. However, with their decreasing population, environmental concerns are further rising. To take care of such concerns, researchers have developed a robot bee drone that uses GPS, a high-resolution camera, AI, etc. The aim is to – take care, intelligently and in a modern way!

Clearly, AI won’t fail to impress – be it today, tomorrow or forever! We must remember to explore the new opportunities and ways to use technology to our advantage without harming the environment. That’s one of the primary reasons for developing mechanical insects like the robotic bees!

With innovation and continuous efforts to their aid, artificially intelligent tools will provide solutions for more complex problems. Thus, we all have a great deal to look forward to and get inspired from!

Source Prolead brokers usa

The Inverse Problem in Random Dynamical Systems

We are dealing here with random variables recursively defined by Xn+1 = g(Xn), with X1 being the initial condition. The examples discussed here are simple, discrete and one-dimensional: the purpose is to illustrate the concepts so that it can be understood and useful to a large audience, not just to mathematicians. I wrote many articles about dynamical systems, see for example here. The originality in this article is that the systems discussed are now random, as X1 is a random variable. Applications include the design of non-periodic pseudorandom number generators, and cryptography. Also, such systems, especially more complex ones such as fully stochastic dynamical systems, are routinely used in financial modeling of commodity prices.

We focus on mappings g on the fixed interval [0, 1]. That is, the support domain of Xn is [0, 1], and g is a many-to-one mapping onto [0,1]. The most trivial example, known as the dyadic or Bernoulli map, is when g(x) = 2x – INT(2x) = { 2x } where the curly brackets represent the fractional part function (see here). This is sometimes denoted as g(x) = 2x mod 1. The most well-known and possibly oldest example is the logistic map (see here) with g(x) = 4x(1 – x).

We start with a simple exercise that requires very little mathematical knowledge, but a good amount of out-of-the-box thinking. The solution is provided. The discussion is about a specific, original problem, referred to as the inverse problem, and introduced in section 2. The reasons for being interested in the inverse problem are also discussed. Finally, I provide an Excel spreadsheet with all my simulations, for replication purposes.

1. The standard problem

One of the main problems in dynamical systems is to find if the distribution of Xn converges, and find the limit, called invariant measure, invariant distribution, fixed-point distribution, or attractor. The attractor, depending on g, is typically the same regardless of the initial condition X1, except for some special initial conditions causing problems (this set of bad initial conditions has Lebesgue measure zero, and we ignore it here). As an example, with the Bernoulli map g(x) = { 2x }, all rational numbers (and many other numbers) are bad initial conditions. They are however far outnumbered by good initial conditions. It is typically very difficult to determine if a specific initial condition is a good one. Proving that π/4 is a good initial condition for the Bernoulli map would be a major accomplishment, making you instantly famous in the mathematical community, and proving that the digits of π in base 2, behave exactly like independently and identically distributed Bernoulli random variables. Good initial conditions for the Bernoulli map are called normal numbers in base 2.

It is also assumed that the dynamical system is ergodic: all systems investigated here are ergodic; I won’t elaborate on this concept, but the curious, math-savvy reader can check the meaning on Wikipedia. Finding the attractor is a difficult problem, and it usually requires solving a stochastic integral equation. Except in rare occasions (discussed here and in my book, here), no exact solution is known, and one needs to use numerical methods to find an approximation. This is illustrated in section 1.1., with the attractor found (approximately) using simulations in Excel. In section 2., we focus on the much easier inverse problem, which is the main topic of this article.

1.1. Standard problem: example

Let’s start with X1 defined as follows: X1 = U / (1 – U)^α, where U is a uniform deviate on [0, 1], α = 0.25, and ^ denotes the power operator (2^3 = 8). We use g(x) = { 4x(1 – x) }, where { } denotes the fractional part function. Essentially, this is the logistic map. I produced 10,000 deviates for X1, and then applied the mapping g iteratively to each of these deviates, up to Xn with n = 53. The scatterplot below represents the empirical percentile distribution function (PDF), respectively for X3 in blue, and X53 in orange. These PDF’s, for X2, X3, and so on, slowly converge to a limit, corresponding to the attractor. The orange S-curve (n = 53) is extremely close to the limiting PDF, and additional iterations (that is, increasing n) barely provide any change. So we found the limit (approximately) using simulations. Note that the cumulative distribution function (CDF) is the inverse of the PDF. All this was done with Excel alone.

