Machine learning is an aspect of computer science, artificial intelligence and robotics, if not the most important aspect of it all which deals with the process of giving computer systems which could be artifical intelligence systems, robots the ability to learn. They make use of statistical techniques to effect this ability to learn.

This helps to actively improve on a task and better execute, process, retain and carry out commands given to the system. Moreso, this enables the system to execute such tasks without being explicitly programmed to precisely do so. Machine learning is the ability for an artificial intelligence system to mimic cognitive abilities, human intelligence and improve on it thus, Machine learning helps these AI systems to respond to situations based on their vast level of programming without being expressly programmed to respond specifically to the task. The machine just responds based on series of variables. It assesses the situation and responds  based on its level of programming. Artificial

Intelligence function in a wide variety of areas thus would be needed to do more than just respond directly to foreseen circumstances. This is where machine learning comes in play. It helps robots and artificial intelligence systems to make judgement calls, simulate consciousness in a way that could pass off as human if not paid attention to properly. Cognisis is a relative term and is being able to be projected into lifeless machines like robots to come alive and simulate realness in a way. 

The term machine learning was coined by Arthur Samuel in 1959. Arthur Samuel, an American pioneer in the field of computer gaming and artificial intelligence, coined the term Machine Learning in 1959 while he was at IBM. Although then, only just starting, machine learning grew out of the quest for artificial intelligence which makes it that you can’t speak of machine learning without talking about Artificial intelligence. 

This was in the early days of AI which was only then regarded as an academic discipline. Some researchers proposed theories that it was possible for machines to learn and respond in tow. They were interested in the idea that machines could learn from data and make decisions without help based on the first push, provision of a vast data bank. They wanted machines to act like they were real, see problems and proffer solutions to the problems. This was in theory, possible seeing the machines would be provided with all the data and information required to draw up conclusions based on statistical inference, repetitive occurrence and other factors. They then proceeded attempt an approach the problem with various symbolic methods of which one of them was what was at that time termed “neural networks” These were mostly perceptrons and other models that were later found to be reinventions of the generalized linear models of statistics which led to one of the many discoveries that machine learning has roots in different fields. Probabilistic reasoning was also employed, especially in automated medical diagnosis where the patient would be scanned for probable causes of illnesses. Machine learning’s application has in some way woven deep into all applications of artificial intelligence and in diverse fields too. 

Machine learning is basically the study and construction of algorithms that are designed to learn from and make predictions, generalizations, derivations based on its vast level of programming. Algorithms such as these overcome obstacles by reasoning our a problem, sifting through possible solutions and picking the most likely to succeed option. This is a mimicry of human cognitive function which becomes more advanced as years go by. Machine learning functions by following a strictly static program instructions by making data based predictions or decisions, through building a model from sample inputs. Machine learning is a vast field that has and is being employed in a range of computing tasks where designing and programming with explicit algorithms is needed. This also where good performance is difficult to accomplish or infeasible. It has been applied successfully to such fields incapable of being filled by any human such as sifting thousands data within seconds to come up with a solution. Some examples of its applications include email filtering, detection of network intruders, and computer vision.

Systems that make use of machine learning is Artificial intelligence systems for example. Machine learning comprises of everything AI runs on as they would be a body without a consciousness so to speak. Artificial intelligence has proven quite useful, filling in as personal assistants. They are extremely dedicated to their work and this in turn has put human personal assistants out of the job. Artificial intelligence can serve as helpers in a company, firm or organization. These AI personal assistants are capable of doing bulky amounts of work in short periods of time, draw up schedules for meetings, events, reminders and they have a knack for increased output. This generation of helpers make repetitive tasks in the workplace much easier. Imagine annoying tasks such as repetitive sales entries, travel bookings, vacations, lunch time and many others. They are tireless machines designed for the sole function of achieving greater success rate in their dealings and deliveries. There are two types  of artificial intelligence systems which are strong artificial intelligence & weak artifical intelligence systems. These are all possible due to machine learning which is their programing to arrive at solid conclusions based on their wide data bank and better serve in these roles much more than a human would. Machine learning has been extremely helpful in a wide range of fields and with efficiency too in such a way that wasn’t known before.

Weak artificial intelligence systems do not use machine learning but however are programmed with some cognitive functions. They’re not as powerful and effective as strong artificial intelligence systems.

