semi-supervised learning Archives

Reading Time: 20 minutes

What is Artificial Intelligence?
How do we measure if the AI is acting like a human?
How Artificial Intelligence works?
What are the three types of Artificial Intelligence?
What is the purpose of AI?
Where is AI used?
What are the disadvantages of AI?
Applications of Artificial Intelligence in business?
What is ML?
What are the different kinds of Machine Learning?
What is Deep Learning?
What is NLP?
What is Python?
What is Computer Vision?
What are neural networks?
Conclusion

What is Artificial Intelligence?

The short answer to What is Artificial Intelligence is that it depends on who you ask. A layman with a fleeting understanding of technology would link it to robots. They’d say AI is a terminator like-figure that can act and think on its own. An AI researcher would say that it’s a set of algorithms that can produce results without having to be explicitly instructed to do so. And they would all be right. So to summarise, AI is:

– An intelligent entity created by humans.

– Capable of performing tasks intelligently without being explicitly instructed.

– Capable of thinking and acting rationally and humanely.

How do we measure if the AI is acting like a human?

Even if we reach that state where an AI can behave as a human does, how can we be sure it can continue to behave that way? We can base the human-likeness of an AI entity with the:

– Turing Test

– The Cognitive Modelling Approach

– The Law of Thought Approach

– The Rational Agent Approach

Let’s take a detailed look at how these approaches perform:

What is the Turing Test?

The basis of the Turing Test is that the AI entity should be able to hold a conversation with a human agent. The human agent ideally should not able to discern that they are talking to an AI. To achieve these ends, the AI needs to possess these qualities:

– Natural Language Processing to communicate successfully.

– Knowledge Representation to act as its memory.

– Automated Reasoning to use the stored information to answer questions and draw new conclusions.

– Machine Learning to detect patterns and adapt to new circumstances.

Cognitive Modelling Approach

As the name suggests, this approach tries to build an AI model-based on Human Cognition. To distil the essence of the human mind, there are 3 approaches:

– Introspection: observing our thoughts, and building a model based on that

– Psychological Experiments: conducting experiments on humans and observing their behaviour

– Brain Imaging – Using MRI to observe how the brain functions in different scenarios and replicating that through code.

The Laws of Thought Approach

The Laws of Thought are a large list of logical statements that govern the operation of our mind. The same laws can be codified and applied to artificial intelligence algorithms. The issues with this approach, because solving a problem in principle (strictly according to the laws of thought) and solving them in practice can be quite different, requiring contextual nuances to apply. Also, there are some actions that we take without being 100% certain of an outcome that an algorithm might not be able to replicate if there are too many parameters.

The Rational Agent Approach

A rational agent acts to achieve the best possible outcome in its present circumstances.

According to the Laws of Thought approach, an entity must behave according to the logical statements. But there are some instances, where there is no logical right thing to do, with multiple outcomes involving different outcomes and corresponding compromises. The rational agent approach tries to make the best possible choice in the current circumstances. It means that it’s a much more dynamic and adaptable agent.

Now that we understand how AI can be designed to act like a human, let’s take a look at how these systems are built.

How Artificial Intelligence works?

Building an AI system is a careful process of reverse-engineering human traits and capabilities in a machine, and using it’s computational prowess to surpass what we are capable of. AI can be built over a diverse set of components and will function as an amalgamation of:

– Philosophy

– Mathematics

– Economics

– Neuroscience

– Psychology

– Computer Engineering

– Control Theory and Cybernetics

– Linguistics

Let’s take a detailed look at each of these components.

Philosophy

The purpose of philosophy for humans is to help us understand our actions, their consequences, and how we can make better decisions. Modern intelligent systems can be built by following the different approaches of philosophy that will enable these systems to make the right decisions, mirroring the way that an ideal human being would think and behave. Philosophy would help these machines think and understand about the nature of knowledge itself. It would also help them make the connection between knowledge and action through goal-based analysis to achieve desirable outcomes.

Also Read: Artificial Intelligence and The Human Mind: When will they meet?

Mathematics

Mathematics is the language of the universe and system built to solve universal problems would need to be proficient in it. For machines to understand logic, computation, and probability are necessary.