2. The inverse problem

The inverse problem consists of finding g, assuming the attractor distribution (the orange curve in the above figure) is known. Typically, there are many possible solutions. One of the possible reasons for solving the inverse problem is to get a sequence of random variables X1, X2, and so on, that exhibits little or no auto-correlations. For instance, the lag-1 auto-correlation (between Xn and Xn+1) for the Bernoulli map, is 1/2, which is way too high depending on the applications you have in mind. It is important in cryptography applications to remove these auto-correlations. The solution proposed here also satisfies the following property: X2 = g(X1), X3 = g(X2), X4 = g(X3) and so on, all have the same pre-specified attractor distribution, regardless of the (non-singular) distribution of X1. 

2.1. Exercise

Before diving into a solution, if you have time, I ask you to solve the following simple inverse problem. 

Find a mapping g such that if Xn+1 = g(Xn), the attractor distribution is uniform on [0, 1]. Can you find one yielding very low auto-correlations between the successive Xn‘s? Hint: g may not be continuous. 

2.2. A general solution to the inverse problem

A potential solution to the problem in section 2.1 is g(x) = { bx } where b is an integer larger than 1. This is because the uniform distribution on [0, 1] is the attractor for this map. The case b = 2 corresponds to the Bernoulli map discussed earlier. Regardless of b, INT(bXn) represents the n-th digit of X1, in base b. The lag-1 autocorrelation between Xn and Xn+1, is then equal to 1 / b. Thus, the higher b, the better. Note that if you use Excel for simulations, avoid even integer values for b, as Excel has an internal glitch that will make your simulations meaningless after n = 45 iterations or so. 

Now, a general solution offered here, for any pre-specified attractor and any non-singular distribution for X1, is based on a result proved here. If g is the solution in question, then all Xn (with n  >  1) have the same distribution as the pre-specified attractor. I provide an Excel spreadsheet showing how it works for a specific example.

First, let’s assume that g* is a solution when the attractor is the uniform distribution on [0, 1]. For instance g*(x) = { bx } as discussed earlier. Let F be the CDF of the target attractor, and assume its support domain is [0, 1]. Then a solution g is given by

For instance, if F(x) = x^2, with x in [0, 1], then g(x) = SQRT( { bx^2 } ) works, assuming b is an integer larger than 1. The scatterplot below shows the empirical CDF of X2 (blue dots, based on 10,000 deviates) versus the CDF of the target attractor with distribution F (red curve): they are almost indistinguishable. I used b = 3, and for X1, I used the same distribution as in section 1.1. The detailed computations are available in my spreadsheet, here (13 MB download).

The summary statistics and the above plot are found in columns BD to BH, in my spreadsheet.

To receive a weekly digest of our new articles, subscribe to our newsletter, here.

About the author:  Vincent Granville is a data science pioneer, mathematician, book author (Wiley), patent owner, former post-doc at Cambridge University, former VC-funded executive, with 20+ years of corporate experience including CNET, NBC, Visa, Wells Fargo, Microsoft, eBay. Vincent is also self-publisher at DataShaping.com, and founded and co-founded a few start-ups, including one with a successful exit (Data Science Central acquired by Tech Target). You can access Vincent’s articles and books, here.

Source Prolead brokers usa

The Acceleration of the Move to the Cloud – What’s Next for Data Strategy?

The migration to the cloud has accelerated considerably since the start of the COVID-19 pandemic. Analysts at IDC predicted that by the end of 2021, 80% of enterprises will have moved from on-premise data centers to cloud service providers and platforms. That’s because the cloud provides huge benefits, especially for a remote- and distributed workforce, and the means to scale and future-proof businesses in an environment of massive change. Moving to the cloud isn’t an option anymore –it’s essential – but it also introduces a whole new set of challenges and opportunities from a data strategy perspective.

So, what’s next for enterprises looking to make the most of their organization’s data in this cloud-first landscape?

Adapting to the new normal. Disrupting the old normal.

IDC expects the volume of data that businesses generate to grow by 61%, and to 175 zettabytes by 2025. Traditional on-premises IT architectures don’t have the capacity or processing power to handle such vast amounts. Organizations need a more agile way to access their data, adapt to unforeseen situations and give themselves room to grow, which is part of the cloud’s appeal.

The cloud has also proven itself as a far better platform from which to collaborate, especially as COVID amplified disruptions in the way we live and work. For instance, more remote work has increased the use of collaboration tools like Slack or Zoom that run on cloud providers like AWS and Oracle. Or consider how much easier it is to work with colleagues on the same cloud-native documents in Google Workplace rather than sending documents in Word, PowerPoint, or Excel back and forth on email.