Weak artificial intelligence systems are also known as narrow AI and are categorized by their non-sentience. They are mainly focused on achieving narrow tasks which gives it its alternative name. Weak artificial intelligence systems mostly fill in for smaller roles compared to the scope Strong AI can occupy. They have their uses and importance like automating traditionally time consuming tasks and also analyzes data super fast too. Weak AI, even though they may appear to posess human consciousness, they don’t. They simulate cognitive consciousness but they lack human consciousness and this makes their functions limited as compared to the Strong AI. Examples of Weak Artificial intelligence (AI) are Apple’s Siri, Facebook’s newsfeed, Amazon’s suggested purchase. Siri is a perfect example of Weak Artificial intelligence with a narrow variety of options. She answers spoken questions and is programed within a defined range of functions. Narrow AI possess no self awareness or genuine intelligence despite their seemingly capability for cognitive responses.

“The iPhone/Siri marriage represents the arrival of hybrid AI, combining several narrow AI techniques plus access to massive data in the cloud.” Ted Greenwald wrote on Forbes magazine in 2011. Some have voiced their fears about weak artificial intelligence systems saying it could cause potential harm. In 2013 George Dvorsky stated via io9: “Narrow AI could knock out our electric grid, damage nuclear power plants, cause a global-scale economic collapse, misdirect autonomous vehicles and robots…”

Weak AI due to their limited range of functions lack the ability to make quick judgement calls whereas strong AI are programmed to do so. 

Weak AI may simulate consciousness but they are the lowest forms of Artificial intelligence machines. They lack machine learning which higher artificial systems possess. That ability for a machine to simulate human consciousness and be able to make judgement calls without specific programming to respond to that particular problem. Machines learning is a very important aspect in Robotics. The foundational basis as you may call it because without robots being able comprehend its surroundings, it becomes useless. 

Weak AI have a very wide variety of importance and are widely used in our everyday life. Your email spam filter is a very good example. They may be the least of AI but they’re also very well used 

Strong artificial intelligence systems like you may have guessed are the direct opposite of what the weak AIS are. They possess the power (Machine Learning) to make judgement calls and are even equipped with cognitive functions and life like deductions. They’re saddled with machine learning and it helps them simulate consciousness and sentience in a way that would nearly pass off as the real thing. They’re also known as artificial general intelligence (AGI) or True intelligence and these have the intelligence to apply common sense to most problems and come up with a reasonable solution. They are as smart as a typical human being. There are also other types of Strong AI that are known as Super intelligence AI and they’re very smart and are by far smarter than the brightest and most intelligent humans on the planet. 

Strong artificial intelligence systems are akin and likened to human intelligence. They are able to learn, reason, solve puzzles, make judgement calls, make plans and communicate like a human. Turing proposed such a test to access their cognitive capabilities in the early stages of Artificial intelligence. 

Machine learning has improved and redefined functionality and efficiency in its various areas of applications rendering the convectional methods and weak artificial intelligence systems on the back seat. 

Machine learning is closely related to, is a combination of and a derivative of several disciplines one way or the other coming together to be one without ever really coming together. Example is, Machine learning is closely related to and is often interwoven in computational statistics. This is so because computational statistics focuses mainly on prediction making, generalization and conclusions through the use of computer systems which in generality is how machine learning functions in artificial intelligence systems. 

Furthermore, machine learning has deep roots in mathematical optimization which delivers methods, theories and application domains to a field. It is also sometimes combined with data mining and the only difference from the both is that this subfield, data mining focuses more on exploratory data analysis and is alternatively known as unsupervised learning. 

Machine learning is applied within this field of data analytics as it is closely related each other. Machine learning is used to create complex algorithms and models that aid themselves in prediction as is the way and purpose of data analytics. This is known as predictive analytics in commercial use as machine learning is a viable aspect of that field as well.

These analytical models help researchers, engineers, data scientists’ ad analysts to be able to generate reliable and predictable results. It also helps to learn the interrelationships between trends and data.

The theory of Machine learning can be said to be based on the knowledge of recurring data. This can be in form of Instances, situations and the ability to respond to it can only be overcome by a program that is able to predict possible events, asses and respond accordingly to the situation at hand. Being expressly programmed for a specific task and performing it according to program is amazing but what is more amazing for a machine is being able to respond without particular programming for the task. It shifts these computer systems one step towards some measure of sentience and even consciousness in a way.

Looking at the history of machine learning, at this time, an increasing emphasis on the logical, more direct knowledge-based approach created a rift between machine learning and artificial intelligence.  It was said that probabilistic systems were constantly hindered by theoretical and practical problems of the acquisition of data and its representation. With this change in the direction, expert systems had soon taken over AI, and statistics was out of vogue and fast became unpopular. Although, much work on symbolic and knowledge-based learning did continue within AI, which lead to inductive logic programming, but the more statistical line of research into machine learning was now very much out of the field of Artificial intelligence properly and this was in information retrieval and pattern recognition; all of which come together to make up the discipline of machine learning. As we earlier stated, machine learning is a discipline with vast other disciplines situated at the centre of it. 