The earliest algorithms were just mathematical pathways to make calculations easy, soon to be followed by theorems, hypotheses and more, which all followed a pre-defined logic to arrive at a computational output. The third mathematical application, probability, makes for accurate predictions of future outcomes on which AI algorithms would base their decision-making.

Economics

Economics is the study of how people make choices according to their preferred outcomes. It’s not just about money, although money the medium of people’s preferences being manifested into the real world. There are many important concepts in economics, such as Design Theory, operations research and Markov decision processes. They all have contributed to our understanding of ‘rational agents’ and laws of thought, by using mathematics to show how these decisions are being made at large scales along with their collective outcomes are. These types of decision-theoretic techniques help build these intelligent systems.

Neuroscience

Since neuroscience studies how the brain functions and AI is trying to replicate the same, there’s an obvious overlap here. The biggest difference between human brains and machines is that computers are millions of times faster than the human brain, but the human brain still has the advantage in terms of storage capacity and interconnections. This advantage is slowly being closed with advances in computer hardware and more sophisticated software, but there’s still a big challenge to overcome as are still not aware of how to use computer resources to achieve the brain’s level of intelligence.

Psychology

Psychology can be viewed as the middle point between neuroscience and philosophy. It tries to understand how our specially-configured and developed brain reacts to stimuli and responds to its environment, both of which are important to building an intelligent system. Cognitive psychology views the brain as an information processing device, operating based on beliefs and goals and beliefs, similar to how we would build an intelligence machine of our own.

Many cognitive theories have already been codified to build algorithms that power the chatbots of today.

Computer Engineering

The most obvious application here, but we’ve put this the end to help you understand what all this computer engineering is going to be based on. Computer engineering will translate all our theories and concepts into a machine-readable language so that it can make its computations to produce an output that we can understand. Each advance in computer engineering has opened up more possibilities to build even more powerful AI systems, that are based on advanced operating systems, programming languages, information management systems, tools, and state-of-the-art hardware.

Control Theory and Cybernetics

To be truly intelligent, a system needs to be able to control and modify its actions to produce the desired output. The desired output in question is defined as an objective function, towards which the system will try to move towards, by continually modifying its actions based on the changes in its environment using mathematical computations and logic to measure and optimise its behaviours.

Linguistics

All thought is based on some language and is the most understandable representation of thoughts. Linguistics has led to the formation of natural language processing, that help machines understand our syntactic language, and also to produce output in a manner that is understandable to almost anyone. Understanding a language is more than just learning how sentences are structured, it also requires a knowledge of the subject matter and context, which has given rise to the knowledge representation branch of linguistics.

What are the 3 types of Artificial Intelligence?

Not all types of AI uses all the above fields simultaneously. Different AI entities are built for different purposes, and that’s how they vary. The three broad types of AI are:

– Artificial Narrow Intelligence (ANI)

– Artificial General Intelligence (AGI)

– Artificial Super Intelligence (ASI)

Let’s take a detailed look.

What is Artificial Narrow Intelligence (ANI)?

This is the most common form of AI that you’d find in the market now. These AI systems are designed to solve one single problem and would be able to execute a single task really well. By definition, they have narrow capabilities, like recommending a product for an e-commerce user or predicting the weather. This is the only kind of AI that exists today. They’re able to come close to human functioning in very specific contexts, and even surpass them in many instances, but only excelling in very controlled environments with a limited set of parameters.

What is Artificial General Intelligence (AGI)?

AGI is still a theoretical concept. It’s defined as AI which has a human-level of cognitive function, across a wide variety of domains such as language processing, image processing, computational functioning and reasoning and so on.

We’re still a long way away from building an AGI system. An AGI system would need to comprise of thousands of Artificial Narrow Intelligence systems working in tandem, communicating with each other to mimic human reasoning. Even with the most advanced computing systems and infrastructures, such as Fujitsu’s K or IBM’s Watson, it has taken them 40 minutes to simulate a single second of neuronal activity. This speaks to both the immense complexity and interconnectedness of the human brain, and to the magnitude of the challenge of building an AGI with our current resources.

What is Artificial Super Intelligence (ASI)?