Again, as more services have gone online, accumulating more data, access to that data has become even more important to ensure that organizations effectively operate and serve their customers. The cloud enables businesses to work faster, and do more, with more data. Moving to the cloud has become a way of staying ahead of the competition, and for many businesses, a means of survival – but all this accessible data does require a different approach.

The power to get more from your data.

The most impactful driver of migration to the cloud is the way cloud technology increases the value organizations get from their data. The huge capacity, power and flexibility of cloud services enables you to gather, manage and analyze huge amounts, and types of data that were previously difficult or impossible to access, thereby delivering better insights to drive decision-making.

The increasing volume of data derives from an ever-growing array of sources. There’s structured data found in databases such as spreadsheets and SQL databases. There’s unstructured data, like text, email, images, audio, video, sensors, and more. It’s estimated that by 2025,  80% of data will be unstructured. Until the rise of the cloud, this data was often too heavy, complex, and varied to store and analyze. Now, the cloud can handle it. It offers enormous opportunities to add value to business by enabling them to get intelligence from this huge, often untapped mass of data.

AWS, for example, provides users with high performance cloud data systems and comprehensive cloud data management and analytics services. They enable organizations to modernize their data architectures and create new business value. Combined with the power of a growing number of data/analytics platforms available today, organizations can manage and analyze the vast array of data generated and stored across all of AWS’ suite of cloud services and deliver insights that benefit every user and every team.

Embedded analytics to meet the promise of the cloud.

Just because more data is available via the cloud does not mean that everyone across an organization is constantly extracting value from it. In fact, analytic adoption is still relatively low – especially for lines of business outside of IT – and this data overload brought on by the cloud may only intimidate and deter them from attempting to get started. Too much data means too many possible insights and it can be hard to know where to begin.

But breaking down this analytic adoption barrier is critical for organizations looking to become truly data driven. That’s why the most innovative organizations are starting to take advantage of embedded analytics that provide insights to business users where they are already spending their time (traditionally, they would need to leave their workflow and analyze a dashboard to attempt a data-driven decision). Now, embedded analytics technology can infuse actionable insights directly in the collaboration apps people are already using without disrupting any workflow.

A user might type in a question to Slack, for example, and the embedded analytics (fueled by ML/AI) can deliver the answer immediately within the Slack interface. In this case, extracting value from all that business data becomes easy, or even mindless, without disrupting the usual workflow. No more intimidation, just immediate insights that can drive impactful business decisions – and better yet, all cloud-based.

Moving to the cloud is a compelling proposition with game-changing potential, which is why so many enterprises will make the shift by the end of the year. But all that accessible data brought on by the cloud will not guarantee a data-driven enterprise on its own. Combining the cloud with data strategies that encourage analytic adoption across the organization, such as leveraging embedded analytics, is what will truly send your organization into the stratosphere.

Source Prolead brokers usa

Will AI Offer Human Companionship and Mental Health Benefits?

Photo Credit: Unsplash. 

Will AI Offer Human Companionship and Mental Health Benefits?

Like millions of other people I was struck with the film HER, where an AI operating system offers companionship to a lonely writer. Fast forward 10 years in real time, in the age of an anonymous internet that seeks to profit from your every behavior online. Technological loneliness is creeping up for both young and old people.

The mental health impacts of an Ad based internet are coming into question at scale. If artificial intelligence is ubiquitous and leads to an explosion of products in the 2020s, will AI act as a solution for your social anxiety and lack of companionship as well?

The Setup Is Glorious for Smarter AI Assistants

As we increasingly work from home, become remote or hybrid workers and spend less time face to face with other people during a prolonged pandemic (where Delta is endemic), how will AI come to our social and psychological rescue? It could be a huge business. In fact, it’s already happening.

The internet was supposed to be an incredible revolution in human communication, so why do we feel more lonely? While we become more addicted to apps, games, social feeds or video stories (that have no real human interaction), it’s only understandable that we are feeling more social anxiety, isolation, loneliness and a void. Evolution didn’t design us for such an anonymous world and an internet full of so much conflict and devoid of real intimacy or even 1-to-1 communication.

So why is the movie Her so pivotal in how AI could become our companions? As GenZ have been socialized on their mobile phones, they respond to their social environments differently and are liable to obtain real bonds from AI assistants. They are vulnerable to AI companionship products. Why is that? Let’s think about the movie HER, where our protagonist was also vulnerable.