Neural networks research just like most of these other disciplines had been abandoned by AI and computer science almost around the same time. It was too continued outside the field of computer science and artificial intelligence, as “connectionism”, by researchers from other fields. These researchers included Hopfield, Rumelhart and Hinton. Their major breakthrough in machine learning came in the mid-1980s with the reinvention of backpropagation. It signified the dawning of a new age for machine learning.

Machine learning later regrouped and reorganized as a separate field which began to do very well for itself in around the 1990s. The discipline then changed its goal from attaining artificial intelligence to combating solvable problems that were practical nature. It shifted its focus away from the rather symbolic approaches it had inherited from artifical intelligence to take on a new and improved path which was towards methods and models borrowed from statistics and probabilistic theory. It also greatly benefited from the increasing availability of digitized information, and the ability to circulate it through the means of the internet.

Machine learning and data mining most often than not employ the same methods and in a lot of places, overlap visibly, but while machine learning is based on the use of prediction, based on known properties learned from the training data, data mining however focuses on the discovery of previously unknown properties in the data. 

This angle of approach is referred to as the analysis step of knowledge discovery in databases which is also one of the most important aspects of machine learning and holds a profound addition to the discipline itself. Data mining makes use of a lot of machine learning methods, but these applications are with different goals altogether but on the other hand, machine learning also applies some of data mining methods. An example of this application is “unsupervised learning” or applied as a preprocessing step to improve learner accuracy. A lot of the confusion that occurs between these two research fields which do often have separate journals and separate conferences with ECML PKDD being a major exception comes from the basic assumptions they work with. In machine learning, performance is usually evaluated with respect to the ability to reproduce known knowledge, while in knowledge discovery and data mining (KDD) the major task is the discovery of previously unknown knowledge. Weighed with respect to the known knowledge, an uninformed (unsupervised) method will easily be outperformed by other supervised methods, while in a typical KDD task, supervised methods cannot be used due to the unavailability of training data.

A more widely quoted and more formal definition of machine learning was provided by Tom M. Mitchell who defined it as the algorithms studied in the machine learning field. He also further defined it as “A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P if its performance at tasks in T, as measured by P, improves with experience E.”

This definition of the tasks which is a great tool in machine learning as a discipline offers a fundamentally functional definition as opposed to defining the field as a whole in cognitive terms as we have done before. This aligns with Alan Turing’s proposal in his paper “Computing Machinery and Intelligence”, where he asked revolutionary and profound questions like “Can machines think?”

However, this is being replaced with the question “Can machines do what we as thinking entities can do?”. In Turing’s proposal the diverse characteristics that could be doned by a thinking machine are properly discussed and exhausted, as well as the implications of building one.

Machine Learning Tasks 

Machine learning tasks are generally classified into various broad categories but discussing a few. 

Supervised learning: in this task, the computer or expert system is provided with example inputs, meaning example tasks and their desired outputs, which is a range of preferable outcomes as acceptable and given by a “teacher”, and the goal for the computer system is to learn a general rule that trails inputs to outputs. For some specific cases, the input signal can be only restricted or partially available to special feedback.

Semi-supervised learning: The computer system here is given only an incomplete training signal, a training set with some, in many cases with the target outputs missing. This is mostly due to the presence of more programs leading the computer to make a decision based on reasoning should data remain inconclusive and the target outputs required to solve the problem remain missing. 

Active learning: The computer here can only be provided with training labels for a limited range of instances also based on a budget, and also has to optimize its choice of objects to acquire labels for. When used in an interactive session, they could be presented to the user for labeling.

Unsupervised learning: this is the opposite of supervised learning where labels are still provided but here, no labels are given to the learning algorithm, thus, leaving it on its own to find structure in its input. Basically, it is left to hash out the problem. Unsupervised learning can be a field on its own, a goal to itself which is the discovery of hidden patterns in data or just a means towards an end (feature learning). 

Reinforcement learning: Data in form of punishments and rewards are given only as feedback to the program’s actions in a controlled, dynamic environment, such as driving a vehicle or playing a game against an opponent. Reinforcement learning is the basis for the creation of AI systems for games such as Chess, Go & other games. This has helped AI take a stand in several fields. An example is it making history when Deep Blue beat Kasparov in chess. 