We’re almost entering into science-fiction territory here, but ASI is seen as the logical progression from AGI. An Artificial Super Intelligence (ASI) system would be able to surpass all human capabilities. This would include decision making, taking rational decisions, and even includes things like making better art and building emotional relationships.

Once we achieve Artificial General Intelligence, AI systems would rapidly be able to improve their capabilities and advance into realms that we might not even have dreamed of. While the gap between AGI and ASI would be relatively narrow (some say as little as a nanosecond, because that’s how fast AI would learn) the long journey ahead of us towards AGI itself makes this seem like a concept that lays far into the future.

What is the Purpose of AI?

The purpose of AI is to aid human capabilities and help us make advanced decisions with far-reaching consequences. That’s the answer from a technical standpoint. From a philosophical perspective, AI has the potential to help humans live more meaningful lives devoid of hard labour, and help manage the complex web of interconnected individuals, companies, states and nations to function in a manner that’s beneficial to all of humanity.

Currently, the purpose of AI is shared by all the different tools and techniques that we’ve invented over the past thousand years – to simplify human effort, and to help us make better decisions. AI has also been touted as our Final Invention, a creation that would invent ground-breaking tools and services that would exponentially change how we lead our lives, by hopefully removing strife, inequality and human suffering.

That’s all in the far future though – we’re still a long way from those kinds of outcomes. Currently, AI is being used mostly by companies to improve their process efficiencies, automate resource-heavy tasks, and to make business predictions based on hard data rather than gut feelings. As all technology that has come before this, the research and development costs need to be subsidised by corporations and government agencies before it becomes accessible to everyday laymen.

Where is AI used?

AI is used in different domains to give insights into user behaviour and give recommendations based on the data. For example, Google’s predictive search algorithm used past user data to predict what a user would type next in the search bar. Netflix uses past user data to recommend what movie a user might want to see next, making the user hooked onto the platform and increase watch time. Facebook uses past data of the users to automatically give suggestions to tag your friends, based on their facial features in their images. AI is used everywhere by large organisations to make an end user’s life simpler. The uses of AI would broadly fall under the data processing category, which would include the following:

– Searching within data, and optimising the search to give the most relevant results

– Logic-chains for if-then reasoning, that can be applied to execute a string of commands based on parameters

– Pattern-detection to identify significant patterns in large data set for unique insights

– Applied probabilistic models for predicting future outcomes

What are the disadvantages of AI?

As is the case with any new and emerging technology, AI has its fair share of drawbacks too such as:

– Cost overruns

– Dearth of talent

– Lack of practical products

– Lack of standards in software development

– Potential for misuse

Let’s take a closer look.

Cost overruns

What separates AI from normal software development is the scale at which they operate. As a result of this scale, the computing resources required would exponentially increase, pushing up the cost of the operation, which brings us to the next point.

Dearth of talent

Since it’s still a fairly nascent field, there’s a lack of experienced professionals, and the best ones are quickly snapped up by corporations and research institutes. This increases the talent cost, which further drives up AI implementation prices.

Lack of practical products

For all the hype that’s been surrounding AI, it doesn’t seem to have a lot to show for it. Granted that applications such as chatbots and recommendation engines do exist, but the applications don’t seem to extend beyond that. This makes it difficult to make a case for pouring in more money to improve AI capabilities.

Lack of standards in software development

The true value of AI lays in collaboration when different AI systems come together to form a bigger, more valuable application. But a lack of standards in AI software development means that it’s difficult for different systems to ‘talk’ to each other. AI software development itself is slow and expensive because of this, which further acts as an impediment to AI development.

Potential for Misuse

The power of AI is massive, and it has the potential to achieve great things. Unfortunately, it also has the potential to be misused. AI by itself is a neutral tool that can be used for anything, but if it falls into the wrong hands, it would have serious repercussions. In this nascent stage where the ramifications of AI developments are still not completely understood, the potential for misuse might be even higher.

Applications of Artificial Intelligence in business?

AI truly has the potential to transform many industries, with a wide range of possible use cases. What all these different industries and use cases have in common, is that they are all data-driven. Since AI is an efficient data processing system at its core, there’s a lot of potential for optimisation everywhere.

Let’s take a look at the industries where AI is currently shining.