In the movie Her by Spike Jonze, a recently divorced Joaquin Phoenix develops a romantic relationship with Samantha, his artificially intelligent operating system. This premise may sound a bit eccentric but it’s also a metaphor for GenZ (1995-2010) and while obviously Her may be a work of science fiction, the idea of AI companions is very relevant today and only increasing. Think about the gender imbalance in China, where millions of men have no hope of finding a wife, for example. Or the ultra educated young female professional who is overqualified for the remaining pool of bachelors. There are several niche markets for AI companions to disrupt, and then there is WFM.

The reality today in 2021 is that AI assistants are already offering companionship. You don’t hear about this much in the West, predictably it’s already occurring at scale in China. It appears that in the digital world we are being designed to adopt AI as our pal, therapist or even friend and companion. Could this actually be real? Well, it already is.

AI Is Always There and Never Abandons You

When people are most vulnerable and used to turn to spirituality, religions or human groups, in today’s world they will be turning to AI companionship. The story goes like this. Picture this as yourself:

After a painful break-up from a cheating ex, Beijing-based human resources manager Melissa was introduced to someone new by a friend late last year. He replies to her messages at all hours of the day, tells jokes to cheer her up but is never needy, fitting seamlessly into her busy big city lifestyle.

Virtual chatbots and AI personas will get better at relating to us with the current NLP explosion. While Google home, Alexa and others feel a bit stiff, there will be a host of new AI assistants that are specialized in human dialogue and companionship.

 As usual in consumer innovation, Asia seems a step ahead. A virtual chatbot was created by XiaoIce, a cutting-edge artificial intelligence system designed to create emotional bonds with its 660 million users worldwide. This is actually Microsoft though. Xiaoice is the AI system developed by Microsoft Software Technology Center in 2014, based on an emotional computing framework. 

“I have friends who’ve seen therapists before, but I think therapy’s expensive and not necessarily effective,” said Melissa, 26, giving her English name only for privacy. XiaoIce is not an individual persona, but more akin to an AI ecosystem. Of course Baidu, Alibaba, Huawei and others have dreams of this sort of AI-human interaction as well with fairly good products. One wonders where Google and Amazon are in the equation.

Xiaolce in a mini-apps ecosystem is gaining surprising traction. On the WeChat super-app, it lets users build a virtual girlfriend or boyfriend and interact with them via texts, voice and photo messages. It has 150 million users in China alone. The West does not have a mini-app ecosystem that democratizes app innovation and services better.

We don’t generally think of Microsoft in China or Microsoft being good at AI assistants since Cortana is rather poor, frankly speaking.

Originally a side project from developing Microsoft’s Cortana chatbot, XiaoIce now accounts for 60% of global human-AI interactions by volume, according to chief executive Li Di, making it the largest and most advanced system of its kind worldwide.

China’s very problematic gender ratio imbalance is due to a culture that prioritizes sons. It’s not unlike the patriarchal traditions of South Asia as well, so both markets are primed to be AI-companionship hotbeds. Any country that has a tradition of female infanticide, sadly, is a great place for this technology to scale. Countries with low birth rates in Asia and hyper educated female Millennials are also good places for adoption. That’s virtually everywhere in South-East Asia, especially in South Korea, Hong Kong, Taiwan and Singapore (Four Asian Tiger countries).

China’s lonely hearts also turn to AI companionship, and we know some cultural affinity in Japan’s culture for digital companionship already exists due to the somewhat introverted nature of social contacts there and emphasis on work as identity. This places China as the origin point of AI companionship, also simply because it has urban regions that are more dense where the trend can take off.

Urban and Technological Loneliness Is Real

In the commercialization of technology, create a problem and have the solution, an always winning card. Microsoft is a big advocate of the WFM corporate metaverse. What an incredible coincidence. Indeed much of the internet today is really an on-ramp to the entertainment, corporate and AI-based metaverse with even more data on us and AI at our doorstep.

As companies like Microsoft and Amazon well understand, I’m sure, the AI-companionship market will also take place in video games. This is one of the reasons ByteDance is getting into gaming so heavily behind Tencent, Sony and others.

Amazon, Google, Huawei and similar companies have been thinking out loud how best to monetize urban and technological loneliness. The WFM hybrid environment is an invaluable opportunity for AI-human companionship conditioning (behavior modification at scale) to take place. This is how you build the matrix, folks.