An AI, AlphaGo was also able to defeat a professional Go player, something that was expected to have taken another decade before artificial intelligence attained such level of intelligence. Firsly, Go is an ancient traditional Chinese game simplified in Chinese; standard Mandarin as ‘Hanyu pinyin’ and it literally means the “encircling board game.” Go is an abstract strategy board game between two players. The aim of this board game just like its literal meaning is to surround more territory or board space than your opponent. It is to circle the opponent, hence earning its name “the encirclement board game.” Go is played on a 19×19 board on a grid of black lines. Its game piecesd are called stones and it is played on the lines’ intersection. Its playing time ranges from twenty to ninety minutes of casual playing time whereas the tournaments could go from one to six hours of playing time.

AlphaGo is a computer program that was created originally for playing the traditional Chinese board game Go, known as Hanyu pinyin in Mandarin. It is an artificial intelligence system that was developed solely for this purpose by DeepMind Technologies Limited; it is a British artificial intelligence company that was founded in 23rdSeptember 2010, well over seven years ago with an employee count of around four hundred employees. Deepmind was acquired by Google in 2014 and have succeded amongst other accomplishments in the creation of a neural network that learns how to play video games just like a human would. It is specifically created to assimilate and understand at the pace of a human and may perhaps surpass which is yet to be seen. It learns in a similar pattern like would a human. It is able to access an external memory like a traditional Turing Machine resulting in machines that mimic on a short-term memory level of a human brain. 

AlphaGo AI won a five-game match series against the world’s best professional Go player without handicaps on a standard sized 19×19 Go board. Before the match, the Go master spoke saying artificial intelligence will one day prevail against humans but strongly maintained that it would in no way diminish the beauty of the ancient Chinese game of strategy Go.

Machine learning has been and is instrumental in the several ground-breaking breakthroughs in various fields. It has a more simplified definition. Machine Learning is the science of making computers to learn and act the way humans do. The goal is also to improve their learning over time in an autonomous manner, by providing them with data and information in the form of observations and real-life interactions.”

There are several definitions of Machine learning and their originators. 

“Machine Learning at its most basic is the practice of using algorithms to parse data, learn from it, and then make a determination or prediction about something in the world.” – Nvidia 

“Machine learning is the science of getting computers to act without being explicitly programmed.” – Stanford

“Machine learning is based on algorithms that can learn from data without relying on rules-based programming.”- McKinsey & Co.

“Machine learning algorithms can figure out how to perform important tasks by generalizing from examples.” – University of Washington

“The field of Machine Learning seeks to answer the question “How can we build computer systems that automatically improve with experience, and what are the fundamental laws that govern all learning processes?” – Carnegie Mellon University. 

Dr. Yoshua Bengio, Université de Montréal:

He said machine learning should not be defined by negatives and defined ML. 

Machine learning research is part of research on artificial intelligence, seeking to provide knowledge to computers through data, observations and interacting with the world. That acquired knowledge allows computers to correctly generalize to new settings.

Dr. Danko Nikolic, CSC and Max-Planck Institute:

 “Machine learning is the science of getting computers to act without being explicitly programmed, but instead letting them learn a few tricks on their own.”

Dr. Roman Yampolskiy, University of Louisville:

Machine Learning is the science of getting computers to learn as well as humans do or better.

Machine learning however posses a lot of ethical questions just like most of Artificial intelligence.

Machine learning poses a vast plethora of ethical questions. Computer systems that are trained on datasets collected with biases may exhibit these biases upon use algorithmic bias, thus digitizing cultural prejudices. An example, one could use job hiring data from a firm with perhaps racist hiring policies which may lead to a machine learning system replicating such a bias by scoring applicants for a job against similarity to previous successful applicants based on their details. Responsible documentation of algorithmic rules used by a system and collection of data thus is a critical part of machine learning; as a matter of fact, it is the very foundational basis of machine learning. 

This is mainly due to the fact that language contains biases and machines trained on language will most certainly also pick up this bias. 

There are other forms of ethical challenges that have been posed at machine learning which are not related to personal biases. These are more visible in fields such as health care and many others. There are however raised brows and sporting concern among health care officials and professionals that these expert systems might not be designed in public interest, but as income generating machines or time saving machines. This is especially true in the United States where there is an existing ethical dilemma of improving health care, but also increasing profits which may infact leave them with few options. For example, the algorithms could be designed to provide patients with unnecessary tests or medication in which the algorithm’s proprietary owners hold stakes in thus increasing expenses for the patients. There is huge chance for machine learning in health care to provide officials in health care a great tool to diagnose, medicate, provide prognosis, and even plan recovery paths for patients, but this can not be possible until things such as personal biases mentioned above, and these “greed” biases are addressed.