Healthcare:

– Administration: AI systems are helping with the routine, day-to-day administrative tasks to minimise human errors and maximise efficiency. Transcriptions of medical notes through NLP and helps structure patient information to make it easier for doctors to read it.

– Telemedicine: For non-emergency situations, patients can reach out to a hospital’s AI system to analyse their symptoms, input their vital signs and assess if there’s a need for medical attention. This reduces the workload of medical professionals by bringing only crucial cases to them.

– Assisted Diagnosis: Through computer vision and convolutional neural networks, AI is now capable of reading MRI scans to check for tumours and other malignant growths, at an exponentially faster pace than radiologists can, with a considerably lower margin of error.

– Robot-assisted surgery: Robotic surgeries have a very minuscule margin-of-error and can consistently perform surgeries round-the-clock without getting exhausted. Since they operate with such a high degree of accuracy, they are less invasive than traditional methods, which potentially reduces the time patients spend in the hospital recovering.

– Vital Stats Monitoring: A person’s state of health is an ongoing process, depending on the varying levels of their respective vitals stats. With wearable devices achieving mass-market popularity now, this data is not available on tap, just waiting to be analysed to deliver actionable insights. Since vital signs have the potential to predict health fluctuations even before the patient is aware, there are a lot of live-saving applications here.

E-commerce

– Better recommendations: This is usually the first example that people give when asked about business applications of AI, and that’s because it’s an area where AI has delivered great results already. Most large e-commerce players have incorporated AI to make product recommendations that users might be interested in, which has led to considerable increases in their bottom-lines.

– Chatbots: Another famous example, based on the proliferation of AI chatbots across industries, and every other website we seem to visit. These chatbots are now serving customers in odd-hours and peak hours as well, removing the bottleneck of limited human resources.

– Filtering spam and fake reviews: Due to the high volume of reviews that sites like Amazon receive, it would be impossible for human eyes to scan through them to filter out malicious content. Through the power of NLP, AI can scan these reviews for suspicious activities and filter them out, making for a better buyer experience.

– Optimising search: All of the e-commerce depends upon users searching for what they want, and being able to find it. AI has been optimising search results based on thousands of parameters to ensure that users find the exact product that they are looking for.

– Supply-chain: AI is being used to predict demand for different products in different timeframes so that they can manage their stocks to meet the demand.

Human Resources

– Building work culture: AI is being used to analyse employee data and place them in the right teams, assign projects based on their competencies, collect feedback about the workplace, and even try to predict if they’re on the verge of quitting their company.

– Hiring: With NLP, AI can go through thousands of CV in a matter of seconds, and ascertain if there’s a good fit. This is beneficial because it would be devoid of any human errors or biases, and would considerably reduce the length of hiring cycles.

What is ML?

Machine learning is a subset of artificial intelligence (AI) which defines one of the core tenets of AI – the ability to learn from experience, rather than just instructions.

Machine Learning algorithms automatically learn and improve by learning from their output. They do not need explicit instructions to produce the desired output. They learn by observing their accessible data sets and compares it with examples of the final output. The examine the final output for any recognisable patterns and would try to reverse-engineer the facets to produce an output.

What are the different kinds of Machine Learning?

The types of Machine learning are:

– Supervised Learning

– Unsupervised Learning

– Semi-supervised learning

– Reinforcement Learning

What is Supervised Learning?

Supervised Machine Learning applies what it has learnt based on past data, and applies it to produce the desired output. They are usually trained with a specific dataset based on which the algorithm would produce an inferred function. It uses this inferred function to predict the final output and delivers an approximation of it.

This is called supervised learning because the algorithm needs to be taught with a specific dataset to help it form the inferred function. The data set is clearly labelled to help the algorithm ‘understand’ the data better. The algorithm can compare its output with the labelled output to modify its model to be more accurate.

What is Unsupervised Learning?

With unsupervised learning, the training data is still provided but it would not be labelled. In this model, the algorithm uses the training data to make inferences based on the attributes of the training data by exploring the data to find any patterns or inferences. It forms its logic for describing these patterns and bases its output on this.

What is Semi-supervised Learning?