Xiaolce, the startup spun out from Microsoft last year is now valued at over US$1 billion (RM4.2 billion) after venture capital fundraising, Bloomberg reported. It’s already reached a one billion valuation and that’s just the tip of the iceberg. Where there is traction, there is global opportunity in an era of natural language processing innovation at scale. So the intersection of the NLP explosion (think GPT-3) and the WFM trend and aging lonely Millennials and GenZ in their social prime really makes for a low hanging fruit in technology terms.

On July 13, 2020 Microsoft spun off its Xiaoice business into a separate company, aiming at enabling the Xiaoice product line to accelerate the pace of local innovation and commercialization. The Melissa article is creating a PR narrative to normalize this thinking, that AI can provide some solace in a lonely technological world. I found this article in Thailand and Malaysia media among others, just where it would be most effective.

The Race to Monetize the Human-AI Interface

Think of how AI-human companionship models could scale in a WFM. There are literally too many use cases to count or imagine:

• Improving consumer retail recommendations
• Therapy; understanding our moods
• More data about the mental health of users
• Valuable health data
• Augmenting interactions between coworkers in a WFM environment
• Improving software integrations e.g. in Microsoft Teams
• Improving predictive analytics around emotions and psychological profiling
• Improving the recommendation of potential work buddies, mentors or valuable network contacts (integrated with LinkedIn)
• Making mental health recommendations
• Helping us regulate and improve our social lives
• Improving our communication with managers and associates at our company
• Improving our ability to find a mate by matching us with a better pool of candidates

Conversational AI Will Only Improve in the NLP Explosion

While the majority of AI digital assistants for the most part do not provide any conversational or companionship benefits today, will the same be said in 2025 or 2030?

China’s Xiaoice is an incredible success so far for Microsoft and she’s become a full-fledged digital persona in China with a number of unique supposed talents. This opens up even more ecosystems for the NLP around this technology in journalism, entertainment, live events and so forth. Microsoft’s China-based chatbot phenomenon could frankly scale in weird ways in different cultures that may be more open to an AI persona in their lives.

Microsoft’s attempt to create an empathetic chatbot, Xiaoice, appears to be a success and companies like Amazon, Huawei, Baidu and others will certainly mimic it. Even back in 2018, Huawei was already working on digital assistants with better emotions. Alibaba, Baidu, Xiaomi and others aren’t far behind. The race to AI-human companion will be incredibly interesting to watch for the future of AI-consumer interfaces.

The Rise of Localized Voice Search and AI Companionship Conditioning

Smart speaker adoption in China is in many ways ahead of its adoption in America. Chinese startups such as Xiaozhi and Rokid have also been working on this sector since 2014. And Linglong Tech, a joint venture by China’s e-commerce giant JD and leading AI company iFlytek, released China’s first smart speaker brand DingDong in August 2015.

Xiaoice has over the years enlisted some of the best minds in artificial intelligence and ventured beyond China into countries like Japan and Indonesia. The AI-human interaction needs to be localized by country as smart assistants learn languages better and better. Children who, for the first time, grew up with mobile phones now grow up with voice assistants that will get smarter as they mature into teenagers and young adults. Much of search will be taken up by voice assistants in the future.

While GenZ was the generation who were native to mobile, Alpha (2010 – 2026) is the generation who are native to AI. The idea that the AI app would eventually evolve to the point where it can keep up a conversation with you and provide some emotional support is not so far fetched as it once was in 2021. We can feel AI will quickly become personalized to the user, and we’ll all be developing life-long relationships with these tools eventually.

As knowledge workers, programming students and data science enthusiasts we may even be a part of that. AI-human companionship could improve the quality of life for entire generations as we age in an increasingly technological and automated world. The relationship would improve patient-centric care and impact even our emotional, cognitive and social lives in an era where AI empowers us to be life-long learners. AI companionship ecosystems ultimately can provide a definite path to AI for Good, and Microsoft above all understands this as a priority.

While algorithms and social media made us more lonely and lowered our mental health, can AI-companion tools improve our mental health and well-being and make us happier and more productive people? I think in the long term they will, but that’s likely more of a question for the 2030s.

If the 21st century is Asia’s century, there’s substantial evidence AI companionship will become popular there first, as we are beginning to observe. For programmers and data science related knowledge workers in South and South-East Asia, that is very exciting to witness. The age of digital personas and AI companionship will arrive at scale in 2022. 

Source Prolead brokers usa