Machine learning has been applied in a wide range of fields and with efficiency at the top of the list, it becomes increasingly favored. Applications for machine learning include

  • Agriculture
  • Automated theorem proving
  • Adaptive websites
  • Affective computing
  • Bioinformatics
  • Brain–machine interfaces
  • Cheminformatics
  • Classifying DNA sequences
  • Computational anatomy
  • Computer Networks
  • Telecommunication
  • Computer vision, including object recognition
  • Detecting credit-card fraud
  • General game playing
  • Information retrieval
  • Internet fraud detection
  • Computational linguistics
  • Marketing
  • Machine learning control
  • Machine perception
  • Automated medical diagnosis
  • Computational economics
  • Insurance
  • Natural language processing
  • Natural language understanding
  • Optimization and metaheuristic
  • Online advertising
  • Recommender systems
  • Robot locomotion
  • Search engines
  • Sentiment analysis (or opinion mining)
  • Sequence mining
  • Software engineering
  • Speech and handwriting recognition
  • Financial market analysis
  • Structural health monitoring
  • Syntactic pattern recognition
  • Time series forecasting
  • User behavior analytics
  • Machine translation

DEEP LEARNING

Deep learning which is also known as heriachal learning or deep structural which is based on the learning of data representations. It is also a broader family of methods in machine learning thus working hand in hand. These data representation methods are as opposed to task specific algorithms where learning can be supervised, unsupervised or semi supervised. 

Deep learning architectures such as deep belief network, deep neural networks as well as recurrent neural networks have also been applied to various fields including speech recognition, computer vision, audio recognition, natural language processing, machine translation, social network filtering, bioinformatics, drug design and board game programs, where they have produced results which are in some cases much more efficient that their human counterparts 

Deep learning models are scarcely inspired by communication patterns and information processing in biological nervous systems yet they have several differences from the functional and structural properties of biological brains, especially the human brain which makes them incompatible with evidences from neuroscience. 

Deep learning is a sector of machine learning algorithms that make use of a wave of several layers which have non-linear processing units that feature extraction and transformation. Each successive layer of algorithm uses the output from the previous layer as input. In other words, each algorithm works based oj response. Successive input uses output from the previous layer meaning it uses information to create more inputs or it uses what it has learned based on previous interactions to create more logical actions. This is what deep learning does and thus enables it to function properly, acting based on the information it receives and treats it accordingly. This is simply what using output from previous layer to create new input means. 

Deep learning also learns in supervised (example being classification) or unsupervised (example, pattern analysis) manners.

Deep learning also learns multiple levels of representations that correspond to different levels of abstraction; the levels form a hierarchy of concepts which is why it is called heriachal learning alternatively. Another major reason for this is because the algorithm does not run another input without successive output, thus giving glory to heirachy. It runs in successions and a heriachal pattern giving deep learning its name. 

Deep learning has been most influential in the artificial intelligence world today and modern deep learning models are solely based on an artificial neural network, although they can also include latent variables or propositional formulas which are organized layer-wise as is the pattern in DL (deep learning) in deep generative models such as the nodes in deep belief networks and deep Boltzmann machines.

In deep learning, each level learns to translate or transform its input data into a tad more abstract yet composite representation. Taking an example of an image recognition application, the raw input may be a matrix of pixels. The first representational layer may isolate & abstract the pixels and encode edges while the second layer may compose and encode arrangements of edges; the third layer may encode an eye or a nose and the fourth layer may recognize that the image contains a face. Importantly, a deep learning process can learn which is how they work and if features to optimally place in which level on its own. However, this does not completely negate the need for hand-tuning; for example, varying numbers of layers and layer sizes can provide different degrees of abstraction which may need human fine tuning yet. 

 Deep learning algorithms can be applied to unsupervised learning tasks. This is beneficial and important mainly because unlabeled data appear to be more abundant than the labeled data. Examples of deep structures that can be trained in an unsupervised manner are deep belief networks and neural history compressors. 

These analytical models help researchers, engineers, data scientists ad analysts to be able to generate reliable and predictable results. It also helps to learn the interrelationships between trends and data.

The theory of Machine learning can be said to be based on the knowledge of recurring data. Instances and situations and the ability to respond to it can only be overcome by a program that is able to predict possible events, asses and respond accordingly to the situation at hand.