This is similar to the above two, with the only difference being that it uses a combination of both labelled and unlabelled data. This solves the problem of having to label large data sets – the programmer can just label and a small subset of the data and let the machine figure the rest out based on this. This method is usually used when labelling the data sets is not feasible, either due to large volumes of a lack of skilled resources to label it.

Read Also: Top 9 AI Startups in India

What is Reinforcement Learning?

Reinforcement learning is dependent on the algorithms environment. The algorithm learns by interacting with it the data sets it has access to, and through a trial and error process tries to discover ‘rewards’ and ‘penalties’ that are set by the programmer. The algorithm tends to move towards maximising these rewards, which in turn provide the desired output. It’s called reinforcement learning because the algorithm receives reinforcement that it is on the right path based on the rewards that it encounters. The reward feedback helps the system model its future behaviour.

What is Deep Learning?

Deep Learning is a subfield of machine learning concerned with algorithms inspired by the structure and function of the brain called artificial neural networks. Deep Learning concepts are used to teach machines what comes naturally to us humans. Using Deep Learning, a computer model can be taught to run classification acts taking image, text, or sound as an input.

Deep Learning is becoming popular as the models are capable of achieving state of the art accuracy. Large labelled data sets are used to train these models along with the neural network architectures.

Simply put, Deep Learning is using brain simulations hoping to make learning algorithms efficient and simpler to use. Let us now see what is the difference between Deep Learning and Machine Learning.

Deep Learning vs. Machine Learning

How is Deep Learning Used- Applications

Deep Learning applications have started to surface but have a much greater scope for the future. Listed here are some of the deep learning applications that will rule the future.

– Adding image and video elements – Deep learning algorithms are being developed to add colour to the black and white images. Also, automatically adding sounds to movies and video clips.

– Automatic Machine Translations – Automatically translating text into other languages or translating images to text. Though automatic machine translations have been around for some time, deep learning is achieving top results.

– Object Classification and Detection – This technology helps in applications like face detection for attendance systems in schools, or spotting criminals through surveillance cameras. Object classification and detection are achieved by using very large convolutional neural networks and have use-cases in many industries.

– Automatic Text Generation – A large corpus of text is learnt by the machine learning algorithm and this text is used to write new text. The model is highly productive in generating meaningful text and can even map the tonality of the corpus in the output text.

– Self-Driving cars – A lot has been said and heard about self-driving cars and is probably the most popular application of deep learning. Here the model needs to learn from a large set of data to understand all key parts of driving, hence deep learning algorithms are used to improve performance as more and more input data is fed.

– Applications in Healthcare – Deep Learning shows promising results in detecting chronic illnesses such as breast cancer and skin cancer. It also has a great scope in mobile and monitoring apps, and prediction and personalised medicine.

Why is Deep Learning important?

Today we can teach machines how to read, write, see, and hear by pushing enough data into learning models and make these machines respond the way humans do, or even better. Access to unlimited computational power backed by the availability of a large amount of data generated through smartphones and the internet has made it possible to employ deep learning applications into real-life problems.

This is the time for deep learning explosion and tech leaders like Google are already applying it anywhere and everywhere possible.

The performance of a deep learning model improves with an increase in the amount of input data as compared to Machine Learning models, where performance tends to decline with an increase in the amount of input data.

What is NLP?

A component of Artificial Intelligence, Natural Language Processing is the ability of a machine to understand the human language as it is spoken. The objective of NLP is to understand and decipher the human language to ultimately present with a result. Most of the NLP techniques use machine learning to draw insights from human language.

What are the different steps involved in NLP?

The steps involved in implementing NLP are:

– The computer program collects all the data required. This includes database files, spreadsheets, email communication chains, recorded phone conversations, notes, and all other relevant data.

– An algorithm is employed to remove all the stop words from this data and normalizes certain words which have the same meaning.

– The remaining text is divided into groups known as tokens.

– The NLP program analyzes results to spot deduce patterns, their frequency and other statistics to understand the usage of tokens and their applicability.

Where is NLP used?

Some of the common applications that are being driven by Natural Language Processing are:

– Language translation application

– Word processors to check grammatical accuracy of the text

– Call centres use Interactive Voice Response to respond to user requests; IVR is an application NLP

– Personal virtual assistants such as Siri and Cortana are a classic example of NLP

What is Python?

Python is a popular object-oriented programming language that was created by Guido Van Rossum and released in the year 1991. It is one of the most widely-used programming languages for web development, software development, system scripting, and many other applications.

Why is Python so popular?

There are many reasons behind the popularity of Python as the much-preferred programming language, i.e.,

– The easy to learn syntax helps with improved readability and hence the reduced cost of program maintenance

– It supports modules and packages to encourage code re-use

– It enables increased productivity as there is no compilation step making the edit-test-debug cycle incredibly faster

– Debugging in Python is much easier as compared to other programming languages

Read Also: Top Interview Questions for Python

Where is Python used?

Python is used in many real-world applications such as:

– Web and Internet Development

– Applications in Desktop GUI

– Science and Numeric Applications

– Software Development Applications

– Applications in Business

– Applications in Education

– Database Access

– Games and 3D Graphics

– Network Programming

How can I learn Python?

There is a lot of content online in the form of videos, blogs, and e-books to learn Python. You can extract as much information as you can through the online material as you can and want. But, if you want more practical learning in a guided format, you can sign up for Python courses provided by many ed-tech companies and learn Python along with hands-on learning through projects from an expert which would be your mentor. There are many offline classroom courses available too. Great Learning’s Artificial Intelligence and Machine Learning course have an elaborate module on Python which is delivered along with projects and lab sessions.

What is Computer Vision?

Computer Vision is a field of study where techniques are developed enabling computers to ‘see’ and understand the digital images and videos. The goal of computer vision is to draw inferences from visual sources and apply it towards solving a real-world problem.

What is Computer Vision used for?

There are many applications of Computer Vision today, and the future holds an immense scope.

– Facial Recognition for surveillance and security systems

– Retail stores also use computer vision for tracking inventory and customers

– Autonomous Vehicles

– Computer Vision in medicine is used for diagnosing diseases

– Financial Institutions use computer vision to prevent fraud, allow mobile deposits, and display information visually

What is Deep Learning Computer Vision

Following are the uses of deep learning for computer vision:

– Object Classification and Localisation: It involves identifying the objects of specific classes of images or videos along with their location highlighted usually with a square box around them.

– Semantic Segmentation: It involves neural networks to classify and locate all the pixels in an image or video.

– Colourisation: Converting greyscale images to full-colour images.

– Reconstructing Images: Reconstructing corrupted and tampered images.

What are Neural Networks?

Neural Network is a series of algorithms that mimic the functioning of the human brain to determine the underlying relationships and patterns in a set of data.

What are Neural Networks used for?

The concept of Neural Networks has found application in developing trading systems for the finance sector. They also assist in the development of processes such as time-series forecasting, security classification, and credit risk modelling.

What are the different Neural Networks?

The different types of neural networks are:

– Feedforward Neural Network: Artificial Neuron: Here data inputs travel in just one direction, going in from the input node and exiting on the output node.

– Radial basis function Neural Network: For their functioning, the radial basis function neural network consider the distance between a point from the centre.

– Kohonen Self Organizing Neural Network: The objective here is to input vectors of arbitrary dimension to discrete map comprised of neurons.

– Recurrent Neural Network(RNN): The Recurrent Neural Network saves the output of a layer and feeds it back to the input for helping in predicting the output of the layer.

– Convolutional Neural Network: It is similar to the feed-forward neural networks with the neurons having learn-able biases and weights. It is applied in signal and image processing.

– Modular Neural Networks: It is a collection of many different neural networks, each processing a sub-task. Each of them has a unique set of inputs as compared to other neural networks contributing towards the output.

What are the benefits of Neural Networks?

The three key benefits of neural networks are:

– The ability to learn and model non-linear and complex relationships

– ANNs can generalize models infer unseen relationships on unseen data as well

– ANN does not impose any restrictions on the input variables

Conclusion

Artificial Intelligence has emerged to be the next big thing in the field of technology. Organizations across the world are coming up with breakthrough innovations in artificial intelligence and machine learning. Hence, there are immense opportunities for trained and certified professionals to enter a rewarding career. As these technologies continue to grow, they will have more and more impact on the social setting and quality of life.